Overview of the Survey of Occupational Injuries and Illnesses Sample Design and Estimation Methodology

Overview of the Survey of Occupational Injuries and Illnesses Sample
Design and Estimation Methodology October 2008
Philip N. Selby, Terry M. Burdette, Erin M. Huband
Bureau of Labor Statistics, 2 Massachusetts Ave NE Room 3160, Washington, DC 20212

Abstract
The Survey of Occupational Injuries and Illnesses (SOII) uses a stratified sample design to produce State and National
estimates for nonfatal workplace injuries and illnesses. Private industry estimates are produced separately for 42
States, the District of Columbia, and three U.S. territories (Guam, Puerto Rico and the Virgin Islands). The level of
industry detail for which State estimates are available varies widely and is based on the needs determined by each State.
Additionally, estimates for injuries and illnesses for State and local government workers are available for 26 of these
States. Future plans will expand the coverage to national estimates for the State and local governments beginning with
data released in 2009. This paper describes the frame development, stratification criteria, sample allocation, