Download:
pdf |
pdfEstimating Consumer Tipping Behavior:
Review and Recommendations
Prepared for: Internal Revenue Service
Prepared by: Fors Marsh Group, LLC
FINAL
Version 2.0
February 2014
The views, opinions, and/or findings contained in this report are those of Fors Marsh Group, LLC and should
not be construed as official government position, policy, or decision unless so designated by other
documentation. This document was prepared for authorized distribution only. It has not been approved for
public release.
Table of Contents
Introduction.................................................................................................................................... 1
Literature Review Summary ....................................................................................................... 2
Review of Methodologies ............................................................................................................ 6
Recommended Approach ......................................................................................................... 17
Appendix A – Reviewed Articles .............................................................................................. 24
Appendix B – Annotated Citations ........................................................................................... 30
Appendix C – Search Engines and Search Terms ............................................................... 69
Introduction
This report is intended to provide guidance to the IRS as it attempts to develop estimates of tipping
and stiffing rates, tipping income, and ultimately, the gap between actual and reported tip income at
the aggregate level and by sector. This guidance is based on the results of past research on tipping
behavior as well as lessons learned from authors’ own work in this area. The first section of this
report summarizes the results from a comprehensive annotated bibliography of academic and
government literature on tipping. This bibliography, which can be found in Appendix B, includes
summaries of research examining average tipping rates as well as individual and establishment
characteristics associated with tipping. In anticipation of subsequent sections, the bibliography also
summarizes articles that do not directly address tipping, but are relevant to the development of
research designs that could be used to collect and analyze data on tipping.
The second section reviews different methods for the collection and analysis of tipping data, and
their potential benefits and drawbacks. Topics addressed include sample sources, specifically
samples drawn from address-, telephone-, and Internet-based samples; the mode used to collect the
data from the sample, including in-person interviews, paper surveys, and Internet surveys; and the
design of the survey, including long-recall cross-sectional, short-recall repeated cross-sectional, and
longitudinal designs. Finally, this section describes potential methods for analyzing the data,
including the use of disaggregated means as well as model-based approaches.
Finally, the third section presents recommended approaches for collecting and analyzing tipping data
based on the reviews in the first two sections. This includes both immediate steps pertaining to
instrument development as well as pilot testing prior to full scale implementation.
FINAL
Page 1
Literature Review Summary
A preliminary set of articles was identified using a bibliography of tipping-related research compiled
by Dr. Michael Lynn.1 Additional articles were identified through backward and forward citation
searches starting from the articles identified in the Lynn bibliography. Google Scholar was used to
identify more recent research that cited the articles from Lynn’s bibliography. Gated articles were
accessed through a local University Library System. However, to mitigate the potential for selection
bias, queries for articles relevant to tipping and survey methodologies were made using several
search engines and archives. This set of search engines and databases included general interest
academic archives and search engines such as Google Scholar, JSTOR, and the Social Science
Research Network (SSRN) as well as specialized business and accounting-related archives such as
Business Source Complete and ProQuest’s Accounting & Tax database. Themes and keywords for
this search were identified based on an initial review of articles obtained from the Lynn bibliography
and the backward and forward searches. From these articles, additional backward and forward
searches were conducted to identify additional articles. From the resulting compilation of articles,
authors influential to the tipping literature were identified based on total numbers of articles
written/published and/or number of citations. These researchers were consulted in order to obtain
any previously unidentified tipping-related papers/research, whether published or unpublished.
Many articles touch on multiple topics that are relevant to determining a methodology for data
collection and analysis of tip-stiffing and tip rates. Consequently, articles cannot be sorted into
mutually exclusive categories based on themes. To facilitate review of evidence from the compiled
literature on specific topics, each citation includes a list of the article’s themes. The reader can use
his or her word processor/PDF reader’s search or find functions to quickly discover articles that
address a given theme. A list of all themes with descriptive text is included in Table 1. A list of the
reviewed articles is provided in Appendix A, with the associated annotations presented in Appendix
B. Descriptions of search engines, search terms, and related themes derived from the search are
provided in Tables 2 and 3 in Appendix C.
1http://tippingresearch.com/uploads/Tip_Bibliography.pdf
FINAL
Page 2
Table 1. Themes
Theme
METHODOLOGY
NATIONAL AVERAGE
TIPPING RATES
INDUSTRY/SERVICE
CASH VERSUS CREDIT
SERVICE CHARGE
BILL SIZE
Description
Methodology used by article along with relevant benefits and drawbacks.
Article’s findings, if any, with respect to U.S.-wide stiffing/tipping rates.
Article’s findings, if any, with respect to differences in average stiffing/tipping
rates across industries/establishment types.
Article’s findings, if any, with respect to differences in stiffing/tipping rates
between establishments/customers who accept/use cash versus credit.
Article’s findings, if any, with respect to differences in stiffing/tipping rates
between establishments that do or do not include automatic tip/service charge.
Article’s findings, if any, with respect to differences in stiffing/tipping rates
between establishments/customers based on bill size.
GEOGRAPHY
Article’s findings, if any, with respect to differences in average stiffing/tipping
rates across geographic regions and jurisdictions.
INCOME
Article’s findings, if any, with respect to differences in average stiffing/tipping
rates of customers with different levels of income.
EDUCATION
Article’s findings, if any, with respect to differences in average stiffing/tipping
rates of customers with different levels of educational attainment.
AGE
Article’s findings, if any, with respect to differences in average stiffing/tipping
rates between customers based on AGE.
GENDER
Article’s findings, if any, with respect to differences in average stiffing/tipping
rates between men and women.
RACE/ETHNICITY
Article’s findings, if any, with respect to differences in average stiffing/tipping
rates of customers with racial/ethnic characteristics.
TIPPING KNOWLEDGE
Article’s findings, if any, with respect to customers’ understanding of tipping
norms (i.e., percent of bill).
Methodology: With respect to the methodologies related to collecting data on consumer/producer
expenditure and reporting, the current literature covers many of the trade-offs between maximizing
data quality, making causal inferences, and ensuring that the sample and their recorded behavior is
representative of the population of interest. Panel-based survey designs, such as the original NPD
Group diary panel (McCrohan & Pearl, 1991; Pearl & McCrohan, 1984) and the Bureau of Labor
Statistics’ Consumer Expenditure Survey, can potentially allow analysts to make inferences about the
effects of interventions on individual behavior because of the ability to control for individual-level
factors that do not vary over time (Parker, Souleles, & Carroll, 2012). However, panel-based survey
designs can also potentially increase respondent burden, leading to increased attrition and selection
bias. In addition, consumer diary panels may induce changes in respondent spending behavior,
leading to less valid predictions for individuals outside the sample (Crossley & Winter, 2012). A
similar trade-off comes with experiments, whether in labs (Alm & Jacobsen, 2007) or fields (List,
FINAL
Page 3
2011)2, which allow for controlled environments, and thus the estimation of treatment effects at the
expense of external validity. Nonpanel surveys, while providing limited ability to make causal
inferences about the effects of different interventions on expenditure and reporting, can potentially
produce more representative samples because of a relatively lower burden being placed on
respondents and consequently higher response rates. However, long recall periods may lead to lower
quality of responses because of the inability of respondents to accurately recall the timing of
spending occasions (Crossley & Winter, 2012).
With respect to the effects of survey modes and instruments, web-based surveys can lead to more
accurate responses than paper-based or in-person/telephone surveys because of the ability of
respondents to more easily skip past irrelevant questions, and, in the case of in-person/telephone
surveys, the increased time respondents have to look up information necessary to accurately answer
questions. In addition, self-administered surveys, which are now primarily web-based, may be more
accurate than in-person or phone interviews because of the anonymity that self-administered webbased surveys afford (Crossley & Winter, 2012). However, a sample of individuals with web access
may not be perfectly representative of the population of interest because of an individual’s
probability of web access being related to individual characteristics as well as his or her geography.
Industry/Service: The majority of tipping research has focused on the restaurant industry, but a few
studies have focused on other industries where tipping is prevalent. For instance, previous studies
have investigated tipping rates for luggage handlers, taxi drivers, bartenders, parking attendants,
hotel bellmen, and barber/hair stylists. Koku (2005) concluded that there is a difference between
tipping rates in the restaurant industry and outside of it by interviewing customers of each sector.
Similarly, Paul and Gardyn (2001) identified higher tip percentages for restaurant servers than for
barbers, taxi drivers, food delivery workers, hotel bellmen, and several other professions. However,
more research is necessary to provide a more direct comparison between customers’ tipping
behavior in restaurants and other service industries.
A relatively significant amount of research has been conducted to investigate alcohol’s effect on
tipping rates. Most of that research demonstrated that customers who consume alcohol provide
higher tip percentages than those who do not. Even after controlling for the relationship between bill
size and tip percentage and a host of other variables, Lynn (1988) identified a significant effect of
alcohol consumption on tip percentage. Similarly, Bodvarsson and Gibson (1997) found that tip rates
varied across establishments, but establishments that were licensed to serve alcohol received higher
tips.
Cash Versus Credit: Several studies have investigated the difference in tipping rates between
restaurant patrons who pay their bill with a credit card and those who pay with cash. Although some
articles failed to find a significant difference between payment methods, the majority of research on
this subject seems to indicate that customers who paid with credit cards tipped at higher rates than
those who paid with cash.
Bill Size: Some research has focused on what is known as the magnitude effect of tipping. The
magnitude effect refers to the tendency for customers to leave bigger proportional tips for smaller
2
List, J. A. (2011). Why economists should conduct field experiments and 14 tips for pulling one off. The
Journal of Economic Perspectives, 25(3), 3-15.
FINAL
Page 4
bills compared with bigger bills. Chapman and Winquist (1998) concluded that customers provide
higher tips on smaller checks, and that the tip percentage decreases as the total bill increases by
demonstrating the effect in restaurants and barbershops/hair salons.
Regional Differences: Geographic difference such as between urban and metropolitan areas and
between different census regions or divisions of the country, have been investigated previously in
studies that contained sufficient sample sizes. Typically, customers in urban areas have been shown
to tip at higher percentages than ones in rural areas, and there are some studies showing significant
differences in tipping behavior and knowledge between customers from the Northeast region of the
country and those in other parts of the country. For instance, McCrohan and Pearl (1983) found that
Northeast customers tipped at higher rates than those from middle parts of the country, and Lynn
(2006) reported that Northeast customers had a higher knowledge of tipping norms than those in
the South. However, the urban/rural difference and regional could potentially be explained by
differences in racial composition as well as education and income levels. Studies have demonstrated
that these variables significantly influence tipping behavior and tipping knowledge, with higher
income levels and educational levels leading to higher tipping and knowledge of tipping norms.
Age and Gender: have been investigated in a number of studies on tipping behavior, and the
influence these variables exert on tipping behavior has been somewhat mixed. Gender differences
have been inconsistent across studies, whereas age has been linked to significantly higher tipping
rates (Pearl & Vidmar, 1988) and greater knowledge of tipping norms (Lynn, 2006). However,
differences in tipping behavior and tipping knowledge due to age could be confounded by other
factors, such as higher income levels, differences in payment methods, and educational differences.
For instance, Lynn (2006) showed in his analysis that higher knowledge of tipping norms with
increasing respondent age was nonsignificant when other factors such as education, income, and
metro status were included in the analysis.
Race Ethnicity: One of the most researched topics in tipping compliance is how tipping behavior
differ between various racial/ethnic groups. Numerous studies have researched this topic,
investigating not only actual tipping behavior, but knowledge of tipping norms as well. The findings of
these studies have some robust conclusions; primarily, that Black customers tip at lower rates than
White customers in the restaurant industry. Though less researched, studies investigating
racial/ethnic differences in other tipping industries have reached similar conclusions: Black
customers usually tip at lower rates, stiff more often, and tend to leave “flat” tipping rates at higher
percentages than White customers, as noted in Lynn’s 2004 study on Black-White tipping
differences among service industries (though this effect differs across certain service industries).
Although a significant amount of research has investigated the differences between White and Black
tipping behavior, some work has looked into differences between Asian and Hispanic customers.
Studies by Lynn in 2006 and 2013 on tipping rates and knowledge of tipping norms indicated that
Hispanics tip at lower rates than Whites and have lower levels of tipping knowledge. Asian customers
in these studies were not shown to tip at significantly different levels or be less knowledgeable about
correct tipping norms. However, these racial/ethnic groups have not been as thoroughly researched
as Black and White customers.
FINAL
Page 5
Review of Methodologies
For the purposes of developing estimates of consumer tipping by industry, multiple approaches can
be taken with respect to the method used to collect the data as well the method used to analyze the
resulting data. A data collection approach is defined along several dimensions; specifically, the
choice of sampling source, survey mode, and survey design. Methodologies used to analyze the data
can be roughly categorized into simple, nonparametric approaches and parametric, model-based
approaches.
Sampling Sources
The primary factors to consider when choosing a sampling source are the representativeness of the
resulting sample with respect to tipping behavior and the costs associated with recruiting and
retaining the sample. Sampling-related bias can result from an unrepresentative frame, nonresponse, incompletes, and, in the case of a longitudinal panel, attrition. In addition, to mitigate the
potential for additional nonresponse that results from transitioning sampled respondents to the
survey, survey sources are often coupled with survey modes. To minimize total survey error, biases
that result from the survey mode have to be considered when choosing a sampling source. Table 2
discusses the benefits and drawbacks of different sampling sources.
RDD Sample: Random digit dialing (RDD) uses randomly generated phone numbers to select a
sample for participation in a survey. RDD sampling is helpful in the sense that it may allow for
coverage of the population that has unlisted numbers, but there are problems associated with cell
phone users. First, in some RDD methodologies, cell phone users are not reachable, which excludes
individuals without household lines from the sample, affecting the generalizability of results.
Second, even if cell phone users can be reached, it may not be possible to determine the
participant’s location from his or her area code because many cell phone users retain their old
numbers when moving to different regions of the country, forcing researchers to rely on selfreporting. Self-reporting of basic demographics has a negative influence on response bias analyses
because little to nothing is known about the nonrespondents. This is compounded by the fact that
RDD response rates continue to decrease with the widespread adoption of caller ID and call
screening.
Address-Based Sample: An address-based sampling (ABS) source relies on home address and
demographic information from the frame file, which is provided by third-party vendors and the U.S.
Postal Service (USPS). This source allows for measureable sample coverage across a population and
a fairly well-known probability of the sample selection. Additional household data can be purchased
and appended to ABS files to assist in more targeted sampling and further response bias analyses.
Although the costs for a mail paper sample are not low, cost per complete has been found to be
lower than that for RDD studies3 with certain populations. However, response rates for mail-based
paper surveys, which are the most commonly used data collection mode with a mail paper sample,
3
Medway, R.L., Viera, L., Turner, S.R. & Marsh, S.M. (2009). List-assisted mail as an alternative to random digit
dial in a survey of the young adult population. Paper presented at the 64th Annual Conference of the
American Association for Public Opinion Research, Hollywood, FL.
FINAL
Page 6
have continued to decrease over time as people increasingly to use the Internet as their medium for
correspondence (Dillman et al., 2009b).
Traditional Internet Sample: Traditional Internet samples are collected via an opt-in procedure where
individuals choose to join a survey administrator’s panel. This panel acts as a potential pool of
respondents who are then queried to participate in individual surveys or diaries. An opt-in sample
might not be representative of the population of interest with tipping or expenditures on tipping
services. Unlike RDD or ABS methods, randomly sampling from an email-based Internet frame is
made difficult by the lack of an algorithm that can randomly select email addresses, due to
inconsistencies in email address conventions (Dillman, 2009). Yet, similar to the RDD and ABS
methods, traditional Internet samples can fit with various data collection modes. Although the online
survey is the most straightforward mode for Internet samples, it should be noted that a longitudinal
diary approach could also be used, such that all those potential respondents are contacted and
recruited to report their tipping behavior for a predetermined amount of time (see Survey Design).
Some examples of different Internet samples include:
GfK Knowledge-Panel®: The KnowledgePanel is an Internet-based panel that uses a
probability-based sampling strategy where the survey frame is derived from the USPS
Delivery Sequence File. Individuals are invited to participate in the panel by mail, followed
by telephone calls for those who do not respond to the initial invitation. Households are
sampled without replacement, avoiding potential bias that may result from respondents
participating in the panel twice. For those individuals selected for participation without
computers or an Internet connection, a netbook is provided. This process attempts to
mitigate the selection bias associated with web surveys while preserving the benefits
associated with a computer interface. The primary benefit of the KnowledgePanel relative
to the opt-in panels described below is that knowing the probability of selection allows
researchers to estimate error. However, these estimates will always be deficient capturing
all aspects of non-response unaddressed by demographic post-stratification. Further, the
procedures used to setup and maintain panel membership and participation serve as an
additional component of error difficult to fully model and correct for.
Blended Online Sample (Ipsos Ampario): Ipsos’ blended sample approach combines the
use of its Ampario online sampling method in addition to its iSAY online panel—an online
panel of 800,000 members and their households. Ampario is a new, nonprobability
sampling procedure that Ipsos has developed that invites respondents by invitations,
banner ads, and other means on 100 to 400 websites that have partnered with Ipsos.
These two methods are combined into a single sample using Ipsos’ proprietary Cortex
routing system, which allocates and reallocates a sample given respondent eligibility.
Simply put, when respondents are not eligible for one survey, they are immediately
redirected to other surveys in progress. In traditional one-off opt-in surveys, noneligible
respondents are lost sample, a considerable cost. Finally, Bayesian methodology, which
requires previous information regarding the overall sample of interest in order to mix with
current information for the final distribution of results, is used to form final distribution. As
is the case with a traditional online sample, Ipsos’ blended sampling could work with
several different data collection modes, but it is best served with an online-based
FINAL
Page 7
questionnaire, which could include a cross-sectional administration or a longitudinal diary
approach. However, because of the opt-in nature of the Blended Sample, it is not possible
to model the probability of response, and thus account for that source of potential bias in
survey estimates.
NPD Group Online Sample: NPD Group utilizes sophisticated techniques both at the
sample design stage and post-survey weighting stage to reduce bias and increase
representativeness of the sample, but it is not a probability sampling technique.. Although
there are certain demographic groups that have less representation online and are not
represented in correct proportions as they would be in the U.S. Census, they are large
enough that they can be sampled appropriately to represent the U.S. population.
Recruitment of panelists is done using a wide variety of opt-in sources (email, affiliate
marketing, co-registration, banners, etc.). The wide variety of sourcing ensures a large
representation from various strata of the U.S. population. All sourcing is balanced and
ensures no single source provides a disproportionate percentage of recruits. A number of
other steps are put in place to prevent fraudulent prospects from joining the panel. This,
combined with other behavioral data collected, is used to monitor recruitment source
quality and guide media planning for recruitment. NPD limits the number of surveys a
panelist can start in a day, week, and month to avoid survey fatigue. Response rates are
tracked at an individual panelist level—if panelists fail to participate consistently over
time, NPD removes them from the active panel.
The sample for a particular study is drawn from this panel to demographically represent
the U.S. population. Sophisticated algorithms take varying response rates by demographic
groups into account to provide stratified quota for each of the targeted cells. Once the
sample is collected, the cells that fall short in demographic representation during
sampling are weighted during processing the data. Again, because of the opt-in nature, it
is not possible to model the probability of response, and thus account for that source of
potential bias in survey estimates.
FINAL
Page 8
Table 2. Summary of Trade-Offs in Sampling Sources
INTERNET SAMPLES
Qualities of
Sampling Plans for
Current Research
General population
coverage
Known probability of
selection
Response rate and
cooperation rate
Cost per complete
Address-Based
Sample
Traditional
Internet
Probability
Sample
GfK
Knowledge
Panel®
Probability
Sample
Ipsos
Nonprobability
Sample
NPD
Nonprobability
Sample
Medium
High
Low
Medium/High
Medium
Medium
Medium
High
Low
High
Low
Low
Low
Medium
UNK
Low
UNK
UNK
High
Medium
Low
Medium
Low/Medium
Medium
RDD
Telephone
Sample
Survey Mode
As is the case for sampling source, the choice of survey mode can impact the representativeness of
the sample by influencing the demographics of those who choose to actually take or complete the
survey. In addition, the burden that a particular choice of survey mode places on respondents can
influence the accuracy of the data obtained from the survey for a given respondent. Issues of
selection bias and measurement error thus have to be considered when choosing the survey mode.
Web-based questionnaire: Some of the many benefits to online surveys include more rapid and
reliable transmission of completed questionnaires as well as more flexibility in skip patterns. This
can also reduce respondent fatigue by withholding non-applicable items (Crossley & Winter, 2012;
Dillman et al., 2009b). Related to this is the lower cost of administering a web survey versus other
modes due to the lack of need to send or code a physical questionnaire or have an interviewer make
contact with the respondent. The accuracy of responses may also be improved relative to in-person
or telephone surveys because of the ability of respondents to retrieve relevant information—a benefit
that results from the ability of respondents to answer web-based questions when convenient.
Another benefit of web-based (and mail-based) surveys is that social desirability effects that result
from the presence of an interviewer can be mitigated. On the other hand, interviewers can diminish
the effects of respondent confusion by helping clarify ambiguous questions or following up on
inconsistent responses—advantages that a web-based or mail based survey may lack. There is also
evidence of a “primacy” effect in responses to visual based surveys (i.e., web- and mail-based ones),
where respondents are more likely to pick the first option given in a list of discrete responses
(Dillman et al., 2009).
Mail-based surveys: Decreased coverage of telephones and difficulty in estimating coverage in web
panels has led to increased use of address-based sampling (ABS) and, subsequently, mail-based
surveys. In addition to this positive association with ABS, mail surveys have actually maintained
relatively higher response rates than telephone- and web-based surveys. However, mail-based
surveys also have a number of weaknesses. They are generally less flexible when it comes to skip
logic than web-based surveys. In addition, a mail-based survey may be significantly more costly than
FINAL
Page 9
a web-based one in terms of time and money because of the significant variable costs needed to
publish survey and mail pieces, transport these pieces between survey administrators and
respondents, and code the responses. This disparity is likely to be greatest for larger survey efforts
requiring big sample sizes and/or significant follow-up.
In-person surveys: Although there may be many variations of this approach, for the current project inperson surveys may involve interviewers waiting outside of restaurants and other service-industry
establishments with the purpose of asking patrons a battery of items associated with their tipping
behavior. This approach could allow for immediate recall of a behavior as well as confidence ratings
of the data if the interviewer was trained (as done by the Bureau of Labor Statistics [BLS] in “the
New Orleans Test”), thus possibly ensuring more reliable data (Crossley & Winter, 2012). However,
the cost can be quite prohibitive, particularly with respect to the number or survey administrators
and/or transportation required to ensure that different demographic groups, establishment types,
and geographic areas are properly represented in the sample. BLS, in particular, has conducted a
number of longitudinal in-person studies over the years under its National Longitudinal Surveys
program, and has used a number of techniques to keep respondent attrition low. These techniques
include giving their researchers access to local resources to track down any respondents who might
have moved or passed away since the previous survey and corresponding with the participant to
encourage survey compliance (thank-you letters and pamphlets highlighting the data and knowledge
gleaned from the survey effort are two examples). To mitigate the social desirability issues previously
discussed with this method, BLS has incorporated computer-based response options so respondents
can listen to sensitive questions with headphones and type in their responses without their
interviewer’s knowledge.
Phone survey: Phone surveys can either be administered by working off a purchased consumer
directory or through RDD. With the advent of cell phones, many households no longer use landline
phones, and that makes a portion of the population difficult to reach in a cost-effective manner (Pew,
2012).4 RDD also lends itself to difficulty in measuring non-response bias given the lack of
knowledge of the sample frame and, specifically, the nonrespondents. Overcoming low response
rates (and potential selection bias) can require frequent calls, increasing the cost of this mode.
Another potential issue is that these types of real-time surveys do not give respondents enough time
to refer to their schedules or other sources of information concerning past expenditures compared
with web and mail surveys (Crossley & Winter, 2012). This will tend to undermine data quality.
Diary study: Following Pearl and McCrohan (1984) and McCrohan and Pearl (1991), a diary panel
can be used to provide data over a certain time span for each observation (i.e., tipping behavior of
interest) and has been used for both servers and customers in the past. The fact that respondents
are expected to record their expenditures near the time when the expenditure was made can
mitigate the effects of recall on response accuracy that would plague a recall-based survey.
However, this lack of recall bias can come at the cost of not properly capturing seasonal fluctuations
in expenditure and tipping behavior if the diary period is short and/or infrequent. In addition,
research burden on the participant is quite high and compliance (in the form of attrition and
recorded expenditures) significantly drops off over time. It is also possible that the act of recording
4
Pew Research (2012) “Assessing the Representativeness of Public Opinion Surveys.” http://www.peoplepress.org/2012/05/15/assessing-the-representativeness-of-public-opinion-surveys/
FINAL
Page 10
expenditures may induce a downward trend in expenditures over time (Crossley & Winter, 2012).
This learning effect is a well-known research confound whereby subjects modify their future behavior
in response to the knowledge and skill they gain by being part of the study.
Mixed-mode surveys: Using multiple survey modes has the potential benefit of increasing response
rates because of differences in mode preferences across different respondents (Dillman et al.,
2009; Crossley & Winter, 2012). However, given that mode has an effect on response quality, data
gathered using different modes will not necessarily produce comparable responses (Dillman et al.,
2009). Measurement error due to mode effects may be exacerbated if modes with low degrees of
recording error are combined with a mode with a high degree of recording error versus the use of a
single mode with a low degree of recording error. There is consequently a trade-off between
nonresponse/selection and the potential for measurement error.
Table 3. Summary of Trade-Offs for Alternative Consumer Study Modes
Qualities of Modes for
Current Research
Web-Based
Questionnaire
Mail-Based
Paper Survey
In-Person
Interview
Phone
Survey
Diary
Study*
Interviewer effects
Low
Low
High
High
Low
Learning/Testing effects
Low
Low
Low
Low
Medium
Respondent controls when to
participate (at a convenient time)
High
High
Low
Low
Medium
Dynamic question branching
High
Low
High
High
High
Quick data turnaround
High
Low
Low
High
Medium
Immediacy of recall
Low
Low
High
Low
High
Administration costs
Low
Medium
High
Medium
High
*Through electronic diary
Study Design
To obtain a picture of expenditure and tipping behavior that is representative of a given period of
interest, several study designs can be employed. These designs would differ with respect to the
number of times individual respondents are interviewed, the period over which the interviews take
place, and the length of the period over which the respondent is required to recall their tipping
behavior. A longitudinal, or diary, study would involve surveying individual respondents about their
tipping behavior multiple times over the course of the period of interest. A cross-sectional study
involves surveying each respondent once over a short period, while requiring that they recall their
tipping behavior for the entire period of interest. Finally, a repeated cross-section would only require
that respondents provide information about their tipping behavior for the period immediately
preceding the interview, but the interviews would be conducted over the entire period of interest.
Longitudinal: A longitudinal study requires repeated observations of the same subject over a specific
length of time. Because the same subjects are tracked over time in a longitudinal study, researchers
FINAL
Page 11
can more reliably attribute a change in behavior to an observed variable. In terms of the proposed
methodologies, a longitudinal diary study could illuminate changes over the course of a week or
across seasons in consumers’ tipping behavior. Longitudinal studies could also be used to track the
tipping rate over time with a multiyear effort. In addition, when examining the causes of tipping
behavior, longitudinal data allows one to control for unobserved individual level factors that affect
tipping, enhancing the researchers’ ability to make causal inferences. However, these two latter
benefits may not be relevant for the purposes of this project. Asking participants to record their
tipping behavior for every service-related purchase immediately afterward over a specified period of
time (e.g., one week) would allow for data collection among several different service industries
without the need for recall. However, attrition among longitudinal studies is certainly higher and
places a higher burden on the respondent. Furthermore, longitudinal studies tend to be more
expensive than cross-sectional studies that merely ask for participation for a short duration of time.
Cross-sectional: Unlike longitudinal studies, cross-sectional studies do not utilize repeated
observations of the same respondent. Instead, cross-sectional studies aim to survey people of
different populations at one point in time, allowing for researchers to compare different populations
simultaneously. At another time, the researcher surveys a different sample that is estimated to be
congruent to the previously surveyed sample. This form of surveying avoids the high costs and high
attrition rates associated with longitudinal studies. All of the proposed data collection methods could
potentially use a cross-sectional approach. Mail-, online-, and phone-based surveys frequently use
single contacts with participants in order to aggregate data for a given population. Similarly, diary
studies can take a cross-sectional approach in the sense that participants are asked to provide
feedback about tipping behavior over a 24-hour period. In the process, they would rely less on
respondent recall, but avoid the burden of high costs and attrition associated with a longitudinal
diary study. However, it is more difficult to be sure that changes in variables of interest within
populations are due to outside factors, because respondents are being grouped as opposed to
following the same respondents over time. In addition, estimates derived from a single crosssectional survey with a short recall would not accurately reflect annual tipping rates if expenditure
and tipping rates vary by season or day of the week.
Repeated Cross-Sectional: Repeated cross-sectional studies, also known as synthetic panels, offer
an alternative to longitudinal and single cross-sectional studies (see Parker, Souleles, & Carroll
(2012). Data from multiple cross-sections of survey data would be pooled and respondents sorted
into strata defined by multiple, unchanging characteristics (gender, ethnicity for individuals,
establishment type and location for establishments/managers). Changes in mean outcome variables
(bill size/tipping) for individual strata could then be tracked over time to discern seasonal trends in
reported tipping. Unlike single cross-sectional studies, this design/methodology allows variation over
time in respondents’ tipping (in the case of a consumer) or tip reporting (in the case of
server/establishment surveys). In addition, these types of studies are less susceptible to issues
associated with longitudinal studies related to survey nonresponse and attrition. The original tipping
studies conducted by IRS/NPD, while using data collected through a diary, treated their data as a
repeated cross-section for the purpose of analysis.
FINAL
Page 12
Analytic Considerations
The goal of the IRS tipping project is to produce estimates of establishment and/or employee tip
income that will inform the development of policies that encourage tip reporting. Given that tip
income, tip reporting propensity, and optimal policies to encourage tip reporting are likely to vary by
sector and geography, estimates of tip income at the industry-location level will likely be more useful
to the IRS than more aggregated data. As individual establishments and employees may be less
likely to provide accurate responses to surveys that ask about tip income (Simpson, 1997),
consumers have been the focus of past research in this area. However, because compliance-based
policies are inevitably going to focus on specific types of establishments and locations, consumer
tipping data is only useful if consumer tipping can be linked to particular industry-locations. Given
that most establishments likely draw the bulk of their customers and tipping revenue locally, this
implies that to produce accurate estimates of tipping revenue for particular industry-locations,
estimates will have to be produced for relatively small geographic units. This section considers two
methods of estimating tipping rates for small geographic areas and their implications for the design
of the data collection instrument: Disaggregated Means (DM), and Multilevel Regression and PostStratification (MRP).
Disaggregated Means: The simplest approach to estimating tipping rates for particular geographic
units would be to simply take the mean tipping rate for all respondents located in a particular
geography. Specifically, the estimate is calculated as:
1
T =
T
Where T is the tipping rate of individual i for sector j in location k and n is the number of
individuals in location k. Besides its simplicity, the advantage of DM is that it makes few
assumptions relative to a model-based approach such as MRP (see below). The disadvantage of DM
is that the number of observed tipping incidents for a given establishment type/location strata may
be very small given a nationally representative survey of typical size (N = 5000).5 Consequently, bill
sizes/tip rates for given sectors/locations from the survey will likely be particularly noisy for a
nationally representative sample. Indeed, for very small levels of geographic aggregation, such as
counties, there may be no observations for a given establishment type to make the estimate. For this
reason, a model-based strategy, like that undertaken by McCrohan and Pearl (1991), may, under
certain assumptions, be used to extract precise predictions of tipping rates at a more disaggregated
level. One such modeling-based approach, MRP, is discussed below.
Multilevel Regression and Post-Stratification: One means of linking customer-level tipping data to
establishments while mitigating issues related to noise in small strata is MRP (Gelman & Little,
19976; see Buttice and Highton, 2013, for a recent review and critique). MRP has attained popularity
Buttice, M. K., & Highton, B. (2013). How does multilevel regression and poststratification perform with
conventional national surveys? Forthcoming, Political Analysis.
6 Gelman, A., & Little, T. C. (1997). Poststratification into many categories using hierarchical logistic regression.
Survey Methodology, 23(2): 127-135.
5
FINAL
Page 13
by social scientists who wish to obtain geographically disaggregated estimates of a quantity of
interest.
Analyzing consumer tipping data using MRP would first involve estimating models of consumer
expenditure and tipping that take the form:
E
= βX
+ αG + C
T
= βX
+ αG + C
Where E is the amount spent by respondent i for a service in sector j in location k; T is a tip rate
calculated by dividing a reported dollar amount in tips by E or by directly asking the respondent for a
tip rate; X is a set of observable respondent-level demographic variables such as race,
socioeconomic status, etc., that are likely to influence both tipping and expenditure; and G is a set of
location-specific factors such as whether the location is part of a rural or urban region that capture
variability in expenditure and tipping by sector that is not explained by differences between locations
in X. Locations are defined as the market area of the establishment. Although it is likely that the size
of a given market area will vary by establishment, it might be more practical to assume that an
establishment draws most of its customers from the county or metropolitan area in which it is
located. Finally, C is a constant. After estimating model parameters β, α, and C, predictions are
generated for strata defined by all N combinations of values of X and G covariates. Poststratification
is then used to generate an average tipping rate for a given establishment type/location:
T =
E P
∑ E P
T
Where P is the population of a given stratum in a given location, taken from census data. Estimates
for the average tipping rate for a given sector/location is thus the average tipping rate across all
strata, weighted by the strata’s expenditure at a given establishment type and the proportion of a
location’s population in the strata. The benefit of using a quasi linear, additive model to produce
predictions for individual strata rather than using nonparametric estimates from the survey is that, if
the linear model provides reasonably accurate estimates of expenditure and tipping rates, the
resulting strata-level predictions are likely to suffer less from sampling variability in small to
moderate sample sizes than nonparametric estimates. The resulting estimated sector-location
tipping rates can be multiplied by an establishment’s reported bill size to arrive at a prediction for its
tip income. This estimate can then be compared with reported tip income to arrive at estimates of tip
reporting.
Note that the model described above is more flexible than that presented in McCrohan and Pearl
(1991) insofar as it (1) disaggregates tipping occasions by industry for the purpose of the regression
and (2) incorporates consumer-level demographic data into predictions. Although the model in
McCrohan and Pearl (1991) only allowed predicted tipping rates to vary by establishment type and
by limited degree geography (size of metropolitan area and census division), MRP may produce
predictions of tipping rates by establishment type for a location that varies not just by metropolitan
status and census division but, because of the poststratification step, also by the demographics of a
particular locality.
FINAL
Page 14
Integrating Data Collection and Data Analysis
Obtaining usable information from consumer tipping data will require that the design of the data
collection instrument anticipate the requirements of the methodology used to analyze the data. With
the assumption that this methodology will incorporate features of both a DM and MRP, this section
reviews some items to consider when designing a survey instrument.
Observable Variables: The poststratification stage of MRP requires counts of demographic strata
defined by the individual-level variables in the regression stage for the geographic units of interest
(i.e., market areas of establishments). Given this requirement, a review of available 2010 Census or
5-year American Community Survey (ACS) data would allow for a determination of what strata counts
are available. This will, in turn, inform the construction of the survey instrument to ensure that
relevant demographic data is obtained from respondents. If, for instance, we could obtain data on
number of individuals of a given age-race-income strata by county, we would want to make sure we
could obtain data—either from the respondents or the survey frame on age, race, and income—
similar to the original IRS tipping study (Pearl & McCrohan, 1984; McCrohan & Pearl, 1991), so as to
post-stratify by income group, age, and region using strata counts taken from Census data.
Geographic Variation: MRP accounts for regional variation in outcomes of interest (in our case,
tipping), by including region-level variables that are thought to predict that outcome. To model the
effect of region-level variables on tipping, we will require that our survey/diary sample be drawn from
variable localities. With respect to geography, the academic literature on tipping has generally
focused on differences in tipping between individuals located in metropolitan and nonmetropolitan
areas. This suggests that our geographic variable should be some indicator of urban status or
population density. However, this might pose a problem for estimating a multilevel regression in a
nationally representative sample given that the overwhelming majority of the country’s population
lives in urban areas. Consequently, it would probably be advisable to oversample rural areas. To do
this, however, it will be necessary to define our urban-rural typology before fielding the survey/diary.
Specifically, we will want to decide on the urbanization categorization. One simple categorization
scheme is the Rural-Urban Continuum Codes (RUCC) produced by the U.S. Department of
Agriculture7. RUCC codes incorporate information on a county’s population density as well as its
proximity (adjacency) to metropolitan areas. The advantage of the use of adjacency is that it may
better reflect the proximity of an individual residing in a county to large numbers of other people than
would be the case if only the county’s population density were considered. One downside to the
RUCC relative to a simple measure such as population density is that it is tied to counties. If we
decide to use a geographic unit other than counties, using the RUCC scheme would require some
means of assigning a status to the alternative unit, which would be simple if the unit were nested
within counties, but less so if counties were nested within the alternative unit or if the borders did
not align with counties, such as in the case of Designated Marketing Areas (DMA).
Temporal Variation: If tipping is seasonal as past research has suggested, computing an annual
average estimate of tipping would be complicated by the potential unrepresentativeness of the
sample with respect to tipping. This would be the case within a short recall cross-sectional survey to
differences in propensity to respond across the year to the day of the week, or in a diary panel
7
http://www.ers.usda.gov/data-products/rural-urban-continuum-codes/documentation.aspx#.UrMWBfRDu6M
FINAL
Page 15
because of attrition. Although this may be mitigated by modeling tipping and expenditure behavior
using time effects in order to create a synthetic panel (in the case of repeat cross sections), if the
lack of variation is extreme enough, then parameter estimates on the time effects will be imprecise.
We might thus want to consider stratifying the sample over days of the week and the year (in the
case of a repeat cross section) or have some means of mitigating panel attrition, perhaps by
oversampling individuals from demographic groups that have a high probability of attriting8 or else by
having some procedure in place to bring on additional panelists.9
Establishment Types/Sectors: One of the goals of the project is to examine variation in tipping rates
by industry and establishment type. This implies the use of an establishment typology. The degree to
which survey design will be affected by the need for an establishment typology will depend on the
type of information we can obtain from respondents. If we can obtain the name of the establishment
where a transaction took place, we may possibly be able to classify the establishment after the
survey has been completed depending upon our needs. If that is not feasible, however, we will likely
need to obtain information on establishment type from the consumers. In that case, we will have to
design the survey such that the options for establishment classification are intuitive and, perhaps
most important, limited enough so as not to increase respondent burden to such a level as to
increase nonresponse, attrition, or otherwise undermine response quality. The original IRS/NPD diary
panel (Pearl & McCrohan, 1984) arguably did a good job of dealing with this trade-off. Individuals
were asked to classify establishments into one of six broad categories and then, in a second
question, asked to name the type of food served. Consequently, respondents were not confronted
with a large typology of establishment types in one list. Defining establishment types and eating
occasions by multiple dimensions and then having a separate question for each dimension allows for
a detailed typology while minimizing respondent burden. The chosen typology will also have to be
meaningful such that the parameters relating the individual and geographic variables to expenditure
and tips will be precisely estimated (i.e., not heterogeneous) when estimated for a given type. Also,
this taxonomy must be extended to include establishments other than restaurants. It is thus
important that we consider how we are likely to obtain information on establishment type, as that will
likely inform the degree of trade-off between collecting accurate information and the precision of the
categorization. Another consideration trade-off with having a large number of establishment types is
the potential lack of variation in terms of expenditure and tipping behavior one will see if the number
of individuals who actually used the service is too small. Larger sample sizes may be necessary to
obtain at least some variation in spending and tipping for establishment types for which individual
patronage is infrequent.
Frankey and Hillygus (2013) found that non-White respondents were more likely to attrit from the American
National Election Study.
9 McCrohan and Pearl (1991), for example, used a panel that was replenished quarterly to match strata
population targets.
8
FINAL
Page 16
Recommended Approach
Based on the benefits and drawbacks of the methodologies reviewed in this report, the following
section provides recommendations for the IRS in developing estimates of tipping and stiffing rates,
tipping income and, ultimately, the gap between actual and reported tip income both at the
aggregate level and by sector. Given many of the unanswered methodological questions in the
literature, this report recommends a two-stage process whereby a small set of methods tests will be
conducted prior to full-scale administration. Specifically, we recommend examining the performance
of a web-based, repeated cross-section survey administered to both a probability and non-probability
internet-based panel. The choice of a probability or nonprobability web panel could be adjudicated in
a validation phase (see below).
Sample Source
As discussed in the earlier section, all the sample sources covered (RDD, ABS, or the traditional
Internet based samples) have a variety of strengths and weaknesses pertaining to sample-related
bias. Although phone and address-based frames may arguably be more representative of the U.S.
population as a whole than Internet-based panels, response rates are generally low and have been
declining over time (Pew, 2012; Keeter et al., 200610; Curtin, 200511). These low response rates
would likely become even more problematic if, as is recommended below, a web-based mode is used
to conduct the survey, given the author’s experience with low conversion rates of individuals
recruited using these methods to a web-based survey. Further, these more traditional methods may
become less mandatory as traditional Internet-based sampling sources continue to evolve,
minimizing deficiencies of idiosyncratic recruiting methods prevalent with single source opt-in
panels. In fact, recent research on “blended” approaches that use multiple online respondent
sources have been found to yield results more similar to dual frame RDD.12 In addition, the GfK
Knowledge Panel® continues to use a probability based sampling strategy where the survey frame is
derived from the USPS Delivery Sequence File.
While none of these methods has a clear advantage with respect to sample-related bias, the same
cannot be said for issues related to cost. As already discussed, recruiting individuals using RDD or
ABS are likely to be very resource-intensive. In the case of the former, it might take many attempts to
contact a given individual before receiving a response, resulting in high labor costs. In the case of
ABS, the requirement that the request be printed and transported to the potential respondent carries
obvious costs, and response times may be slow. By contrast, recruiting a sufficient number of
individuals from Internet-based panels will likely be less costly because of the panelists’ stated
willingness to participate and the ease of scaling given relatively low variable costs. Even in the case
of the GfK Knowledge Panel®, which recruits its panelists using more costly ABS methods, the
Keeter, S., Kennedy, C., Dimock, M., Best, J,. & Craighill, P. (2006). Gauging the impact of growing
nonresponse on estimates from a national RDD telephone survey. Public Opinion Quarterly, 70, 759-779.
11Curtin, R. (2005). Changes in telephone survey nonresponse over the past quarter century. Public Opinion
Quarterly, 69, 87-98.
12 Vidmar, J., Bricker, D., Young, C., Clark, J., Roshwalb, A., & El Dash, N. (2013). Using non-probability online
surveys for exit polling: The case of the 2012 U.S. Presidential Elections. Paper presented at the 68th
annual meeting of the American Association for Public Opinion Research (AAPOR), October 7, 2013.
10
FINAL
Page 17
recruitment costs would be lower than those for phone- or address-based frames because of the low
costs associated with contacting individuals through email.
Consequently, we recommend the use of an Internet-based sample. Further, we recommend pilot
testing both probability and non-probability samples in an attempt to validate the quality of the data
resulting from samples recruited from each source.
Survey Mode
With respect to the survey mode, this report recommends the use of a web-based survey. The
primary reasons being minimization of measurement error and relative cost. Because the survey will
require individuals to record their expenditures and tips and categorize the types of establishment
for at least a day, the amount of information they may potentially have to recall and enter is
substantial. In fact, the shear amount of possible survey branches and associated instruction would
make a paper-/mail-based survey extremely burdensome, increasing the probability of nonresponse,
attrition, or otherwise incomplete, inaccurate documentation of tipping occasions, undermining the
quality of the data. With respect to in-person and phone-based surveys, data quality issues may arise
because of interviewer effects as well as the inability of the respondent to invest time in recalling
accurate information about his or her tipping behavior. By contrast, a computer-based interface can
make finding the type of establishment and entering tipping expenditures relatively easy, through
dynamic branching, instruction, and look-ups.
Another clear advantage of web-based modes is related to cost. In-person, phone-based, and mailbased surveys all have high variable costs which are likely to be substantial due to the large number
of people that will be required to estimate tipping rates on low frequency behaviors like casino
gambling. By contrast, web-based modes can be scaled at relatively low cost.
Survey Design
The primary considerations for survey design are the ability of a specific design to obtain information
on tipping that is representative across both individual and time as well as the degree to which
different designs increase respondent burden, and thus risk nonresponse/attrition and/or poor data
quality. Given these considerations, this report recommends the use of a repeated cross-sectional
design. Given that each individual is only surveyed once, in contrast to a consumer diary (longitudinal
design), where an individual is expected to record the details of tipping occasions multiple times,
respondent burden, and thus the unrepresentativeness of the final sample can be considerably
improved. The one-shot nature of the cross-sectional design may also mitigate the risk that the
survey itself will influence behavior. One of the primary benefits of a longitudinal design, the
potential to make inferences about the causes of individual expenditure behavior, is arguably of
limited relevance in this context as the IRS is primarily concerned with estimating tipping and stiffing
behavior rather than explaining individual differences related to consumer tipping. Finally, the costs
associated with gaining longer term commitments and incentivizing participation can be considerably
higher for longitudinal designs.
With the repeated cross-sectional design, we further recommend a short-recall period to increase the
accuracy of recall, reduce respondent burden, and consequently minimize the role of measurement
error. Shorter recall periods mean that the tipping occasion reported by a given respondent is not
FINAL
Page 18
representative of their yearly tipping. However, because of seasonal differences in tipping behavior
and the frequency of tipping occasions for specific industries, the repeated nature of the survey
increases the potential for variation in both the days of the week and season for tipping occasions in
the sample. This variation then allows for the further development of period-specific estimates of
tipping using poststratification weighting techniques.
To obtain a large enough sample of respondent-day observations to ensure that there is sufficient
variability in low frequency tipping occasions for analysis, the number of respondents used in a
repeated cross-sectional study may have to be very large or the recall length extended with the
implied increase in measurement error. It should be noted that the IRS’ initial tipping study
conducted 30 years ago roughly averaged 60,000 respondent-day observations each year
(approximately 4,200 respondents over a 14 day period each year). Although this sample size was
largely driven by the existing NPD diary data collection this IRS study was attached to, this is roughly
the magnitude that we would expect would be necessary to adequately capture the “opportunity for
tipping” on low frequency behaviors like casino gambling. For example, as seen in Table 4 below, we
estimate needing approximately 76,000 respondent-days to capture 350 casino gambling
occasions. This would entail 76,000 respondents if the recall was 24 hours and fewer if the recall
length was extended. Although we strongly recommend a short recall period, the day or days this
represents should be determined in the piloting stage of the study as prior research does not provide
explicit guidance on this key detail. Table 4 provides estimates for the sample size required to obtain
different frequencies of tipping occasions by sector. A one-day recall is assumed to remain
conservative with the projected estimates.
An alternative to relying on a large nationally representative sample to capture sufficient variation in
infrequent activities is to oversample from regions where the activity is expected to be more
frequent. This strategy would be most suitable for activities like gambling, where establishments are
geographically clustered. Potential complications that result from oversampling arise from the fact
that individuals residing in gambling localities may not be representative of the total U.S. population
with respect to tipping rates. These differences may reflect the fact that gamblers in high-gambling
localities are less likely to be on vacation when they gamble. There may also be systematic
differences with respect to demographic characteristics between high gambling and low gambling
regions that influence gambling-related tipping. In a model-based approach such as MRP, this could
be accounted for by including an indicator for residence in a tipping locality as well as an indicator if
the individual were on vacation when the gambling took place. If the assumptions of the model were
accurate, relevant differences between gamblers in high gambling regions, gamblers in low gambling
regions, and those who gamble on vacation could be accounted for in the final estimate through
post-stratification. An alternative approach that avoids the model based assumptions would be to
calculate a weighted mean tipping rate, where respondents from oversampled localities would be
given a smaller weight such that the weighted sample is representative of the national population
with respect to geography. However, this would result in a smaller effective sample of gamblers and
gambling occasions, which would increase variance in the final estimate, potentially limiting the
benefits of oversampling.
FINAL
Page 19
Table 4. Estimated Annual Occurrence
Estimated Number of Occasions for a given Sample
Size
Occasions
per year
Likelihood
per day
Required
sample
for 350
10,000
30,000
60,000
120,000
240,000
183.0
0.501
698
5,014
15,041
30,082
60,164
120,329
68.0
0.186
1,879
1,863
5,589
11,178
22,356
44,712
Salon**
6.3
0.017
20,373
172
515
1,031
2,062
4,123
Hotels/
motels**
0.6
0.002
223,826
16
47
94
188
375
Taxi/Limo**
0.6
0.002
210,415
17
50
100
200
399
Casino***
1.7
0.005
76,314
46
138
275
550
1101
Eating
out/take-out
fast food*
Eating out/sit
down*
Notes: * Estimates of occasions per day taken from Pearl and McCrohan (1981). ** Estimates of occasions
per day generated from the detailed monthly expenditure file of the Consumer Expenditure Survey13.
***Estimate is an average based on data taken from Shinogle, Norris, Park, Volberg, Haynes, & Stokan (2011)
and Volberg, Nysse-Carris, and Gerstein (2006)14.
Next Steps
This report lays out a general recommended approach; it also leaves open a number of key choices—
such as the use of a probability or nonprobability sample, the period of recording/recall, and the type
of model (MRP versus DM). These choices are critical as they may lead to invalid predictions due to
the data (e.g., selection bias and measurement error) and issues with the model (e.g., included
variables and functional form assumptions). Both issues can be relatively difficult to remedy after
data has been collected. If the data is measured with error or if there is substantial response bias, it
will be unclear what precisely is being modeled and additional rounds of data collection might be
prohibitively expensive.
If the dataset does not contain a large range of potentially observable respondent characteristics,
then testing alternative model specifications might be impossible. For this reason, before settling
upon a final method, we believe it will be important to conduct a set of method studies to examine
the validity and feasibility of our recommended approaches.
For the purposes of calculating the number of occasions per year, a non-zero monthly expenditure on a given
activity is assumed to equate with one occasion in that month for the individual respondent. The number of
occasions per year is then the fraction of person-months with non-zero expenditure multiplied by 12. Note the
assumption that an individual engages in a maximum one expenditure a month likely depresses the number of
occasions. Consequently, these estimates should be viewed as conservative.
14 Shinogle, J., Norris, D. F., Park, D., Volberg, R., Haynes, D., & Stokan, E. (2011). Gambling prevalence in
Maryland: A baseline analysis. Volberg R.A., Nysse-Carris K.L., and Gerstein D.R. (2006). 2006 California
Problem Gambling Prevalence Survey. Estimates based on Table 4.15 on pg. 26 and Table 3 on pg. 31,
respectively. Respondents who list “Past Year Participation,” assumed to gamble at a Casino once per year;
“Monthly Participation,” 12 times per year; “Weekly Participation,” 52 times a year. Note that casino gambling
is legal in both Maryland and California. In addition, California is in close proximity to Nevada. Consequently,
the fraction of the population who reports gambling at a casino, and especially those who visit the casino
frequently, may be larger than in the national population. As a result, Table 4 may inflate the number of casino
gambling occasions that would be obtained in a nationally representative sample.
13
FINAL
Page 20
Instrument Development: The first step, which would focus on instrument development and choice of
recall length, should occur even before a pilot study is initiated. Survey usability testing can be used
to identify problems in the self-administration of the surveys and interpretation of survey items and
instructions. In its most basic form, usability testing is a pretest in which participants are asked to
think aloud while completing the survey instrument and describe their thought process for
determining their answer to the survey item. Hearing participants vocalize this “inner speech”
provides insight into the respondents’ understanding of the question wording, response categories,
and survey organization. After completion of the survey, additional cognitive probing can be done to
explore understanding of concepts that did not emerge during the “think aloud” process. If issues
are identified, the survey can be refined and additional cognitive interviews will be conducted to
verify the changes. In this respect, survey development and usability should be performed iteratively.
One of the primary focuses of this test would be to understand the process through which people
recall their expenditures in order to make a consistent decision on one or multiple days of recall. If
usability testing, for example, demonstrates that user’s performance is similar in both one- and twoday recall, we would suggest including this variable in subsequent pilot testing.
Pilot Testing: Once an instrument or instruments have been developed, we would suggest a pilot test
to further examine the measurement characteristics of the instrument while also examining the use
of probability and nonprobability internet panels. As discussed above, the trade-offs between cost
and quality are not entirely clear between these two sample sources and would benefit from an
empirical test prior to full scale implementation. In addition, to the degree that the usability testing
yields ambiguous results with respect to the effects of recall length on accuracy, recall length may
also be used to define the set of instruments subject to testing in the pilot phase. We would
recommend conducting a test of approximately 20,000 respondent-days (10,000 each method),
within one month, spread over approximately 30 Designated Market Areas (DMA). Initial analyses
would include an examination of relative differences in estimates, indicators, as well as response
characteristics. Although this will provide some evidence as to the consistency of these methods, it
will provide little by way of validation evidence. For this, it will be critical to identify a benchmark data
source.
One potential source of validation data is point of sale (POS) electronic billing records. Organizations
like Restaurant Sciences collect electronic billing records/guest checks to compile useful data for
the restaurant industry. This data, including bill and tip totals, can also be purchased by third parties.
However, because not all tips are paid using a credit or debit card, such estimates will likely provide
an underestimate of total tip income, and therefore cannot be taken as accurate. One way of
generating comparable predictions would be to only model expenditures and tipping rates that are
paid using a debit or credit card. The dependent variables of the tipping and expenditure models
would then be zero if the payment or tip were made using cash, and equal to the amount expended
or the tip rate otherwise.
One issue with this type of validation is that the validation metrics would only apply to electronic
tipping in restaurants, and would not necessarily say much about the ability of the model to predict
nonelectronic, nonrestaurant tip revenue. This type of selection bias would be expected if
restaurants that report electronic payments were systematically different from those that do not,
with respect to their tip rates. This would be the case, for example, if restaurants with the means to
FINAL
Page 21
report their electronic tips were generally better organized. Better organization may be reflected in
better service quality and thus higher tips. Another issue with this type of validation is that electronic
payment data will likely only be available for restaurants, and thus this data has less to say about the
validity for model predictions for nonrestaurant sectors. With these caveats in mind, this out-ofsample data source could provide an extremely valuable source of validation independent of
respondent survey data.
Model Validation: Implementing a MRP approach places additional requirements on the data
collection instrument. Specifically, for the model to be estimated, the sample will likely have to be
stratified geographically in order to obtain variation in the geographic variables. This takes the
unweighted sample away from being representative and thus potentially leads to less precise
national-level estimates. In addition, depending on the proposed model specification, obtaining
information for the individual or geographic variables may increase respondent burden and thus the
risk of non-response or attrition. Consequently, model based approaches, and specifically MRP,
should be validated in the Pilot stage with respect to its ability to predict regional-level tipping rates.
In the spirit of Buttice and Highton (2013), a potential means of validating the model would be to use
the disaggregate mean estimates of tipping from relatively large Restaurant Sciences samples for a
set of approximately 30 geographic regions. The number of observations in the given region will be
larger than in the primary survey, allowing more precise, non-parametric estimates of tipping
behavior in that region. Regions should be chosen for the validation exercise based on dimensions
relevant to tipping rates. Specifically, based on prior literature on tipping, we may believe population
density or proximity to an urban center is associated with tipping rates. In that case, the sample of
validation regions should vary with respect to their level of urbanization. Note that, because the
limited number of observations in the pilot sample, urbanization categories may have to be more
aggregated than for the final sample in order to obtain sufficient variation in the geographic
covariates (i.e. to obtain observations from less dense, rural regions). If the additive assumptions
underlying MRP hold, the MR estimates would be expected to look similar to estimates from these
region-specific surveys. Of course, this latter validation step does not account for potential
systematic measurement error that can affect the accuracy of responses to any survey.
The deviation between the prediction and the ‘observed’ establishment level revenue can be
modeled using establishment-level and locality level covariates to provide further guidance with
respect to sources of bias. Specifically, we can estimate:
|T − T | = βO
+ αG
In this equation T , is the observed tip rate of restaurant o. The left hand side is therefore the
difference between the predicted tip rate of establishments in its sector and locality. We model this
as a function of both establishment-specific characteristics, O, and locality characteristics (G). Note
that the locality definitions and characteristics do not have to match those in the models of
consumer tipping behavior. This is important because it allows us to incorporate additional
geographic information that explains model error. We might find, for instance, that zip-code
tabulation area income explains some of the error in the predicted tip rates. In that case, that would
suggest in the full survey, we would want to ensure that we are able to identify the zip code of the
respondents for the purpose of modeling. We might also find that, within establishment types,
FINAL
Page 22
organizational features such as the size of the establishment affects error. To account for this, for
the final data collection instrument, we might want to ensure that we are able to collect relevant
information about the establishment in order to incorporate those characteristics into our sector
typology for the purposes of either DM or MRP, even if it comes at the price of increased respondent
burden and risk of selection bias.
FINAL
Page 23
Appendix A – Reviewed Articles
Alm, J., & Embaye, A. (2013). Using dynamic panel methods to estimate shadow economies around
the world, 1984–2006. Public Finance Review, 41(5), 510–543. Themes: METHODOLOGY
Alm, J., & Erard, B. (2013). Using public information to estimate informal supplier income. Working
paper. Themes: METHODOLOGY, INDUSTRY/SERVICE
Alm, J., & Jacobson, S. (2007). Using laboratory experiments in public economics. National Tax
Journal, 60(1), 129–152. Themes: METHODOLOGY
Anderson, J. E., & Bodvarsson, O. B. (2005). Do higher tipped minimum wages boost server pay?
Applied Economics Letters, 12, 391–393. Themes: GEOGRAPHY, NATIONAL AVERAGE
TIPPING RATES
Anderson, J. E., & Bodvarsson, O. B. (2005). Tax evasion on gratuities. Public Finance Review, 33,
466–487. Themes: GEOGRAPHY
Ayres, I., Vars, F. E., & Zakariya, N. (2005). To insure prejudice: Racial disparities in taxicab tipping.
The Yale Law Journal, 114, 1613–1674. Themes: RACE/ETHNICITY
Azar, O. H. (2007). The social norm of tipping: A review. Journal of Applied Social Psychology, 37(2),
380–402. Themes: BILL SIZE
Bodvarsson, O. B., & Gibson, W. A. (1997). Economics and restaurant gratuities: Determining tip
rates. American Journal of Economics and Sociology, 56(2), 187–203. Themes:
INDUSTRY/SERVICE, BILL SIZE, GEOGRAPHY
Borzekowski, R., & Kiser, E. K. (2008). The choice at the checkout: Quantifying demand across
payment instruments. International Journal of Industrial Organization, 26(4), 889–902.
Themes: GEOGRAPHY
Boyes, W. J., Mounts, W. S., Jr., & Sowell, C. (2004). Restaurant tipping: Free-riding, social
acceptance, and gender differences. Journal of Applied Social Psychology, 34(12), 2616–
2625. Themes: INDUSTRY/SERVICE, GENDER, INCOME
Brewster, Z. W. (2012). Racialized customer service in restaurants: A quantitative assessment of the
statistical discrimination explanatory framework. Sociological Inquiry, 82(1), 3–28. Themes:
RACE/ETHNICITY
Brewster, Z. W., & Mallinson, C. (2009). Racial differences in restaurant tipping: A labour process
perspective. The Service Industries Journal, 29(8), 1053–1075. Themes: RACE/ETHNICITY
Chapman, G. B., & Winquist, J. R. (1998). The magnitude effect: Temporal discount rates and
restaurant tips. Psychonomic Bulletin & Review, 5(1), 119–123. Themes: BILL SIZE,
INDUSTRY/SERVICE
FINAL
Page 24
Crossley, T. F., & Winter, J. K. (2012). Asking households about expenditures: What have we learned?
In Improving the measurement of consumer expenditures, National Bureau of Economic
Research. Themes: METHODOLOGY
Curtin, R. (2005). Changes in telephone survey nonresponse over the past quarter century. Public
Opinion Quarterly, 69, 87-98.
Davis, S. F., Schrader, B., Richardson, T. R., Kring, J. P., & Kieffer, J. C. (1998). Restaurant servers
influence tipping behavior. Psychological Reports, 83, 223–226. THEMES: GENDER,
GEOGRAPHY
Even, W. E., & Macpherson, D. A. (in press). The effect of the tipped minimum wage on employees in
the U.S. restaurant industry. Southern Economic Journal. Themes: GEOGRAPHY, NATIONAL
AVERAGE TIPPING RATES, METHODOLOGY
Fan, W., & Yan, Z. (2010). Factors affecting response rates of the web survey: A systematic review.
Computers in Human Behavior, 26, 132–139. Themes: METHODOLOGY
Feinberg, R. A. (1986). Credit cards as spending facilitating stimuli: A conditioning interpretation.
Journal of Consumer Research, 13(3), 348–356. Themes: CASH VERSUS CREDIT
Fernandez, G. A. (2004). The tipping point—gratuities, culture, and politics. Cornell Hospitality
Quarterly, 45(1), 48–51. Themes: TIPPING KNOWLEDGE, RACE/ETHNICITY
Filion, K., & Allegretto, S. A. (2011). Waiting for change: The $2.13 Federal subminimum wage
(Briefing Paper No. 297). Economic Policy Institute and Center on Wage and Employment
Dynamics. Themes: GEOGRAPHY, NATIONAL AVERAGE TIPPING RATES, GENDER
Frankel, L. L., & Hillygus, D. S. (2013). Looking beyond demographics: Panel attrition in the ANES and
GSS. Political Analysis. Advance online publication. doi:10.1093/pan/mpt020 Themes:
METHODOLOGY
Frash, R. E., Jr. (2012). Eat, drink, and tip: Exploring economic opportunities for full-service
restaurants. Journal of Foodservice Business Research, 15, 176–194. Themes:
INDUSTRY/SERVICE
Garrity, K., & Degelman, D. (1990). Effect of server introduction on restaurant tipping. Journal of
Applied Social Psychology, 20(2), 168–172. Themes: CASH VERSUS CREDIT
Green, L., Myerson, J., & Schneider, R. (2003). Is there a magnitude effect in tipping? Psychonomic
Bulletin & Review, 10(2), 381–386. Themes: INDUSTRY/SERVICE, BILL SIZE
Greenberg, A. E. (2014). On the complementarity of prosocial norms: The case of restaurant tipping
during the holidays. Journal of Economic Behavior & Organization, 97, 103–112. Themes:
GEOGRAPHY
FINAL
Page 25
Harrison, G. W., & List, J. A. (2004). Field experiments. Journal of Economic Literature, 42(4), 1009–
1055. Themes: METHODOLOGY
Hill, D. J., & King, M. F. (1993). An exploratory investigation into consumer knowledge of tipping
etiquette: Accuracy, antecedents and consequences. In W. Darden & R. Lusch (Eds.),
Proceedings of the symposium on patronage behavior and retail strategy: Cutting edge III
(pp. 121–135). Themes: TIPPING KNOWLEDGE
Jargon, J. (2013, September 4). IRS rule leads restaurants to rethink automatic tips. The Wall Street
Journal. Retrieved from
http://online.wsj.com/news/articles/SB10001424127887323893004579055224175110
910. Themes: SERVICE CHARGE
Keeter, S., Kennedy, C., Dimock, M., Best, J. & Craighill, P. (2006). Gauging the impact of growing
nonresponse on estimates from a national RDD telephone survey. Public Opinion Quarterly,
70, 759-779.
Kerr, P. M., Domazlicky, B. R., Kerr, A. P., & Knittel, J. R. (2006). An objective measure of service and
its effect on tipping. The Journal of Economics, 32(2), 61–69. Themes: INCOME, GENDER,
CASH VERSUS CREDIT
Klee, E. (2004). How people pay: Evidence from grocery store data. Federal Reserve Board.
Retrieved from
http://www.newyorkfed.org/research/conference/2006/Econ_Payments/Klee_b.pdf
Themes: AGE, INCOME
Kleven, H. J., Knudsen, M. B., Kreiner, C. T., Pedersen, S., & Saez, E. (2011). Unwilling or unable to
cheat? Evidence from a tax audit experiment in Denmark. Econometrica, 79(3), 651–692.
Themes: METHODOLOGY
Koku, P. S. (2005). Is there a difference in tipping in restaurant versus non-restaurant service
encounters, and do ethnicity and gender matter? Journal of Services Marketing, 19(7), 445–
452. Themes: INDUSTRY/SERVICE, GENDER, RACE/ETHNICITY
Koku, P. S. (2007). Some significant factors that influence tipping in service encounters outside the
restaurant industry in the United States. Services Marketing Quarterly, 29(1), 23–45.
Themes: INDUSTRY/SERVICE
Lynn, M. (1988). The effects of alcohol consumption on restaurant tipping. Personality and Social
Psychology Bulletin, 14(1), 87–91. Themes: BILL SIZE, INDUSTRY/SERVICE
Lynn, M. (2004). Black-White differences in tipping of various service providers. Journal of Applied
Social Psychology, 34(11), 2261–2271. Themes: INDUSTRY/SERVICE, RACE/ETHNICITY
Lynn, M. (2004). Ethnic differences in tipping: A matter of familiarity with tipping norms.
FINAL
Page 26
Cornell Hospitality Quarterly, 45(1), 12–22. Themes: TIPPING KNOWLEDGE,
RACE/ETHNICITY
Lynn, M. (2006). Geodemographic differences in knowledge about the restaurant
tipping norm. Journal of Applied Social Psychology, 36(3), 740–750. Themes: TIPPING
KNOWLEDGE, RACE/ETHNICITY, AGE, INCOME, EDUCATION, GENDER, GEOGRAPHY
Lynn, M. (2006). Tipping in restaurants and around the globe: An interdisciplinary review. In M.
Altman (Ed.), Handbook of Contemporary Behavioral Economics: Foundations and
Developments (pp. 626–643). M. E. Sharpe Publishers. Themes: CASH VERSUS CREDIT, BILL
SIZE, RACE/ETHNICITY, GENDER
Lynn, M. (2011). Race differences in tipping: Testing the role of norm familiarity. Cornell Hospitality
Quarterly, 52(1), 73–80. Themes: RACE/ETHNICITY
Lynn, M. (2012). The contribution of norm familiarity to race differences in tipping:
A replication and extension. Journal of Hospitality & Tourism Research. Advance online
publication. doi:10.1177/1096348012451463. Themes: RACE/ETHNICITY
Lynn, M. (2013). A comparison of Asians’, Hispanics’, and Whites’ restaurant tipping. Journal of
Applied Social Psychology, 43(4), 834–839. Themes: BILL SIZE, RACE/ETHNICITY
Lynn, M., & Gregor, R. (2001). Tipping and service: The case of hotel bellmen. International Journal
of Hospitality Management, 20, 299–303. Themes: INDUSTRY/SERVICE
Lynn, M., & Latane, B. (1984). The psychology of restaurant tipping. Journal of Applied Social
Psychology, 14(6), 549–561. Themes: CASH VERSUS CREDIT, GENDER, BILL SIZE
Lynn, M., & McCall, M. (2000). Gratitude and gratuity: A meta-analysis of research on the servicetipping relationship. The Journal of Socio-Economics, 29(2), 203–214. Themes:
METHODOLOGY
Lynn, M., & McCall, M. (2000). Beyond gratitude and gratuity: A meta-analytic review of the
predictors of restaurant tipping. Working paper, School of Hotel Administration, Cornell
University. Themes: CASH VERSUS CREDIT, BILL SIZE, GENDER, RACE/ETHNICITY
Lynn, M., & Thomas-Haysbert, C. D. (2003). Ethnic differences in tipping: Evidence, explanations, and
implications. Journal of Applied Social Psychology, 33(8), 1747–1772. Themes:
RACE/ETHNICITY, AGE, INCOME, EDUCATION, GENDER
Lynn, M., & Williams, J. (2012). Black-White differences in beliefs about the U.S. restaurant tipping
norm: Moderated by socio-economic status? International Journal of Hospitality
Management, 31(3), 1033–1035. Themes: RACE/ETHNICITY
Lynn, M., Zinkhan, G., & Harris, J. (1993). Consumer tipping: A cross-country study. Journal of
Consumer Research, 20, 478–488. Themes: GEOGRAPHY
FINAL
Page 27
McCall, M., & Belmont, H. J. (1996). Credit card insignia and restaurant tipping: Evidence for an
associative link. Journal of Applied Psychology, 81(5), 609–613. Themes: CASH VERSUS
CREDIT
McCrohan, K. F., & Pearl, R. B. (1983, August). Tipping practices of American households: Consumer
based estimates for 1979. 1983 Program and Abstracts Joint Statistical Meetings, Toronto,
CA. Themes: NATIONAL AVERAGE TIPPING RATES, GEOGRAPHY, INCOME, CASH VERSUS
CREDIT
McCrohan, K. F., & Pearl, R. B. (1991). An application of commercial panel data for public policy
research: Estimates of tip earnings. Journal of Economic and Social Measurement, 17, 217–
231. Themes: NATIONAL AVERAGE TIPPING RATES, CASH VERSUS CREDIT,
INDUSTRY/SERVICE, GEOGRAPHY
Morran, C. (2013, September 5). Are these the final days of automatic 18% tips at restaurants?
Consumerist. Retrieved from http://consumerist.com/2013/09/05/are-these-the-final-daysof-automatic-18-tips-at-restaurants/. Themes: SERVICE CHARGE
Neuman, S. (2013, September 5). IRS to count automatic gratuities as wages, not tips. NPR.
Retrieved from http://www.npr.org/blogs/thetwo-way/2013/09/05/219290573/irs-tocount-automatic-gratuities-as-wages-not-tips. Themes: SERVICE CHARGE
Noll, E., & Arnold, S. (2004). Racial differences in tipping: Evidence from the field. Cornell
Hospitality Quarterly, 45, 23–29. Themes: RACE/ETHNICITY
Papp, T. G., & Burkhammer, A. L. (2001, March). An investigation of server posture and gender on
restaurant tipping. Paper presented at the 22nd Annual Industrial Organizational Psychology
and Organizational Behavior Graduate Student Conference, Pennsylvania State University.
Themes: GENDER
Parker, J. A., Souleles, N. S., & Carroll, C. D. (2012). The benefits of panel data in consumer
expenditure surveys. National Bureau of Economic Research. Themes: METHODOLOGY,
INDUSTRY/SERVICE
Paul, P., & Gardyn, R. (2001). The tricky topic of tipping. American Demographics, 23(5), 10–11.
Themes: NATIONAL AVERAGE TIPPING RATES, INDUSTRY/SERVICE, GEOGRAPHY
Pearl, R. B. (1984). A survey approach to estimating the tipping practices of consumers. Special
report on regression analysis to the Internal Revenue Service under contract TIR-81-21,
Survey Research Laboratory, University of Illinois, Champaign, IL. Themes: GEOGRAPHY
Pearl, R. B., & McCrohan, K. F. (1984). Estimates of tip income in eating places, 1982. Statistics of
Income Bulletin, 3(4), 49–53. Themes: METHODOLOGY, NATIONAL AVERAGE TIPPING RATES,
INDUSTRY/SERVICE
FINAL
Page 28
Pearl, R. B., & Sudman, S. (1983). A survey approach to estimating the tipping practices of
consumers. Final report to the Internal Revenue Service under Contract TIR-81-21, Survey
Research Laboratory, University of Illinois. Themes: NATIONAL AVERAGE TIPPING RATES,
CREDIT VS CASH, INCOME, GEOGRAPHY, INDUSTRY/SERVICE
Pearl, R. B., & Vidmar, J. (1988). Tipping practices of American households in restaurants and other
eating places: 1985–86. Supplementary report to the Internal Revenue Service under
Contract TIR 86-279, Survey Research Laboratory, University of Illinois, Champaign, IL.
Themes: CASH VERSUS CREDIT, GEOGRAPHY, INCOME, EDUCATION, AGE
Pearl, R. B., & Vidmar, J. (1988). Tipping practices of American households in restaurants and other
eating places: 1985–86. Supplementary report to the Internal Revenue Service under
Contract TIR 86-279, Survey Research Laboratory, University of Illinois, Champaign, IL.
Themes: GEOGRAPHY, INDSUTRY/SERVICE
Pew Research Center. (2012). Assessing the representativeness of public opinion surveys. 125. http://www.people-press.org/2012/05/15/assessing-the-representativeness-of-publicopinion-surveys/
Rind, B. (1996). Effects of beliefs about weather conditions on tipping. Journal of Applied Social
Psychology, 26(2), 137–147. Themes: INDUSTRY/SERVICE
Sanchez, A. (2002). The effect of alcohol consumption and patronage frequency on restaurant
tipping. Journal of Foodservice Business Research, 5(3), 19–36. Themes: RACE/ETHNICITY,
AGE, CASH VERSUS CREDIT, INDUSTRY/SERVICE
Schwer, R. K., & Daneshvary, R. (2000). Tipping participation and expenditures in beauty salons.
Applied Economics, 32, 2023–2031. Themes: SERVICE/INDUSTRY, INCOME, AGE, GENDER
Seiter, J. S., & Weger, H., Jr. (2013). Does a customer by any other name tip the same? The effect of
forms of address and customers’ age on gratuities given to food servers in the United States.
Journal of Applied Social Psychology, 43, 1592–1598. Themes: METHODOLOGY, AGE
Simpson, H. (1997). Tips and excluded workers: The New Orleans test. Compensation and Working
Conditions, Bureau of Labor Statistics, 32–36. Themes: SERVICE/INDUSTRY, GEOGRAPHY
Speer, T. (1997). The give and take of tipping. American Demographics, 19(2), 51–54. Themes:
INDUSTRY/SERVICE, GEOGRAPHY, INCOME, GENDER.
Star, N. (1988). The international guide to tipping: When, where, and how much to tip in the U.S. and
around the world. New York, NY: Berkley Books. Themes: INDUSTRY/SERVICE, GEOGRAPHY
Thomas-Haysbert, C. D. (2002). The effects of race, education, and income on tipping behavior.
Journal of Foodservice Business Research, 5(2), 47–60. Themes: INDUSTRY/SERVICE,
RACE/ETHNICITY, INCOME, EDUCATION
FINAL
Page 29
Appendix B – Annotated Citations
Alm, J., & Embaye, A. (2013). Using dynamic panel methods to estimate shadow economies around
the world, 1984–2006. Public Finance Review, 41(5), 510–543.
METHODOLOGY: Article uses a model-based approach to estimate the size of the shadow economy
for 111 countries across the world for the period 1984 to 2006. The shadow economy is defined as
the production of goods and services that are not included in government accounts. To estimate the
shadow economy, the authors model the demand for currency, defined as the amount of cash over
M2. Cash-based transactions are assumed to be relatively easy to hide from the state. Consequently,
economies dominated by shadow activities are expected to also be cash-based, all other things
being equal. Cash demand is modeled as a function of proxies for levels of development such as
urbanization and per capita income as well as country-level characteristics that are thought to
influence the incentive to conceal income from the government (thus increasing the demand cash),
including bureaucratic quality, the tax rate, and the level of inflation. The use of panel data provides
more observations and thus degrees of freedom than prior country-specific, time-series based
analysis of cash demand while also allowing the authors to correct for endogeneity in the predictors.
The resulting model is used to predict cash demand as well as a counterfactual set of predictions
where there is no incentive to hide income (when government quality, and thus enforcement is at its
maximum, the tax rate is zero, and there is no inflation). The predictions for cash demand where
there is no shadow economy is subtracted from the total predicted cash demand to arrive at an
estimate of cash demand that is due to tax evasion. This estimate is then multiplied by money
velocity and divided by GDP to arrive at an estimate of the shadow economy as a fraction of GDP.
The results indicate a negative association between the size of the shadow economy and the level of
development. However, the mean for OECD (Organization for Economic Cooperation and
Development) countries across the entire period is still a substantial 16.9%, and 13.3% for the
United States in 2006.
Alm, J., & Erard, B. (2013). Using public information to estimate informal supplier income. Working
paper.
METHODOLOGY: Authors use responses from the 2001 Current Population Survey (CPS) to estimate
informal supplier (self-employment) income and tax noncompliance. Specifically, they develop
estimates of national informal supplier income using income information provided by self-employed
respondents working in 11 industry categories in which informal suppliers will be prominent. To
estimate the income of “Food Caterers and Roadside Stands,” the authors use responses from the
Bureau of Labor Statistics’ Consumer Expenditure Survey (CES). They then compare these estimates
FINAL
Page 30
with income reported to IRS National Research Program (NRP) in these industries to arrive at
industry category-level estimates of tax noncompliance. Supplementary CPS surveys were used to
identify second jobs and second job income was imputed based on the assumption that secondary
income comprised 26.5% of wages. This fraction was in turn estimated using a subsample of
respondents who reported income for both jobs. In addition, self-employed, informal income
misclassified as wages was assumed to comprise 4.08% of wages. The resulting estimates of total
self-employed income ($156.4 billion) for the 11 CPS industry categories exceeded reported income
estimated from the NRP ($50.9 billion), but was lower than an estimate of total income derived from
NRP data (reported income + audit detected + estimated non-detected).
Alm, J., & Jacobson, S. (2007). Using laboratory experiments in public economics. National Tax
Journal, 60(1), 129–152.
METHODOLOGY: Provides a review of literature using laboratory experiments in the field of public
economics. The article lays out the requirements for the successful expectation of an experiment
studying the effect of incentives on behavior, including control over the experimental environment
such that monetary incentives be explicitly linked to behavior, that instructions are clear, that the
experiment not be too long or complicated, and that instructions should not use terminology that
hints at the research question that the experiment addresses, which the authors argue could
potentially influence the subjects’ behavior. Common criticisms of experiments include the argument
that the mainly university student subject pool of most laboratory experiments is not representative
of the wider population whose behavior and motivations the experiment is trying to analyze/explain
(though the authors argue this concern is unfounded), that subjects modify their behavior as a result
of the awareness that they are participating in an experiment, and that certain factors that affect
behavior in the real world, such as the threat of prison time, cannot be plausibly simulated in a
laboratory setting. Consequently, results of an experiment may not generalize outside of the
laboratory setting. The article also discusses the use of laboratory experiments to address questions
related to the determinants of tax compliance behavior. These experiments typically find that audits
increase compliance (though there are diminishing marginal returns as the audit rate increases),
that the fine rate increases compliance (though the effect is small), and that higher marginal tax
rates lead to lower compliance. Higher income is found to lead to greater compliance. Targeted
audits have been found to be more effective in increasing compliance than random audits.
Democratic participation and an effective social norm supporting tax compliance increase individual
compliance.
FINAL
Page 31
Anderson, J. E., & Bodvarsson, O. B. (2005). Do higher tipped minimum wages boost server pay?
Applied Economics Letters, 12, 391–393.
DESIGN OVERVIEW: Authors investigate if there is any difference in server pay between states with
varying levels of subminimum wages and tip credits for tipped staff. A probit analysis was used, and
there were 100 total observations in the analysis: one observation for waiters and one observation
for bartenders for each of the 50 states (Washington, D.C., was not mentioned in the article and was
likely excluded). Data was pulled from the Bureau of Labor Statistics “Wages by Area and
Occupation” file (additional data was pulled from the U.S. Census Bureau, the National Restaurant
Association, and the Bureau of Economic Analysis). Analysis controlled for the percentage of firms
exempted from state and federal minimum wage laws, and restaurants’ revenue as a proportion of
the GDP, in addition to other control variables such as age and whether the state has a state income
tax.
AVERAGE TIPPING RATES: OLS regression findings indicate that there was a very small difference
between states with no minimum wage or tip credit versus states with no tip credits and wages that
exceed federal standards, but that overall there was no noticeable difference between the minimum
wage of waiters and reported wages.
Anderson, J. E., & Bodvarsson, O. B. (2005). Tax evasion on gratuities. Public Finance Review, 33,
466–487.
DESIGN OVERVIEW: The authors used state-level data from the Bureau of Labor Statistics (BLS) to
determine if total reported pay is affected by factors that are hypothesized to affect underreporting
of tips. The BLS’ Occupational Employment Statistics (OES) surveys are used to estimate the mean
and median hourly pay for over 750 occupations, and the authors used restaurant-related
occupations for testing their model. Two variables are included to proxy average customer tipping
rate (i.e., premium full-service restaurants as a percentage of full restaurants in the state and the
percentage of each state’s population living in urban areas). They also included several control
variables to account for slight differences in job characteristics and locations.
GEOGRAPHY: Reported pay is higher in areas with a higher tipped minimum wage and in states with
no income tax. IRS audit rates do not appear to have an effect on reported pay by restaurant
employees. The most important result from their analyses was that higher tax rates raise the
employee’s reported pay, such that one percentage point increase in a state’s minimum income tax
rate results in servers reporting 13 cents more in pay.
FINAL
Page 32
Ayres, I., Vars, F. E., & Zakariya, N. (2005). To insure prejudice: Racial disparities in taxicab tipping.
The Yale Law Journal, 114, 1613–1674.
DESIGN OVERVIEW: 12 taxicab drivers (6 Black, 4 White, and 2 “other minorities”) completed
surveys immediately after dropping off customers for a total of 1,066 completed surveys. Tips were
calculated by subtracting the fare from the total cost of the ride. Drivers recorded sex, race, age,
passenger dress (proxy for wealth), and driver experience. They also recorded other interaction
characteristics, including whether they paid with cash.
RACE/ETHNICITY: White drivers were tipped 61% more than Black drivers (20.3% versus 12.6%) and
64% more than “other minority” drivers (20.3% versus 12.4%). Black drivers were 80% more likely to
be stiffed than White drivers (28.3% versus 15.7%) and “other minority” drivers were 131% more
likely (36.4% versus 15.7%). The mean tipping percentage of Black customers was 42% of the mean
tipping percentage of White customers (9.2% versus 21.6%). Hispanic customers’ mean tipping
percentage was just over half of White customers’ mean tipping percentage (12.0% versus 21.6%).
Asians tipped 75% of the White customers’ mean tipping percentage (16.2% versus 21.6%). White
customers stiffed the driver (10.6%) less frequently than Blacks (39.2%), Hispanics (34.3%), and
Asians (15.8%). Using a regression analysis and controlling for random driver effects, time, manner,
and place effects, Black drivers are tipped 9.1% less than White drivers. In the most complete
regression, Black passengers tipped 9% less than White passengers.
Azar, O. H. (2007). The social norm of tipping: A review. Journal of Applied Social Psychology, 37(2),
380–402.
DESIGN OVERVIEW: A literature review of various tipping-related areas, including both theoretical
motivations behind tipping behavior and empirical studies on the subject. Areas of focus include the
relationship between service quality and tipping behavior, patronage frequency, bill size, service
quantity, and other variables.
Bodvarsson, O. B., & Gibson, W. A. (1997). Economics and restaurant gratuities: Determining tip
rates. American Journal of Economics and Sociology, 56(2), 187–203.
DESIGN OVERVIEW: Authors test several hypothesis derived from economic theory on the
determinants of tipping. Data is based on 697 respondents to a survey conducted in 7 Minnesota
restaurants. Data collected included bill and tip size, number of food and drink items ordered,
number of people at the table, whether the respondent visited the establishment at least once a
month, and an assessment of service quality. To account for potential measurement error in tipping
due to social desirability bias, the tip rates reported by customers were passed by the servers who
FINAL
Page 33
gave an assessment of their plausibility. Their answer was affirmative. Tip amounts and tip rates
were analyzed using both descriptive statistics and multivariate regression analysis.
INDUSTRY/SERVICE: Tip rates varied across establishments; establishments that were licensed to
serve alcohol received higher tips.
BILL SIZE: Tip amount was positively related to tip amount bill size, and only marginally related to
service quality, consistent with the existence of a lower bound on the amount customer’s tip.
GEOGRAPHY: Tips (amounts and rates) were higher in restaurants located in St. Paul than in St.
Cloud, consistent with tips being higher in larger urban areas.
Borzekowski, R., & Kiser, E. K. (2008). The choice at the checkout: Quantifying demand across
payment instruments. International Journal of Industrial Organization, 26(4), 889–902.
DESIGN OVERVIEW: Article examining roughly 1,500 households over the course of three months
from March through May of 2004. The survey was conducted as part of the University of Michigan
Survey of Consumers, a telephone-based survey that covers various aspects of consumer behaviors
and attitudes. Various scenarios were presented to respondents, including one suggesting that a
“flash” debit service has been introduced to see changes in behavior and another that attempts to
“age” the cohort to see changes in behavior. Overall, it was reported that debit cards were
overcoming the use of cash and checks for consumers. However, given that the scenarios presented
ask about usage when purchasing items from a supermarket, payment methods will likely be very
different for tipping situations, because checks are often not appropriate or accepted for tipping
situations or establishments.
GEOGRAPHY: Of the four regions, the West region had the highest predicted market share of debit
and credit usage (53% for the two) compared with 46.5% for the South, 41.6% for the Northeast, and
38.4% in the Midwest.
Boyes, W. J., Mounts, W. S., Jr., & Sowell, C. (2004). Restaurant tipping: Free-riding, social
acceptance, and gender differences. Journal of Applied Social Psychology, 34(12), 2616–2625.
DESIGN OVERVIEW: Study investigating tipping behavior using in-person survey intercepts at 18
different restaurant locations, 10 surveys per restaurant. Analysis was used to determine if social
acceptance and free-riding influence tipping behavior. Additional variables included customer
gender. In-person intercepts were used at each restaurant, asking respondents various questions
about their demographics, the size of their party, whether they are a local resident of the area (used
as a proxy to determine if they were a repeat customer), how often they eat out and how often they
have eaten at the restaurant in the past month, and ratings about the quality of their meal.
FINAL
Page 34
Respondents were also asked if they had any alcohol or not. Surveys were only asked during dinner
hours to maintain consistency; roughly 90% of respondents agreed to respond to the survey, and a
third of surveys were confirmed with the servers of the restaurant for accuracy.
Furthermore, restaurants were classified into four different restaurant types
INDUSTRY/SERVICE: Alcohol consumption had a significant impact on the tipping percentage such
that respondents who indicated they had consumed alcohol left higher tips.
GENDER: Men tipped less than women, even when other factors were held constant. In addition,
men’s tips were found to be more significantly influenced by party size.
INCOME: Higher levels of income were related to higher tipping rates.
Brewster, Z. W. (2012). Racialized customer service in restaurants: A quantitative assessment of the
statistical discrimination explanatory framework. Sociological Inquiry, 82(1), 3–28.
DESIGN OVERVIEW: A paper survey was given to servers from a sample of 18 chain-style restaurants.
Overall, 200 completed surveys were gathered. The aim of the survey was to determine whether
servers discriminate against customers of various races (based on questions asking if the quality of
service will vary by race). The author acknowledged that explicit questions about racial tendencies in
this way will lead to some lack of variability in reporting behaviors because people will wish to report
in a way consistent with a social-desirability bias. Respondents were given a series of five scenarios
(in which the customer race was held constant as Black customers in various configurations) and
asked whether the customers were good or bad tippers (on a 5-point scale). Respondents were also
asked what they considered to be good and bad attributes of diners and to provide ratings of the
dining behaviors of the Black individuals in the scenarios. Servers were also asked if they preferred
to serve various situations (such as groups with or without children, social classes of their clients,
etc.).
RACE/ETHNICITY: Overall, nearly 1 in 5 servers reported an explicit preference for serving White
clients. In addition, on the 4-point scale regarding service-quality discrimination (1 = never and 4 =
always), the mean score was 1.49, indicating that a reasonable number of servers were willing to
report some discriminatory behaviors against their customers. Findings seem to indicate that once
discriminatory tendencies toward other groups are taken into consideration (such as children, etc.),
that servers who report more positivity toward Blacks are less likely to discriminate against them in
their service. However, given their use of a proxy variable for discriminatory behaviors, the findings
have to be considered with caution.
FINAL
Page 35
Brewster, Z. W., & Mallinson, C. (2009). Racial differences in restaurant tipping: A labour process
perspective. The Service Industries Journal, 29(8), 1053–1075.
DESIGN OVERVIEW: Literature review of two theoretical frameworks that try to explain the reasons
for lower tipping behavior among Blacks. The two frameworks that are discussed are that (1) Blacks
are unaware of tipping norms, hence leading to lower tipping behavior and (2) that Blacks tip at
lower rates because service providers (i.e., waiters) treat Black customers poorly because they
anticipate poor tips, creating a cyclical problem.
Chapman, G. B., & Winquist, J. R. (1998). The magnitude effect: Temporal discount rates and
restaurant tips. Psychonomic Bulletin & Review, 5(1), 119–123.
DESIGN OVERVIEW: Subjects included 50 undergraduate students participating for course credit.
Subjects completed a questionnaire that included two sections: an intertemporal choice and three
tipping scenarios. The tipping scenarios comprised a taxi ride, a restaurant dinner, and a haircut.
Each scenario included a brief description and asked how much the participant would tip based on
bill size. They were presented with four different magnitudes for each tipping setting. Participants
were also asked how much they had paid and tipped the last time they had used each of the service
scenarios.
INDUSTRY/SERVICE AND BILL SIZE: Tip percentages decreased with bill magnitude for each of the
three tipping scenarios, but ANOVA revealed a significant effect of magnitude for the haircut and
restaurant dinner scenarios. The magnitude effect (i.e., tip percentages decrease significantly as the
bill size increases) was found to be present in both of these scenarios, indicating that participants
reported leaving bigger tips for smaller bills.
Crossley, T. F., & Winter, J. K. (2012). Asking households about expenditures: What have we
learned? In Improving the measurement of consumer expenditures, National Bureau of Economic
Research.
METHODOLOGY: Article reviews literature examining the benefits and drawbacks of different
methods of collecting household expenditure data through surveys. There is little evidence to
suggest the superiority of single survey modes (face-to-face interviews, telephone interviews, selfadministered questionnaires); while self-administered questionnaires may increase response rates
and quality by allowing respondents time to recall their expenditure patterns and reduce
confidentiality relative to modes requiring an immediate response to the interviewer, interviewers
may be able to provide more assistance to respondents who have issues with question
comprehension. Recall surveys may lead to downward biases in reported expenditure due to poor
FINAL
Page 36
recall relative to diaries as well the inclusion of expenditures from before the survey reference
period, but diaries may lead to respondent attrition and a decline in the accuracy of responses as
time passes due to the greater imposition on respondents. This may lead to a downward bias in
expenditure estimates in diaries versus recall surveys, and has been found to be problematic in the
case of expenditures on food. Expenditure data collected from diaries with short time frames may
also show greater variance due to the fact that respondents report expenditures as they are made,
and there may be a large degree of variance in expenditures in short time periods, particularly with
respect to infrequent expenditure categories. The keeping of diaries may also influence respondents’
expenditure patterns, resulting in biased estimates of population expenditure patterns. Diary
respondents may also tend to aggregate different expenditures when they are made at the same
time.
The format of survey questions has also been found to have an effect on data quality; open-ended
formats lead to rounding of responses, while closed formats may lead respondents to choose
categories that they perceive as reflecting their relative expenditures (high spender, high-spending
bin) as opposed to their true expenditure. Aggregated expenditure categories tend to lead to lower
total expenditure estimates, perhaps due to an inability of respondents to recall every type of
expenditure. On the other hand, more disaggregated expenditure categories may put a greater
burden on respondents and thus lead to lower quality (less accurate) responses. Using single
respondents to solicit information on household expenditures may lead to lower-quality estimates,
but using multiple respondents per household may place a greater burden on the household and
consequently result in lower response rates. Incentives for completing the survey or diary may
increase both response rates and data quality. Data quality can also be improved by asking
respondents to reassess their expenditure estimates when they are inconsistent with previously
given information, such as total budget.
Davis, S. F., Schrader, B., Richardson, T. R., Kring, J. P., & Kieffer J. C. (1998). Restaurant servers
influence tipping behavior. Psychological Reports, 83, 223–226.
DESIGN OVERVIEW: Twenty-eight servers from a pair of restaurants (one in a small Midwestern town,
12 servers; and another in an urban area, 16 servers) recorded their tips for a four-week period
while alternating whether they stood or squatted by tables in order to determine if that increased tip
size. Aside from varying the squat/standing procedure, other descriptive measures including whether
the meal was for lunch or dinner and what the gender of the server was were maintained for
analysis. Of the 12 servers in the rural area, 7 were women and 5 were men, and there was an even
8/8 split in the urban area. Servers maintained all of the recordings, including the dollar amount of
FINAL
Page 37
the meal and the tip that they received. Possible issues with this study are that there is no mention
of an incentive for the servers to maintain accurate record-keeping and that they might be
misreporting their tips as a whole.
GEOGRAPHY: The study found that people from urban areas tipped significantly more than those
from rural areas, but because the servers were not able to determine any kind of socioeconomic
variables such as income or education, this might be a spurious effect caused by other variables.
GENDER: Female servers received significantly greater tips than male servers (15.6% compared with
14.1%, though this was the smallest of the significant findings).
Even, W. E., & Macpherson, D. A. (in press). The effect of the tipped minimum wage on employees in
the U.S. restaurant industry. Southern Economic Journal.
DESIGN OVERVIEW: Two sets of regression analyses were run (specifically, the regressions were a
version of “difference-in-difference estimation”—additional details and citations about this regression
method can be found in the article): one using data from the Quarterly Census of Employment and
Wages (QCEW) and the other using data from the Census Bureau’s Current Population Survey from
1990 through 2011. The regression equation controlled for changes due to season and various
demographic variables that would change earnings in the industry, and accounted for both the
federal minimum wage and the subminimum wage, among other factors.
Both data sources have their advantages. The QCEW data is pulled from unemployment insurance
reports, ensuring essentially mandatory compliance for reporting. However, this data does not
provide work hours for workers, nor does it give characteristics of the workers. CPS data, on the
other hand, provides such characteristics, but because of methodology the sample for certain
industries and states can be quite small and introduce the possibility of error. Both data sets were
acknowledged to have specific strengths and weaknesses for their analysis.
NATIONAL AVERAGE TIPPING RATES: Findings from analyses of both data sources indicate that the
salary of tipped workers does increase along with minimum wage increase, though the QCEW data
points out that this only occurs among full-service restaurants and is not seen among limited service
restaurants. Further findings indicate that increases in the minimum wage for tipped employees has
a negative influence on the employment of this population and that raises in this minimum wage
lead to reduced hours worked per week in addition to higher wages.
Fan, W., & Yan, Z. (2010). Factors affecting response rates of the web survey: A systematic review.
Computers in Human Behavior, 26, 132–139.
FINAL
Page 38
METHODOLOGY: Article reviews literature addressing factors that affect web response rates. Factors
related to survey content include: the sponsor of the survey, with response rates being higher when
the survey’s sponsor is an academic or government agency; the content of the survey, with surveys
asking questions concerning highly salient issues typically receiving higher response rates than
those whose subject is less relevant to potential respondents; the length of the survey, with longer
surveys having lower response rates. Sample design and contact methods also influence response
rates: web panel designs typically yield higher response rates than single-shot surveys, while emailbased contact can result in low response rates because of spam filters. However, the use of
personalized messages, prenotifications, and reminders can raise response rates. Empirical work
examining the influence of incentives (such as an electronically mailed gift certificate) on response
rates has generally found small (or even negative) effects on participation. The survey frame also
affects response rates, with surveys of the general population generally yielding lower response rates
than surveys of specific populations such as employees, though top managers are less likely to
respond than lower-level managers/employees. Populations with low socioeconomic status are less
likely to respond because of limited Internet access, though this effect persists even after controlling
for such access. The personalities of potential respondents also influence response rates, with more
conscientious individuals having a greater propensity to respond.
Feinberg, R.A. (1986). Credit cards as spending facilitating stimuli: A conditioning interpretation.
Journal of Consumer Research, 13(3), 348–356.
DESIGN OVERVIEW: One hundred and thirty-five customers were observed at random intervals over a
one-week span at a local restaurant. Servers recorded party size, check amount, mode of payment,
and amount of tip. The author also conducted four experiments investigating characteristics of credit
card spending, but none of them dealt with tipping or the service industry.
CASH VERSUS CREDIT: A 2 (payment method) x 4 (check size divided into quartiles) ANOVA revealed
that when credit card stimuli were present, customers left a significantly higher tip. For each quartile
of check size, customers paying with credit cards provided higher tips. Credit card–paying customers,
on average, left a tip that was 16.95% of the total bill, while cash-paying customers left a tip that was
14.95% of the total bill.
Fernandez, G. A. (2004). The tipping point—gratuities, culture, and politics. Cornell Hospitality
Quarterly, 45(1), 48–51.
DESIGN OVERVIEW: Discussion about knowledge of tipping behavior, how the knowledge is passed
on, and a discussion about what underlies the racial differences in tipping. Some topics that are
FINAL
Page 39
discussed are underlying psychological issues that might be at work within the Black community,
including how the segregation of service in restaurants in the past might be the cause of certain
behaviors in the present. The author calls for a national study to look at this subject, with enough of
a sample to investigate racial differences across different areas with the sufficient detail needed to
draw concrete conclusions.
Filion, K., & Allegretto, S. A. (2011). Waiting for change: The $2.13 Federal subminimum wage
(Briefing Paper No. 297). Economic Policy Institute and Center on Wage and Employment Dynamics.
DESIGN OVERVIEW: Analysis was conducted using the Census Bureau’s Current Population Survey
from 2008–2009. Descriptive results of reported wages were split by several demographic groups,
including worker gender, race, age, education, and across various states with differing levels of
wages for tipped employees.
NATIONAL AVERAGE TIPPING RATES: Overall, it was found that states with higher levels of
subminimum wages had higher reported hourly wages for waiters and tipped workers than states
with lower tipped minimum wages for tipped workers. However, it is worth noting that the median
wage of workers was higher in those states overall, indicating that the relative affluence of those
states are driving these changes.
GENDER: Demographic splits indicate that while females constitute the majority of tipped workers
and waiters (72.9% and 72.4%, respectively) they earn less on average than male workers,
particularly among waiters ($9.04 for females and $9.87 for males).
Frankel, L. L., & Hillygus, D. S. (2013). Looking beyond demographics: Panel attrition in the ANES
and GSS. Political Analysis. Advance online publication. doi:10.1093/pan/mpt020
METHODOLOGY: Article examines the determinants of respondent attrition in the American National
Election Studies (ANES), an online panel survey, and the General Social Survey (GSS), a face-to-face
interview panel survey using logit regression. Both respondent demographics and survey experience
characteristics are included as predictors of attrition. Among the demographic characteristics, age,
education, and employment were negatively associated with attrition in the ANES, while non-English
preferences and the number of young children were positively associated with attrition. Age and
education had a statistically significant negative association in GSS, while foreign born and single
member household status were positively associated with the probability of attrition. Among the
survey experience characteristics, respondents to the ANES who reported a monetary motivation,
had a negative experience, and/or took a long time to complete the survey were more likely to
attrite, as were those who refused to answer the survey in the first wave. For the GSS, interviewer
FINAL
Page 40
experience was found to be negatively associated with the probability of attrition, and respondents
who were interviewed by females were less likely to attrite.
Frash, R. E, Jr. (2012). Eat, drink, and tip: Exploring economic opportunities for full-service
restaurants. Journal of Foodservice Business Research, 15, 176–194.
DESIGN OVERVIEW: The author pooled point-of-sale (POS) processed guest checks and their
associated credit card checks from two restaurants (one fine dining establishment and one casualtheme full-service restaurant). One hundred and fifty checks were randomly selected from each
restaurant’s weekly pool and each check had to meet several conditions, namely that the checks
had to include both food and alcoholic beverages, be from restaurants’ dining rooms (i.e., no checks
from the bar), be tendered after 5:00 p.m., paid by only one party, and not include any promotional
or employee discounting. From the guest and credit card checks, the author recorded reliably
accurate information for the guest check dollar amount, percentage of the guest check spent on
alcoholic beverage purchases, server’s gender, dollar tip amount, and tip percentage. Time the guest
check was rendered and day of the week were also recorded.
INDUSTRY/SERVICE: Two hundred and ninety-seven guest checks were included in the final analysis
from the two restaurants. The median percentage of the guest check that was attributable to
alcoholic beverages was 26.8%, the median guest check was $40.67, and the median tip
percentage was 20.6%. A multiple regression was performed to predict the tip percentage from
percentage of the guest check used on alcoholic beverages. A positive relationship was found
between the percentage of guest check attributable to alcoholic beverages and the tip percentage of
the whole guest check.
Garrity, K., & Degelman, D. (1990). Effect of server introduction on restaurant tipping. Journal of
Applied Social Psychology, 20(2), 168–172.
DESIGN OVERVIEW: Forty-two, 2-person parties that ordered a Sunday brunch at a restaurant were
randomly assigned into two interaction conditions. In one condition, the server greets the customer
while introducing herself; in the other condition, the server just greets the customer.
CASH VERSUS CREDIT: Customers that used a credit card as a form of payment left, on average,
larger tips than those using cash (22.6% versus 15.9%).
Green, L., Myerson, J., & Schneider, R. (2003). Is there a magnitude effect in tipping? Psychonomic
Bulletin & Review, 10(2), 381–386.
FINAL
Page 41
DESIGN OVERVIEW: In order to determine if there is a magnitude effect in tipping (i.e., as bill size
increases, percentage tipped decreases), researchers had two taxicab drivers, four restaurant
servers (from two restaurants), and four hair stylists (from two salons) record the total bill size and
the amount of the tip for each customer over several months. This amounted to nearly 1,000 service
encounters.
INDUSTRY/SERVICE AND BILL SIZE: The author’s regressed percentage tipped on the total amount of
the bill for all bills less than $100. The regression slopes were negative in each of the six cases (two
taxicabs, two hair salons, and two restaurants), indicating a magnitude effect. Linear regression
results for each of the six establishments demonstrate that as the total bill amounts get even larger,
the slope of the regression line becomes less negative, approaching zero.
Greenberg, A. E. (2014). On the complementarity of prosocial norms: The case of restaurant tipping
during the holidays. Journal of Economic Behavior & Organization, 97, 103–112.
DESIGN OVERVIEW: Data was pulled from all credit card transactions from a restaurant chain in
upstate New York over the course of one year. All transactions required both a correct bill and tip
amount, so that situations when no tip was left on the credit card were dropped from the analysis
(because those situations likely included a cash tip since it was reported that instances of complete
“stiffing” among credit card customers were quite rare).
For their analysis, the “holiday period” was determined to be the weeks prior and post-Christmas
Day. Furthermore, other holiday days were added into the regression equation as a separate
variable. Customers were restricted in the analysis to those who were observed as having dined at
least once during the holidays and during the non-holiday period.
GEOGRAPHY: Forthcoming paper looking at whether prosocial behaviors (tipping behavior in general
and generosity during the holidays) compete with one another, leading to no change in tipping
behavior during the holidays, or whether they would complement one another such that people
would tip at higher rates during the holidays. Overall findings were that people tipped higher during
the holidays, but when the population was split, it was determined that this finding was skewed and
that while bad tippers tipped better, “good” tippers tipped even more.
Findings were that tips during the holiday period were 3.7% higher than in the non-holiday period
(24.3% overall).
Harrison, G. W., & List, J. A. (2004). Field experiments. Journal of Economic Literature, 42(4), 1009–
1055.
FINAL
Page 42
METHODOLOGY: Article discusses the use of field experiments in economic research. In contrast to
traditional means of collecting data for the purpose of economic research—such as the use of
naturally occurring data, where treatment and control status are not assigned at random, or
laboratory experiments, where treatment status is randomly assigned but the setting is artificial—
field experiments feature the use of randomly assigned treatment status but in a natural setting.
They thus potentially allow the researcher to make causal inferences while simultaneously mitigating
issues of external validity that are prevalent in laboratory experiments. The article briefly discusses
findings from three types of field experiments that allow for varying degrees of external validity:
artifactual field experiments, where the subjects are aware of the experiment and the activity that
they undertake does not directly correspond to naturally occurring activities, but where the subject
pool represents a naturally occurring population of interest; frame field experiments, where, like
artifactual experiments, the subjects are aware that they are participating in an experiment but
where the subject’s activity in the experiment more closely corresponds to naturally occurring
phenomena; and natural field experiments, where the activity induced by the experiment is
something the subjects would do naturally and they are simultaneously unaware that they are
participating in an experiment, maximizing the chances that observed responses to the treatment
would hold outside of the context of the experiment.
Hill, D. J., & King, M. F. (1993). An exploratory investigation into consumer knowledge of tipping
etiquette: Accuracy, antecedents and consequences. In W. Darden & R. Lusch (Eds.), Proceedings of
the symposium on patronage behavior and retail strategy: Cutting edge III (pp. 121–135).
DESIGN OVERVIEW: Sample was roughly 150 business majors ages 20 to 42 used for the analysis.
They were asked to provide responses to what the appropriate tipping levels were for various
services (not listed by the author, though the articles that they based these “correct” answers on
were listed). They created a battery of tipping-related items and used a factor analysis to determine
that there were five factors concerning tipping knowledge. Respondents were also asked a series of
27 developed questions on variables that were determined to influence tipping behavior from
literature reviews and one-on-one interviews on this subject. The 27 questions were determined to
have five useful factors: (1) social tipping orientation (their belief in the “social value of tipping”), (2)
tipping experience, (3) tipping confidence (their belief that their knowledge of tipping behavior was
correct), (4) tipping response (belief that poor service should receive poor tips, etc.), and (5) parental
influence.
TIPPING KNOWLEDGE: Ultimately, most of the factors were not found to be correlated to correct
tipping knowledge. The only two that were related were the parental influence (such that those who
FINAL
Page 43
learned more from their parents had more correct knowledge) and the age they first tipped (which
also makes sense given that the earlier they started tipping, the more guidance they likely got from
their parents and practice they have with tipping behavior).
Jargon, J. (2013, September 4). IRS rule leads restaurants to rethink automatic tips. The Wall Street
Journal. Retrieved from
http://online.wsj.com/news/articles/SB10001424127887323893004579055224175110910.
SERVICE CHARGE: Article reporting on the change in how the IRS counts tips automatically added to
the bill for large parties and the change that will occur starting in 2014. Under the new rules,
restaurants will have to take those automatic tips and add it to the servers’ actual wage at the end of
the pay cycle and withhold taxes from it. This means that servers will have to wait for that money, as
opposed to getting it at the end of the night, to ensure taxes are filed properly (which could mean
less income for servers), and cause more paperwork and costs to restaurants to manage additional
records.
This article was later cited by other websites, including NPR and the Consumerist (see Neuman,
2013; and Morran, 2013, citations).
Kerr, P. M., Domazlicky, B. R., Kerr, A. P., & Knittel, J. R. (2006). An objective measure of service and
its effect on tipping. The Journal of Economics, 32(2), 61–69.
DESIGN OVERVIEW: Author investigated how service quality, measured by the amount of time it took
to deliver the meal, influenced the tip size. Other variables included in the analysis were gender, race
(White vs. all others), and income of the served location. Some information was added to the
analysis based on census information, particularly the income variable. Two delivery drivers from the
same restaurant measured all data in this study aside from “income,” which was added based on
census information on the location of the delivered food. The type of payment and the magnitude of
the bill were also considered in the analysis.
However, it is worth noting that this article does not specify how many observations are being
analyzed, or provide any information about the drivers other than state that the “personal attributes
of the drivers were quite similar.”
INCOME: Higher-income areas were more likely to leave better tips than lower-income areas.
GENDER: Males were found to tip marginally better than females.
CASH VERSUS CREDIT: Cash-paying customers were actually found to tip better than credit card
customers, but this effect was nonsignificant when the magnitude of the bill was considered as part
of the regression equation.
FINAL
Page 44
Klee, E. (2004). How people pay: Evidence from grocery store data. Federal Reserve Board.
Retrieved from http://www.newyorkfed.org/research/conference/2006/Econ_Payments/Klee_b.pdf
DESIGN OVERVIEW: Examination of household data from the Survey of Consumer Finances from
1995, 1998, and 2001. Findings indicate that the share of credit card and debit card usage has
increased over the years, while the usage of checks has decreased. However, these market shares
and usage rates will not apply to many tipping situations, and should only be considered for
demographic groups that have credit or debit cards.
AGE: Credit card usage differed somewhat by age, such that very young heads of households and
those over the age of 75 have lower credit card usage than other age groups, while debit card usage
differed significantly. Debit card usage was highest among the youngest cohort and decreased as
age increased.
INCOME: For both credit and debit cards, usage rates increased along with rising income brackets,
indicating that more wealthy individuals are more likely to have credit and/or debit cards.
Kleven, H. J., Knudsen, M. B., Kreiner, C. T., Pedersen, S., & Saez, E. (2011). Unwilling or unable to
cheat? Evidence from a tax audit experiment in Denmark. Econometrica, 79(3), 651–692.
METHODOLOGY: Article reports results from a field experiment conducted on Danish tax filers where
tax filers were initially randomly assigned to one of two groups, where one group is subject to
rigorous audits while the other is not. Subjects are then randomly assigned to three groups, where
one group does not receive a notice of a future audit while the other two groups receive notices that
they will be audited with different probabilities (50% or 100%). Subjects in different treatment groups
are compared based on the difference in the amount of income that they report and baseline audit
data, with income broken down into that income that is subject to third-party reporting (i.e., there are
records kept by employers, etc., against which self-reported income can be checked) and income
that is purely self-reported. The authors hypothesize that only self-reported income should be
affected by past audits and the threats of future audits. Consistent with the hypothesis, the effect of
the enforcement treatments on evasion is close to zero for income subject to third-party reports, but
having been audited in the past and the prospect of future audits reduces evasion for self-reported
income. Evasion was generally substantially higher for self-reported income. Higher marginal tax
rates were found to increase evasion, though the effect was relatively small. The authors argue that
the results support the importance of enforcement through third-party reporting in explaining why
compliance is generally high in developing countries despite low audit probabilities and fines.
FINAL
Page 45
Koku, P. S. (2005). Is there a difference in tipping in restaurant versus non-restaurant service
encounters, and do ethnicity and gender matter? Journal of Services Marketing, 19(7), 445–452.
DESIGN OVERVIEW: Thirty-five participants were randomly selected for seven different service sector
businesses (245 total participants) that they indicated they had patronized within the past three
months. Service sector business included restaurants, barbershops/hair salons, spas, golf club
shops, auto detailing shops, auto mechanics’ shops, and valet parking. Participants were provided a
questionnaire that asked them if they tipped 15% or more of the total bill, less than 15%, or did not
tip at all. They were also given a space to provide a reason for their tipping decision.
INDUSTRY/SERVICE: For analysis purposes, the researchers combined all non-restaurant services to
compare against restaurant tipping. They also combined all those who said they tipped less than
15% and those who did not tip at all. Using a chi-square test, the researchers determined that there
is a difference between the reasons people tip in the restaurant industry and outside of it.
RACE/ETHNICITY: The researchers also compared White versus non-White respondents on tipping
tendencies outside the restaurant industry, and failed to find any difference.
GENDER: They only found a marginal difference between men and women in tipping outside the
restaurant industry.
Koku, P. S. (2007). Some significant factors that influence tipping in service encounters outside the
restaurant industry in the United States. Services Marketing Quarterly, 29(1), 23–45.
DESIGN OVERVIEW: The sample included 12 MBA students (6 male, 6 female) who indicated that
they had used another service-sector business in addition to the restaurant industry in the past 3
months. Other service-sector businesses included spas/body massage, barbershop/hair salons, auto
mechanics’ shops, plumbing services, auto detailing shops, valet parking, and lawn care services.
There were two sessions. All participants met in the first session for two hours and were asked about
service encounters in which they tipped in the past month and what led them to do so, as well as
service encounters in which they did not tip and why. The second session included 30-minute
individual sessions.
INDUSTRY/SERVICE: Using the framework of transaction cost analysis (TCA), the authors propose
several factors that influence a consumer’s tip in other service-sector businesses (i.e., service
industries other than restaurants). From information gleaned in interviews, the authors propose that
the customer’s decision to tip is influenced by (1) quality of service, (2) the length of time to be
served or have his or her issue resolved in an emergency situation, (3) the likelihood of repeat
purchase (which is influenced by service quality), and (4) budgetary constraints.
FINAL
Page 46
Lynn, M. (1988). The effects of alcohol consumption on restaurant tipping. Personality and Social
Psychology Bulletin, 14(1), 87–91.
DESIGN OVERVIEW: The author became employed as a waiter at the restaurant where the study took
place. For just over a month, he recorded information for 207 dining parties, including bill size, tip
amount, whether alcohol was consumed and number of drinks, customer’s gender, and payment
method.
BILL SIZE: A regression of tip amount on bill size indicated that tipping is strongly, positively related
to bill size. The resulting equation found a y-intercept of .32 (32 cents) with an additional tip of 11%
of bill size; this accounted for 50% of the variance in tip amount.
INDUSTRY/SERVICE: After controlling for the relationship between bill size and tip amount and a host
of other variables, a hierarchical multiple regression found a significant effect for alcohol. The results
indicate that alcohol (but not number of drinks) consumption increases tipping.
Lynn, M. (2004). Black-White differences in tipping of various service providers. Journal of Applied
Social Psychology, 34(11), 2261–2271.
DESIGN OVERVIEW: A randomized telephone-based survey was conducted to determine the
difference in tipping behavior among various service industries. This data was acquired by Lynn in
order to conduct follow-up analysis regarding tipping differences between Whites and Blacks.
Waiters, bartenders, barbers, taxi drivers, food-delivery people, hotel maids, masseuses, bellhops,
and ushers at theatres or sporting events were the occupations of interest. In the final analysis, 894
respondents (811 White and 83 Black respondents) were used. Respondents were asked, “If you
received good service from ____ would you tip them a percent of the total cost of the service, tip
them a flat amount, or not give them a tip?” Respondents were asked this question nine times for
different service industries: waiter or waitress; bartender; barber, hair stylist, or cosmetician; cab or
limousine driver; food-delivery person; hotel maid; skycap or bellhop; masseuse; and usher at
theater, sporting events, etc. Respondents were then further questioned about the amount they
would tip if they indicated that they would tip a percentage or flat amount.
INDUSTRY/SERVICE: Waiters received the most tips among Whites, though barbers also had a high
tip percentage amount among both Whites and Blacks.
RACE/ETHNICITY: Blacks are less likely to base restaurant tips on bill size than are Whites. Black
percentage tippers leave a smaller average percentage of the bill than do White percentage tippers
across many service contexts. Finally, Black flat tippers leave larger average dollar tips than do White
flat tippers across many service contexts (e.g., bartenders, barbers, hotel maids, and masseuses).
FINAL
Page 47
Lynn, M. (2004). Ethnic differences in tipping: A matter of familiarity with tipping norms. Cornell
Hospitality Quarterly, 45(1), 12–22.
DESIGN OVERVIEW: The survey results were the same as reported in Lynn’s 2004 article on tipping
knowledge among various racial groups in which respondents were collected by random-digit-dialing
(RDD) telephone methods. Respondents were asked how much it was customary to tip waiters and
waitresses in the United States with “15% to 20%” considered to be the right answer. Roughly 1,000
total completes were gained, but only 99 were from Black respondents. It is also important to note
that respondents were asked about customary practices rather than their own tipping behavior.
RACE/ETHNICITY: Overall, most Whites (over 70%) indicated that they knew the correct amount to tip
a waiter or waitress, compared with only 37.4% of Black respondents. Furthermore, 12.1% of Black
respondents reported that they did not know the correct amount compared with only 2.4% of White
respondents.
Lynn, M. (2006). Geodemographic differences in knowledge about the restaurant tipping norm.
Journal of Applied Social Psychology, 36(3), 740–750.
DESIGN OVERVIEW: A phone survey was conducted by Taylor Nelson Sofres using random-digit-dial
sampling for a total sample of slightly over 1,000 respondents. The primary question of interest was
how much people are expected to tip waiters and waitresses in comparison to how much they
typically tip. The “correct” response was considered to be 15% to 20%.
RACE/ETHNICITY: Significantly more Whites (72%) have the correct knowledge of tipping conventions
compared with Hispanics and Blacks (33% of both). These effects were still significant once other
variables were controlled for.
AGE: Age was initially significant, such that respondents in their 40s to 60s had higher levels of
knowledge compared with older and younger respondents, but once other factors such as race, sex,
education, income, metro status, and region were controlled for, it became insignificant.
GEOGRAPHY: Metro status was marginally significant before controlling for other variables but nonsignificant after control variables were considered. That said, the Northeast region had higher levels
of tipping knowledge compared with the South region, but there were no other significant differences
between other regions.
INCOME: Higher levels of income were related to higher levels of knowledge of correct tipping norms
even when controlling for other variables.
EDUCATION: Higher levels of education were related to higher levels of knowledge of correct tipping
norms even when controlling for other variables.
FINAL
Page 48
GENDER: Knowledge did not vary by sex when directly compared with men, but when other variable
were controlled for, it was found that women had a significantly higher level of tipping knowledge
than men.
Lynn, M. (2006). Tipping in restaurants and around the globe: An interdisciplinary review. In M.
Altman (Ed.), Handbook of Contemporary Behavioral Economics: Foundations and Developments,
(pp. 626–643). M. E. Sharpe Publishers.
DESIGN OVERVIEW: Lynn examines results from the literature for anything related to tipping. This
includes the determinants of restaurant tipping, including bill size, payment method, gender, and
race/ethnicity. A meta-analytic review conducted by Lynn and McCall (2000) found that 69% of the
variability of dollar tip amounts within a restaurant can be explained by the bill size. Several studies’
results support a “magnitude effect” where dollar tip amount increases with bill size, but percentage
tip decreases. Several studies have demonstrated that patrons paying by credit card tend to leave a
larger tip than those paying with cash (Feinberg, 1986; Garrity & Degelman, 1990; Lynn & Latane,
1984). Furthermore, the presence of a credit card company insignia induces higher tip amounts
(McCall & Belmont, 1996). There has been some support for men leaving larger tips than women
and waitresses receiving larger tips than waiters. Results indicate that patrons are more likely to
provide a higher tip for a server of the opposite sex. Black restaurant patrons are more likely than
their White counterparts to tip a flat amount rather than a percentage and tip a lower percentage.
Studies have shown these results even when controlling for education, income, and perceptions of
service quality (Lynn & Thomas-Haysbert, 2003).
Lynn, M. (2011). Race differences in tipping: Testing the role of norm familiarity. Cornell Hospitality
Quarterly, 51(1), 73-80.
DESIGN OVERVIEW: This study was a web-based survey from a consumer panel (Zoomerang.com) in
which the aim was to test and determine if tipping knowledge mediates the relationship between
race/ethnicity and tipping behavior, because no work up to this point had tested if this relationship
existed. Multiple waves of invitations were sent until the desired demographics groups were
gathered (100 respondents from both White and Black races, and with a separate split of those with
and without a college education, 831 total observations in all).
As with previous studies, respondents were asked how much people in the United States are
generally expected to tip waiters and waitresses, with 15% to 20% being considered the correct
answer. Later in the survey, they were also asked about their tipping behavior for waiters and
FINAL
Page 49
waitresses that gave them good service in order to determine not just their knowledge about tipping
behavior, but also their own behavior as well.
For other industries and services, such as hotel maids and bartenders, the respondent was simply
asked if that industry was generally tipped at all. Respondents who indicated that the various other
services were tipped were considered as having some knowledge of the norm for that occupation.
As with waiters and waitresses, respondents were further asked how often they tipped members of
the other professions.
RACE/ETHNICITY: Analyses indicate that tipping norm awareness did predict racial differences
between Black and White tipping behavior for restaurant tips, both for tip type (whether a percentage
of the bill was left versus a flat amount) and the percentage left.
No racial differences were found in the tipping/stiffing of hotel maids and luggage handlers, but
racial differences were found for the other investigated services.
Finally, a moderated relationship for norm awareness was also tested for, but this was not found to
be statistically significant.
INDUSTRY/SERVICE: A few significant differences were found for other professions. Specifically, they
found that norm awareness mediated racial differences in stiffing behaviors for haircutters and pizza
delivery, but not for bartenders, parking valets, or cab drivers.
Lynn, M. (2012). The contribution of norm familiarity to race differences in tipping: A replication and
extension. Journal of Hospitality & Tourism Research. Advance online publication.
doi:10.1177/1096348012451463
DESIGN OVERVIEW: Web-based survey was sent out to members of a consumer survey panel.
Response rates were not calculated as probability of panel selection is not captured. The final
sample included 180 respondents after s cleaning the original data set for outlier responses (such
as suggesting they gave tips over 100% of their bill) or improbable completion times...
Respondents were asked how much they would tip for one of two randomly assigned bill amounts,
$21.32 or $46.23, if the service was determined to be unusually good, average, or unusually bad.
Finally, respondents were also asked how much people in the United States are expected to tip a
waiter for adequate to good service and given typical response options.
RACE/ETHNICITY: Controlling for age, sex, income, education, and bill size, Black and Hispanics were
found to tip less, and were also less aware of the standard 15% to 20% tipping norm. Furthermore, it
was found that tipping knowledge significantly affected tip size after controlling for race, indicating a
partially mediated relationship.
FINAL
Page 50
Lynn, M. (2013). A comparison of Asians’, Hispanics’, and Whites’ restaurant tipping. Journal of
Applied Social Psychology, 43(4), 834–839.
DESIGN OVERVIEW: An online survey was conducted via a large multistate restaurant, yielding 1,274
final observations after 64 subjects who refused the race/ethnicity question were dropped from the
analysis. The survey asked respondents about service and restaurant quality in addition to the size
of their bill and tip size. Service quality was used as a control when observing the differences
between the different racial groups.
This study asked respondent race/ethnicity as a single-item question, as opposed to how the U.S.
Census Bureau asks two questions, one for race and one for ethnicity. In this setup, respondents
could indicate that they were Hispanic or Black, but not both.
BILL SIZE: Flat-dollar tips increased along with bill size while percent tips decreased in the same
span.
RACE/ETHNICITY: Hispanics tip significantly less than Whites but there are no differences between
Asians and Whites. However, given the relatively low N of the Asian population (roughly 75
observations) the findings have to be taken with caution.
Lynn, M., & Gregor, R. (2001). Tipping and service: The case of hotel bellmen. International Journal
of Hospitality Management, 20, 299–303.
DESIGN OVERVIEW: A hotel bellman interacted with 50 different customers while delivering one of
two conditions of level of service, at a small luxury hotel. In the “limited” service condition, the
bellman met customers at their cars with a cart and loaded their bags and then accompanied them
to their hotel room after they checked in, opened the door, and brought the luggage to their room.
They then asked guests if there was anything else they needed before collecting any tips and leaving
the room. The “full” service condition included the same treatment as the “limited” condition, but the
bellman also demonstrated how to use the television and thermostat, opened the blinds, and offered
to get ice for the guest. The bellman recorded the guests’ experimental condition, sex, apparent age,
and tip following each interaction.
INDUSTRY/SERVICE: The hotel bellman received significantly higher tips for providing the “full”
service condition ($4.77) than the “limited” service condition ($2.40). The effect of increases in tips
based on service condition was similar among men, women, younger guests, and older guests.
Lynn, M., & Latane, B. (1984). The psychology of restaurant tipping. Journal of Applied Social
Psychology, 14(6), 549–561.
FINAL
Page 51
DESIGN OVERVIEW: In the first study, 169 groups of customers were interviewed as they exited an
IHOP. Only those who paid the bill were questioned, or if two or more paid the bill, their responses
were combined. Participants were questioned about party size, restaurant atmosphere, food quality,
service quality, bill size, tip size, and improvements for the restaurant; respondent gender was also
recorded. All servers were female.
In the second study, 4 waiters and 5 waitresses collected data for 206 dining groups over a 1-week
period. They recorded the number of people on the check, number of people at the table, number of
checks at the table, bill size, gender of person(s) paying the check(s), method of payment, amount
left as a tip, and server’s level of effort spent serving the table. They recorded this information for
parties of five people or less or larger parties without a reservation because of the automatic gratuity
applied to larger parties.
BILL SIZE: In the first study, the average bill size was $3.16 and the average tip per person was $.42.
Customers tipped an average of 15.6% of their bill size. A hierarchical, multiple linear regression of
customer’s gender, party size, number of separate checks, atmosphere, service, food ratings, and
per-person bill size on percent tipped was performed. After controlling for other variables, per-person
bill size predicted a significant amount of variance in percent tipped. The larger the per-person bill
size, the smaller the percentage tip of the total check. In the second study, the average bill size per
person was $13.01 and the average tip per person was $2.01. Customers tipped an average of
15.5% of their bill size. In the hierarchical, multiple linear regression, per-person bill size was
unrelated to percent tip, which the authors speculate is due to the high price of the restaurant where
the study was conducted compared with that of a café in the first study where some groups only
ordered coffee or a snack.
PAYMENT METHOD: In the second study, a hierarchical, multiple linear regression of customer’s
gender, server’s gender, party size, number of separate checks, effort ratings, per-person bill size,
and payment method on percent tipped was performed. After controlling for other variables, payment
method predicted a significant amount of variance in percent tipped. Customers paying their checks
with credit cards tipped a larger percentage of the bill than cash-paying customers (16.9% versus
14.5%).
GENDER: In the first study, using the same hierarchical, multiple linear regression of customer’s
gender, party size, number of separate checks, atmosphere, service, food ratings, and per-person bill
size on percent tipped, after controlling for other variables, gender predicted a significant amount of
variance in percent tipped. Men tipped significantly more than women (17.4% versus 9.5%). For the
second study, in the hierarchical, multiple linear regression of customer’s gender, server’s gender,
party size, number of separate checks, effort ratings, per-person bill size, and payment method on
FINAL
Page 52
percent tipped, after controlling for other variables, customer’s gender also predicted a significant
amount of variance in percent tipped. Men tipped slightly more than women (15.7% versus 14.6%).
Lynn, M. & McCall, M. (2000). Gratitude and gratuity: A meta-analysis of research on the servicetipping relationship. The Journal of Socio-Economics, 29(2), 203–214.
DESIGN OVERVIEW: Meta-analysis conducted on a combination of published and unpublished
studies that had variables concerning tipping behavior and service quality, yielding observations for
2,547 dining parties across 20 different restaurants. The unit of analysis used was the N of
restaurants, as the authors argue that as tipping expectations and norms can vary by establishment,
that is the most appropriate level for analysis. Some splits were done to determine the relationship
between service quality and tipping behavior based on the metric used in the analysis and the
person providing the data, as some of the relationships were based upon a server’s estimation of the
service quality rather than the customer’s.
Ultimately, it was determined that there was a significant relationship between service quality and
tips, but that it accounted for less than 2% of variance in tipping behavior. This value was stronger
(almost 5%) among studies that had stronger measures of service quality. However, there was no
such relationship found for measures that recorded the perceptions of servers, indicating that
servers do not see a link between service quality and tipping behavior.
Lynn, M., & McCall, M. (2000). Beyond gratitude and gratuity: A meta-analytic review of the
predictors of restaurant tipping. Working paper, School of Hotel Administration, Cornell University.
DESIGN OVERVIEW: The authors limited the meta-analysis to research concerned with the restaurant
industry where the data were collected about an individual service encounter from one of three or
more modes: (1) restaurant checks, charge receipts, and comment cards; (2) records kept each
evening by restaurant servers; and/or (3) interviews with patrons as they departed restaurants. A
total of 22 published studies and 14 unpublished studies were included in the meta-analysis. The
authors meta-analyzed the relationships of tip size to bill size and of bill-adjusted tip size to 23
predictors from the tipping literature, including weather, payment method, and alcohol consumption.
BILL SIZE: The meta-analysis indicated that tip amounts were positively related to bill size. In fact,
the authors found that bill size accounted for about two-thirds of the variability in tip amounts.
GEOGRAPHY: Meta-analysis results indicate that patrons left larger bill-adjusted tips when the
weather was sunny.
FINAL
Page 53
CASH VERSUS CREDIT: Patrons left larger bill-adjusted tips when they used a credit card as their
method of payment or when they received their bill on a tip tray embossed with a credit card
company’s insignia.
INDUSTRY/SERVICE: Alcohol consumption was not related to bill-adjusted tips.
Lynn, M., & Thomas-Haysbert, C. D. (2003). Ethnic differences in tipping: Evidence, explanations,
and implications. Journal of Applied Social Psychology, 33(8), 1747–1772.
DESIGN OVERVIEW: A pair of studies were conducted to investigate racial differences in tipping. The
first study was based on the data from the 1997 Speer article. The first study was based on the data
from the 1997 Speer article. The second study was based on a collection of data sets based on five
tipping articles that either interviewed customers after they had left their restaurant or the servers
after the customers had had their meal.
The first study used the data from Speer (1997), with an N of about 1,000 from a telephone survey
and about 100 Black respondents. The combination of data sets in the second study resulted in an
N of about 1,800 respondents, with 94 Black respondents, 149 Asian respondents, and 113
Hispanic respondents. All the restaurants in the five studies used in the second study came from in
or near Houston, Texas.
RACE/ETHNICITY: The first study showed the same results as in previous studies in that Blacks
tipped less than Whites, but additional mediating analyses were conducted. Age, income, education,
and tip size were all found to be partial mediators of the race/ethnicity relationship.
The second study found that Whites left significantly higher tip sizes compared with both Blacks and
Asians, but not Hispanics. Another finding of note was that Asians and Hispanics were more likely to
tie the percent tip to service quality than Whites and Blacks.
Lynn, M., & Williams, J. (2012). Black-White differences in beliefs about the U.S. restaurant tipping
norm: Moderated by socio-economic status? International Journal of Hospitality Management, 31(3),
1033–1035.
DESIGN OVERVIEW: A pair of phone surveys were used for the analysis that used separate, but very
similar, questions. One survey asked, “Thinking about tipping overall, not your own practices, how
much is it customary for people in U.S. to tip waiters and waitresses?” The other survey asked,
“Thinking about restaurant tipping norms, how much are people in the U.S. expected to tip waiters
and waitresses?” Both questions were open-ended and results were coded into predetermined
response options, such as “15%–20%.” Tipping knowledge was considered to be either partial (in
FINAL
Page 54
terms of knowing that it was customary to tip waiters and waitresses) or complete (that it was
customary and that 15% to 20% was the correct amount).
A measure of Socio-Economic Status (SES) was crafted based on a pair of questions, one asking
income and the other asking education background. These were standardized and then averaged
together to form one scale.
RACE/ETHNICITY: The significant difference between White and Black tipping knowledge was
mediated by SES for partial tipping knowledge but not for complete tipping knowledge. This would
seem to indicate that all low SES individuals in general are unaware that tipping is customary in
certain situations, but that the “correct” tipping amount is not influenced by SES, and still seems to
involve a racial component.
Lynn, M., Zinkhan, G., & Harris, J. (1993). Consumer tipping: A cross-country study. Journal of
Consumer Research, 20, 478–488.
DESIGN OVERVIEW: The authors used information about tipping in 33 service professions across 30
different countries (Star, 1988). Each service in each country was coded as either “tipped” or “not
tipped” and aggregated. The authors obtained data for four different work-related indices that they
posit are related to tipping differences across countries. The study used 116,000 questionnaire
responses from industrial corporation employees in 50 different countries (Hofstede, 1983). The
four indices are “power distance,” where a high score reflects an acceptance for hierarchical
structure and a low score reflects the opposite; “uncertainty avoidance,” where a high score reflects
a culture that is concerned with following the rules and a low score reflects one that is willing to take
risks; “individualism,” where a high score is associated with a culture that is concerned with
individuals’ independence and a low score reflects a culture of collectivism; and “masculinity”
reflects a culture whose values are primarily masculine.
GEOGRAPHY: There was a correlation of .46 between the power distance index and number of
services that get tipped, indicating a strong relationship between high power distance scores and the
number of services that are tipped. There was a correlation of .55 for uncertainty avoidance and
tipping, indicating that tipping occurred more often in countries that were less tolerant of uncertainty.
There was a correlation of -.39 between the individualism index score and tipping, indicating that
tipping was more common in collectivistic countries. There was a correlation of .47 for masculinity
index and tipping, indicating that tipping occurred more often in countries with masculine values.
Japan was an outlier in all four analyses and was omitted.
FINAL
Page 55
McCall, M., & Belmont, H. J. (1996). Credit card insignia and restaurant tipping: Evidence for an
associative link. Journal of Applied Psychology, 81(5), 609–613.
DESIGN OVERVIEW: For the first experiment, data were collected from 77 paying customers at a
family restaurant; men were most frequently the paying customer (59 men and 18 women). Patrons
tended to be people vacationing at a nearby ski resort. The independent variable was what type of
tip tray the diner received with the check, either a blank tip tray or a tip tray with the credit card
insignia of a major credit card company in the center of the tray. Servers recorded the amount of the
bill, number of patrons in the dining party, the sex of the individual paying the bill, the method of
payment, and the total amount tipped.
For the second experiment, data were collected from 27 paying customers from a café in a separate
town from Experiment 1, whose main clientele is university students. The sample included 13 men
and 12 women, and two missing cases where gender was not recorded. The methodology of
Experiment 2 replicated Experiment 1 except that the credit card insignia on the tip trays was from a
different credit card company.
CASH VERSUS CREDIT: In the first experiment, an analysis of covariance (ANCOVA) revealed that a
credit cue significantly affected percentage tipped. Specifically, individuals that were given the tip
tray with a credit card insignia tipped a significantly higher percentage (19.77%) than those who
received a blank tip tray (15.48%).
In the second experiment, all paying customers used cash. Data were analyzed the same way for
Experiment 2 as they were in Experiment 1. Similar to Experiment 1, the ANCOVA demonstrated a
significant effect of credit cue on percentage of the bill tipped, where the presence of a credit card
insignia resulted in a tip percentage of 21.91% compared with those who received a blank tip tray
(17.53%). While the following two experiments did not compare tipping by method of payment used,
these were the basis for a lot of method-of-payment research in the future.
McCrohan, K. F., & Pearl, R. B. (1983, August). Tipping practices of American households: Consumer
based estimates for 1979. 1983 Program and Abstracts: Joint Statistical Meetings, Toronto, CA.
DESIGN OVERVIEW: Diary population was recruited via telephone recruitment and auto-registration
listings, creating an estimate of $5.7 billion in tipping revenue. Demographic targets were based on
census data. Two samples were used: 10,000 family households and an additional 1,500 nonfamily
households. The sample populations were recruited via telephone recruitment and auto-registration
listings. Reports were given on a quarterly basis over the course of the entire year. Families reported
over a two-week span every quarter and were staggered such that there were diaries coming in from
FINAL
Page 56
some of the sample every week. However, the nonfamily sample only reported during one quarter in
the entire year.
NATIONAL AVERAGE TIPPING BEHAVIOR: Of the $72.7 billion that was spent on dining out in 1979,
31% was considered to be spent on tipping occasions and such occasions accounted for over half of
all revenue. Of this revenue, tipping behavior constituted $5.7 billion, or roughly 14.4% of tipping
occasion behaviors.
After examining the data and determining what types of establishments should be classified as
“tipping occasions,” they determined that the true stiffing rate was somewhere around 20%, though
that included some situations where people ordered hasty meals or snacks.
GEOGRAPHY: Findings indicate that tipping was higher in the northeast region of the country
compared with the middle parts of the country and that metro areas tipped at higher rates.
INCOME: Very small differences were found relating to income, such that the highest income group
tipped at about 1% greater rate than the lowest income group.
CASH VERSUS CREDIT: Credit transactions tipped at a somewhat higher rate than cash transactions
(1% difference), but at this point they were only used in less than 3% of all dining transactions.
McCrohan, K. F., & Pearl, R. B. (1991). An application of commercial panel data for public policy
research: Estimates of tip earnings. Journal of Economic and Social Measurement, 17, 217–231.
DESIGN OVERVIEW: Authors expand on the analysis of the consumer diary data discussed in Pearl
and McCrohan (1984). The diary panel of restaurant patrons now includes the years 1982, 1983,
and 1984. The authors find that tipping occurs in only 29% of eating occasions, but that tipping
occasions account for approximately half of all expenditures. A regression analysis was also
undertaken to examine the determinants of the tipping rate (tip amount over total expenditure, for
both tipping and non-tipping occasions) for a given occasion.
NATIONAL AVERAGE TIPPING RATES: Across all periods, tip rates averaged approximately 14.4% and
that the average was relatively invariant across the types of eating establishments (inside, outside,
or non-tipping), though stiffing behavior varied by type, with tipping type restaurants (family,
atmosphere, and coffee shop) accounting for 90% of all tips.
CASH VERSUS CREDIT: Findings from the regression analysis indicate that tipping rates are higher
when establishments accept credit cards.
INDUSTRY/SERVICE: Findings from the regression analysis indicate that tipping rates are higher
when establishments serve alcohol.
GEOGRAPHY: Findings from the regression analysis indicate that tipping rates are higher when
establishments are located in metropolitan areas.
FINAL
Page 57
Morran, C. (2013, September 5). Are these the final days of automatic 18% tips at restaurants?
Consumerist. Retrieved from http://consumerist.com/2013/09/05/are-these-the-final-days-ofautomatic-18-tips-at-restaurants/.
SERVICE CHARGE: Report on the change in how IRS considers the automatic 15% to 20% gratuity in
restaurants, citing the piece by Jargon (2013) in The Wall Street Journal. Darden Restaurants, parent
company of Olive Garden, Red Lobster, and LongHorn Steakhouse, has already reported that it was
going to drop the automatic gratuity policy because of this issue.
Neuman, S. (2013, September 5). IRS to count automatic gratuities as wages, not tips. NPR.
Retrieved from http://www.npr.org/blogs/thetwo-way/2013/09/05/219290573/irs-to-countautomatic-gratuities-as-wages-not-tips.
SERVICE CHARGE: Blog post on the IRS’s change in how automatic gratuities are counted. The blog
post covers an original Wall Street Journal article on this issue (see Jargon, 2013, for original report).
Noll, E., & Arnold, S. (2004). Racial differences in tipping: Evidence from the field, Cornell Hospitality
Quarterly, 45, 23–29.
DESIGN OVERVIEW: Two unpublished studies, both of which were reported by servers from a large
restaurant chain, were used. In the first study, approximately 100 servers were asked a variety of
questions regarding supposed “tip predictors” such as race, alcohol use, and gender. The second
study aimed to investigate whether servers were accurately reporting their tip sizes as that
misreporting could significantly damage the results that were found in the first experiment. Two
servers in the same restaurant chain (but in another state) agreed to note their tips over a two-week
period. Overall, tips were recorded from 151 sets of customers.
RACE/ETHNICITY: Nearly all of the servers in the first study reported that they were aware of the
differences in tipping by race. Three-quarters of the servers indicated that their Black customers
were less likely to provide a tip, and when a tip was provided, more likely to tip below 15% than
White customers. In the second study, the two reporting servers reported similar findings for
differences between White and Black customers (though it is worth noting that outliers of tips over
26% were removed for both White and Black customers prior to analysis).
GENDER: In the first study, it was also found that male customers tipped more than female
customers.
INDUSTRY/SERVICE: In the first study, it was reported that customers who consumed alcohol gave
significantly higher tips than those who did not consume alcohol.
FINAL
Page 58
CREDIT/CASH: In the first study, servers reported that credit card customers tipped significantly
more than customers who paid with cash. However, in the second study it was reported that
customers who paid with cash gave marginally higher tips than those who paid with credit cards.
Papp, T. G., & Burkhammer, A. L. (2001, March). An investigation of server posture and gender on
restaurant tipping. Paper presented at the 22nd Annual Industrial Organizational Psychology and
Organizational Behavior Graduate Student Conference, Pennsylvania State University.
DESIGN OVERVIEW: Servers were recruited and asked to record information for 10 different dinners,
alternating between squatting and standing in order to determine how this changed tipping behavior.
Servers recorded bill size, tip size, and gender of the diners. Servers were recruited from campus
and by sending out survey packets to various restaurants in the area, yielding a final sample of 107
observations across 12 different servers. Eight of the final servers were female and four were male.
Each server was instructed to record five meals squatting and five meals standing, and to only record
this for small dinners or two or fewer diners.
GENDER: The only effect found was a marginally significant difference between male and female
servers such that male servers had more tips, but this was the only effect that was found.
Parker, J. A., Souleles, N. S., & Carroll, C. D. (2012). The benefits of panel data in consumer
expenditure surveys. National Bureau of Economic Research.
METHODOLOGY: Article reviews the benefits of the panel nature of the Consume Expenditure (CE)
Survey. The authors argue that repeating questions for individual respondents increases response
accuracy by increasing familiarity and understanding of the survey. In addition, repeat interviews
allow for respondents and interviewers to check the consistency of the responses, thus further
mitigating measurement error. On the other hand, requiring repeat interviews increases the burden
on respondents and thus potentially increases sample attrition and thus selection bias, though the
authors argue that there is little evidence that those who drop out of the sample are different in a
way that would influence expenditure. Repeat measures also help reduce noise in individual
respondent expenditures that could result from irregular expenditures taking place in individual
interview periods or measurement error that results from using long recall periods. When modeling
expenditure, panel data allows researchers to control for individual-level unobserved fixed effects,
allowing the researcher to potentially make causal inferences concerning the effect of some timevarying factors on individual expenditure. Controlling for unobserved individual fixed effects may also
reduce variability in estimated effect sizes, increasing the precision of estimates. Panel data also
allows the researcher to assess the dynamics of expenditure for a given household.
FINAL
Page 59
Paul, P., & Gardyn, R. (2001). The tricky topic of tipping. American Demographics, 23(5), 10–11.
DESIGN OVERVIEW: The article used the same data source that was mentioned in the Lynn piece on
differences between Blacks and Whites among various service types (2004). Roughly 900 total
phone numbers were randomly called to get the survey population. The professions that were listed
in the article were waiters, bartenders, barbers, taxi drivers, food delivery workers, hotel maids,
skycaps or bellhops, masseuses, and ushers at theater or sporting events.
INDUSTRY/SERVICE: Waiters were tipped far more often on a percentage basis than all other listed
professions (74% were tipped a percentage compared with 22% who got a flat tip), and were also
tipped the highest amount when tipped by percentage (along with barbers, both at 17%). Of all other
professions, the percentage of respondents who said they were tipped a percentage was much lower
than that for waiters, ranging between 5% for ushers to 31% for taxi drivers and food delivery
workers.
NATIONAL AVERAGE TIPPING RATE: Waiters were also stiffed the least of all the professions, with
only 2% reporting stiffing behaviors. Of the other professions, only masseuses (25%), hotel maids
(26%), and ushers (70%) were stiffed at rates greater than 20%, while bellhops were stiffed the least
of the other professions at 10%.
GEOGRAPHY: Various regional differences were discussed, such as respondents from the Northeast
region gave higher tips to waitstaff and busboys (16% to 20%, respectively) compared with other
regions, but they tipped cab drivers less than other regions (21% only gave a dollar or less for cab
rides compared with 13% from the rest of the country).
Pearl, R. B. (1984). A survey approach to estimating the tipping practices of consumers. Special
report on regression analysis to the Internal Revenue Service under contract TIR-81-21, Survey
Research Laboratory, University of Illinois, Champaign, IL.
DESIGN OVERVIEW: Special analysis of the 1982 data using regression. Analyses were run using
both a weighted and unweighted approach in order to examine both the propensity to tip and the
tipping percentage on occasions where a tip was left. Regressions using scaled weights produced
somewhat better regressions and were used in the final analysis. These analyses produced R2 values
of .20 for tipping behavior, but only .13 for regressions related to the actual tipping rate. Propensity
to tip was mostly predicted by whether it was for full-scale restaurants or for snack places.
GEOGRAPHY: Metro areas tipped at higher levels than nonurban areas.
FINAL
Page 60
Pearl, R. B., & McCrohan, K. F. (1984). Estimates of tip income in eating places, 1982. Statistics of
Income Bulletin, 3(4), 49–53.
METHODOLOGY: Authors attempt to improve upon prior attempts at estimating tipping income for
restaurants through the analysis of a large (N = 10,000 households of two or more related persons +
2,800 households of one or two unrelated persons) diary panel of restaurant patrons for 1982.
Respondents kept a diary where they recorded information about all eating occasions over the
course of a two-week period in a given quarter. The large sample (weighted to be representative of
the U.S. population in the given years) allowed for the more precise estimates, while querying
customers rather than employees or managers of establishments on tipping behavior mitigated bias
that may have resulted from the incentive of employees to underreport tipping income or managers
to exaggerate tipping income in order to justify subminimum wages. The authors argue that the use
of a diary as opposed to a survey increases the accuracy of the information provided, because
details of dining occasions are recorded closer to the time of the meal. In addition, they maintain
that the use of a diary lowers the probability that respondents will exaggerate the size of the tip in
order to impress the interviewer.
NATIONAL AVERAGE TIPPING RATES: The results of their analysis of data from the diary imply that
tips comprised approximately 7.4% of all expenditures and 14.3% of all expenditures on meals
where tipping actually occurred.
INDUSTRY/SERVICE: Respondents were asked to categorize establishments in six types (family,
atmosphere/specialty, coffee shop, cafeteria, fast-food and drive-in, and take-out) where the first
three categories were classified by the authors as “tipping establishments.” Within the tipping
establishments, sit-down and specialty establishments received tips on 60% of occasions. Within
this group, tips made up 12.9% of all expenditures and 14.5% of all expenditures on occasions
where a tip was actually given.
Pearl, R. B., & Sudman, S. (1983). A survey approach to estimating the tipping practices of
consumers. Final report to the Internal Revenue Service under Contract TIR-81-21, Survey Research
Laboratory, University of Illinois.
DESIGN OVERVIEW: Methodology was very similar to the previous report on 1979 tipping behavior
that was conducted by NPD, with a sample of 10,000 families and an additional 2,800 households
containing one or two unmarried people. The study was updated to include tipping behavior not only
in restaurant situations, but also in other industries, including bars, hotels, barbershops, and taxi
services. In this case, each household maintained records of tipping behavior at eating places during
a one-week period each quarter, with half of the sample doing this in addition to a supplementary
FINAL
Page 61
two-week diary study over the course of two quarters that covered additional services that might get
tipped (over 50 other industries were identified as having been tipped, but four of them accounted
for 80% of such situations and were the primary focus in the report). They were also asked to provide
some brief information about the type of establishment that they ate at to determine whether it was
a situation that tipping was expected in order to determine a true stiffing rate.
In order to determine if there were sources of bias in the data, an additional phone survey was
conducted with 935 households during the summer months to validate the data that was being
obtained via the diary studies. The validation study reported somewhat lower tipping rates for each
service, but they were within sampling error and might be due to the change in methodology between
a recall-question telephone survey and a diary survey.
AVERAGE TIPPING BEHAVIOR: In restaurants, the tipping rate was 14.3% overall, though only onefifth of responses came within the 14%–16% band, one-fifth of responses exceeded 20%, and
another one-seventh of responses reported less than 10%. Tipping rates also decreased along with
increasing household size.
The true stiffing rate was determined to be similar to levels reported in 1979, in that roughly 21.2%
of tipping situations for restaurants were stiffed and about 10% of expenditures. As noted in the prior
study, it is impossible to determine which purchase included snacks and small items that might not
be considered to be tip-worthy. Stiffing rates were the lowest for credit card purchases.
GEOGRAPHY: The overall tipping rate was found to be somewhat higher in the Northeast region of
the country and in metro areas.
INCOME: As noted in the previous study, as income levels increased, the tipping rate also increased
somewhat with greater income, but not to a large degree.
CASH VERSUS CREDIT: Credit card users gave higher tips (14.9%) than cash users (14.3%).
INDUSTRY/SERVICE: The tipping estimates that were reported for other industries, notably bars,
differed substantially from independent reports. Bars and taxi services reported receiving tips of 19%
to 20% overall, while barbers received 11.6%. The average tip at hotels and motels could not be
accurately assessed based on percentages, and the average tip amount was $1.89, though this
amount was still higher than that reported for the other services overall.
Stiffing rates are very difficult to assess for these other noted industries because hotels might be
considered to be “stiffed” even if it was simply a one-night stop at a motel, as 70% of hotel instances
did not get a tip.
FINAL
Page 62
Pearl, R. B., & Vidmar, J. (1988). Tipping practices of American households in restaurants and other
eating places: 1985–86. Supplementary report to the Internal Revenue Service under Contract TIR
86-279, Survey Research Laboratory, University of Illinois, Champaign, IL.
DESIGN OVERVIEW: Report on tipping behavior from 1986, including some comparisons with
previous years. It was found that roughly $6.76 billion was spent on tipping in restaurants and other
eating establishments compared with $6.67 billion in 1985 and $5.85 billion in 1979. However, the
percentage of money spent at eating-style restaurants compared with all eating places dropped from
39% to 34% in 1986. As in previous reports, restaurants were separated into categories that were
determined to be “tipping style” restaurants, though even when split in this manner the “stiffing rate”
seemed higher than it should be at 30%. Given this, they were recategorized based on the main type
of food in order to create a group of “high tipping–type restaurants.” This category was found to have
tipping incidences of more than 80% on most occasions. They also note that the estimates of tipping
revenue that they produce are lower than those provided by the U.S. Census Bureau and higher than
those generated by the Bureau of Economic Analysis.
In addition to their standard analyses, a regression analysis was conducted specifically using the
variables and information that might be available to the IRS in order to create a framework for future
use and identification of tipping discrepancies. Scaled weights and a combination of scaled and
expenditure weights were used in the analysis. The run with the expenditure weights was done to try
to correct for some of the downward bias that occurs when bill size increases. The expenditurebased approach accounted for a higher R2 than the scaled approach only (16.8% versus 13.1%).
Predictions using the scaled weights alone also showed somewhat higher tipping rates than were
accurate.
AGE: Middle-aged and older populations had higher rates of tipping incidence compared with
younger groups.
GEOGRAPHY: Regional differences were found such that the Northeast area (which consisted of the
New England and Middle Atlantic Census divisions) tipped at higher rates. Nonmetropolitan areas
had one of the highest negative predictive values in the analysis. Metro areas had higher rates of
tipping incidence than nonurban areas and their respective census regions. Metro areas were also
significant in the regression analysis.
INCOME: As in previous studies, they found some differences in tipping behavior based on income
levels. In this particular report, they found that tipping incidence was higher with higher
socioeconomic statuses. The difference in tipping rate between the highest and lowest income group
was only about 1%, so the range in this type of tipping behavior was not too great.
FINAL
Page 63
EDUCATION: Education had a similar effect on tipping incidence as did income, but had a great range
of tipping rates. Tipping rates were also 1.5% higher among the highest education group compared
with the lower groups.
CASH VERSUS CREDIT: Credit cards had the largest coefficient in the regression analysis, showing
that credit card users had higher tip percentages than those who paid with cash.
INDUSTRY/SERVICE: Establishments that served alcohol were not found to be as important to the
regression analysis as had been found in previous reports.
Rind, B. (1996). Effects of beliefs about weather conditions on tipping. Journal of Applied Social
Psychology, 26(2), 137–147.
DESIGN OVERVIEW: In the first study, 266 adult hotel guests (181 males and 85 females) were put
into four conditions. A room-service server reported one of the four weather conditions (sunny, partly
sunny, cloudy, or rainy) when asked or volunteered the information (if the guests didn’t ask) while
delivering food or drinks. He always reported temperatures in the 50s. The windows of the hotel
rooms were soundproof and dark-tinted that gave the impression it was cloudy even under sunny
conditions.
In the second study, 205 adult hotel guests (115 males and 90 females) were randomly assigned to
four conditions. A room-service server reported one of the four weather conditions (cold and rainy,
cold and sunny, warm and rainy, warm and sunny) when asked or volunteered the information (if the
guests didn’t ask) while delivering food or drinks.
GEOGRAPHY: For the first study, a linear contrast analysis revealed a significant positive association
between believed weather conditions and tipping. Tipping percentages improved as the conditions
went from rainy (18%) to cloudy (24%) to partly sunny (26%) to sunny (29%).
For the second study, an ANOVA demonstrated that hotel guests in the sunny condition tipped
significantly higher percentages than those who were told it was rainy. However, there was no effect
for the temperature conditions.
Sanchez, A. (2002). The effect of alcohol consumption and patronage frequency on restaurant
tipping. Journal of Foodservice Business Research, 5(3), 19–36.
DESIGN OVERVIEW: A waitress at a steakhouse restaurant collected data for 164 tables during
dinnertime over a three-month period; however, only 138 tables (158 parties) were included in the
analysis. The waitress recorded several variables of interest, including group ethnicity, group size,
number of parties (number of checks), party size, customers’ and paying patron’s ages (ages
estimated), customers’ and paying patron’s gender, number of alcoholic beverages (for the party and
FINAL
Page 64
for paying patron), food bill per party, bill size per party, tip amount per party, and payment method.
Several other variables, including the number of children per party and patronage frequency, were
recorded and analyzed.
RACE/ETHNICITY: For analysis purposes, customers were either identified as Caucasian or nonCaucasian. Ethnicity did not have any significant effect on tipping behavior. Caucasians tipped
slightly less ($7.42) than non-Caucasians ($7.49).
AGE: Results indicated that estimated age of the paying patron by the server was a good predictor of
tips. Older, paying patrons tipped more than those paying patrons judged to be younger.
CASH VERSUS CREDIT: Paying patrons’ choice of payment method (i.e., cash, check, or credit) did
not have any relationship with the total tip amount. Those patrons who paid with cash or check
tipped slightly more ($7.49) than those using credit ($7.42).
INDUSTRY/SERVICE: Consumption of alcoholic beverages was found to significantly affect the tip
amount. Tips from paying patrons who had one alcoholic drink ($10.19) or more than one ($9.52)
tipped significantly more than those who did not drink an alcoholic beverage ($6.44).
Schwer, R. K., & Daneshvary, R. (2000). Tipping participation and expenditures in beauty salons.
Applied Economics, 32, 2023–2031.
DESIGN OVERVIEW: A stratified, convenience sample of 317 respondents was selected for this
survey. This sample included a mix of respondents from banks, university staff and students,
government employees, and customers of barbershops and beauty salons. Furthermore, the survey
was conducted over time periods during the spring and summer of 1995. Questions on the survey
dealt with patronage, what barbershops or beauty salons they go to, important qualities in the salon
or barbershop they go to, and various demographic and socioeconomic questions.
Analyses were conducted using a combination of probit and Tobit regressions. Two Tobit regressions
were used, a censored version as well as a truncated run. The truncated, two-step Tobit model
showed the better fit of the Tobit models.
SERVICE/INDUSTRY: Overall, while Post (1992) recommended tipping 15% to 20% for hair
salon/barbershops, it was determined that all customers tipped at 8% of their bill, and 9% when
customers who left no tip were excluded.
INCOME: Income was included in the analyses, but significant findings were only discovered in the
probit analysis. In both cases the dummy variables showed marginally significant findings.
AGE: Results from the truncated Tobit analysis indicate a marginally significant finding for tipping
behavior, such that tipping rates decrease with the age of the respondent, though no such significant
finding was discovered in any of the other data runs.
FINAL
Page 65
GENDER: Mixed findings regarding gender were found between the probit and truncated Tobit
models. The probit model showed a marginally higher tipping total from women than men, but an
opposite finding was reported in the truncated Tobit analysis.
RACE: White respondents were found to tip marginally more in only one of the three models (the
censored Tobit) and race was generally found to be a nonsignificant variable.
Seiter, J. S., & Weger, H., Jr. (2013). Does a customer by any other name tip the same? The effect of
forms of address and customers’ age on gratuities given to food servers in the United States. Journal
of Applied Social Psychology, 43, 1592–1598.
METHODOLOGY: A field experiment of diners (N = 142) at two Utah restaurants was conducted to
examine the effects of differences in how servers addressed customers (first name, Mr./Mrs., etc.)
on tip rate. A regression analysis was conducted that included form of address effects, customer
age, and the interaction between age and form of address. Data was collected by three
student/servers.
AGE: In the regression model without the interaction (i.e., just form of address and age), customer
age had a negative association with tip amount, but the estimated relationship was not found to be
statistically significant at the 5% level, but was at the 10% level (p = .09). This negative relationship
was stronger when the customers were addressed by their first name.
Simpson, H. (1997). Tips and excluded workers: The New Orleans test. Compensation and Working,
Bureau of Labor Statistics, 32–36.
DESIGN OVERVIEW: Data was gained from “BLS field economists” in face-to-face interviews for the
most part. Of the 359 establishments that were sampled, 77% provided some data, but only 11
provided tipping data, indicating that the findings in this article are to be considered as preliminary
without any significance testing. Besides information regarding the number of tipped workers at the
establishment and the dollar amount of tips collected, the BLS workers also gave a rating for their
confidence in the data that was provided. However, while the majority (82%) of the data for “hours
worked” was determined to be good, only 55% of the tip data was considered to be good, and 27%
was considered “poor.” This indicates that the data in the article might be flawed and underscores
the difficulty of obtaining reliable tipping data.
SERVICE/INDUSTRY: Of the occupations that met publication criteria (certain number of workers
from a certain number of establishments at least), waiters had the highest amount of average tips
per hour ($6.10), followed by hostesses ($5.73), bussers ($4.86), and bartenders ($3.70).
FINAL
Page 66
GEOGRAPHY: The article reports that tipped employees would underreport their tips during the busy
months and overreport during the slower months in order to balance things out for their bosses and
create less hassle. Similarly, the months when data was collected (July–August) were considered to
be slower tourist months, so the data might be skewed somewhat by that.
Speer, T. (1997). The give and take of tipping. American Demographics, 19(2), 51–54.
DESIGN OVERVIEW: Random telephone survey of roughly 1,000 adults in 1996. Respondents were
asked what the largest determining factor was regarding their tipping behavior, and service was
often claimed as the most important thing, though this percentage was smaller among nonrestaurant services.
INDUSTRY/SERVICE: Roughly 28% of respondents indicated that they never tipped the individual in
the hotel who replaces their towels and bed sheets. Also of note was that 36% of respondents
indicated that they always carried their baggage at hotels and airports, and were thus unable to
answer any questions about tipping this particular profession. Similarly, roughly half of adults
reported that they don’t use taxi cabs or limo drivers, so they were unable to answer any such
questions about tipping behavior. Finally, 40% of respondents indicated that they are never served
by bartenders.
Also worth noting is that this article has a chart that indicates the percentage of respondents who
indicate specific tipping percentages for a number of different industries.
INCOME: Higher-income ($50,000 or higher) individuals reported that the reason they tipped was
that they tipped to help some individuals (notably parking valets, luggage handlers, and taxi drivers).
Lower-income individuals were less likely to tip at all because they reported that the bill should
reflect the full cost of the service, though this behavior does not extend to waiters.
GEOGAPHRY: Southerners were more likely to say they would never tip for some services, mostly taxi
drivers, waitstaff, and barbers, while Midwesterners were the most likely to say that they would never
tip parking valets, bartenders, maids, and luggage handlers. Northerners tipped the highest of the
groups when split by region, or reported as much.
GENDER: In this study, women were reported as more likely to leave a tip than men, particularly
when it comes to services other than taxis or waitstaff. Women are more likely to report that they tip
based on the impact that it has on others when compared with men.
Star, N. (1988). The international guide to tipping: When, where, and how much to tip in the U.S. and
around the world. New York, NY: Berkley Books.
FINAL
Page 67
DESIGN OVERVIEW: Star’s book discusses cross-country differences in tipping. Specifically, the
author describes expectations and norms for tipping across 38 professions in 34 different countries.
The 38 professions cover a diverse set of service-related professions including restaurant jobs (e.g.,
severs, bartenders, hostesses, etc.), guides, hotel staff, and hair stylists. According to Lynn, Zinkhan,
and Harris (1993) that had correspondence with Star, her tipping suggestions and summaries were
primarily based on questionnaires sent to hotels, national railroads, resorts, restaurants, tour
groups, and so on in each of the 34 countries.
Thomas-Haysbert, C. D. (2002). The effects of race, education, and income on tipping behavior.
Journal of Foodservice Business Research, 5(2), 47–60.
DESIGN OVERVIEW: Phone surveys were conducted on a population of 1,005 respondents. The
phone survey was conducted by Market Facts for American Demographics and methodology of the
phone survey is discussed in greater detail in another article (Speer, 1997). Questions were asked
regarding whether respondents tipped various service-industry workers such as servers, bartenders,
taxi drivers, parking attendants, and luggage handlers, and why they tip or did not tip.
INDUSTRY/SERVICE: Luggage handlers were tipped the most (98% said they always tipped this
group, followed by servers, parking attendants, taxi drivers, and bartenders.
INCOME: Income was found to significantly affect tipping behavior and when used as a dichotomous
controlling variable it nullified the influence of race on tipping behavior.
EDUCATION: Same effect as income was found in that it is significantly related to tipping behavior
and when used as a control it nullifies the effect of race on tipping behavior.
RACE/ETHNICITY: White respondents tipped every category of worker significantly more often than
Black respondents, but this effect was nonsignificant once education and income levels were
considered for all service workers except for taxi drivers. However, Black respondents were more
likely to indicate that service quality was more important to them than White respondents and that
they tipped more to ensure better service in the future. Blacks were also more likely than Whites to
indicate that they did not tip because they felt that it should be included in the bill. Black
respondents reported that they tipped less than Whites but this effect was nullified when income
and education were incorporated into the model.
FINAL
Page 68
Appendix C – Search Engines and Search Terms
Table 5. Search Engines
FINAL
Search Engine
Description
University Library System
Online database of journal
articles maintained by local DC
Metro University.
Google Scholar
Google search engine that
produces links to both gated
and ungated scholarly articles.
JSTOR
Archive of peer-reviewed
articles published in academic
journals.
Social Science Research
Network (SSRN)
Archive of social science
working papers.
Business Source Complete
(EBSCOhost)
Database containing archived
peer-reviewed articles
published in business-related
journals.
ABI/INFORM Complete
(ProQuest)
Database containing peerreviewed articles published in
business-related economics,
business, accounting, and
marketing journals.
Accounting & Tax (ProQuest)
Database containing peerreviewed articles published in
high-impact accounting,
auditing, tax management, and
tax law journals, as well as
trade publications.
PsycINFO
Database of peer-reviewed
behavioral science and mental
health articles.
Page 69
Table 6. Search Terms
Search Term
Themes
Gratuity, tipping, tip giving,
stiffing behavior, tip reporting
GENERAL TIPPING, NATIONAL
AVERAGE TIPPING RATES
Internet, mail, and mixed-mode
surveys: the tailored design
method.
METHODOLOGY
Regional, urban versus rural,
metropolitan tipping differences,
holiday differences, seasonal
effects, tourist tipping
GEOGRAPHY
Income, education, age, gender,
SES, salary tipping
differences/restaurant tipping
differences
INCOME, EDUCATION, AGE,
GENDER
Black-White/Asian/Hispanic/
racial tipping differences
RACE/ETHNICITY
Tipping knowledge, tipping
norms
TIPPING KNOWLEDGE
Service charge law change,
mandatory service charge,
mandatory restaurant tips
SERVICE CHARGE
Tipping differences by industry,
non-restaurant tipping, tipping in
services industries, alcohol and
tipping
INDUSTRY/SERVICE
Method of payment tipping,
credit card/cash tipping, cash
differential
CASH VERSUS CREDIT
FINAL
Page 70
| File Type | application/pdf |
| File Modified | 2019-07-29 |
| File Created | 2014-02-05 |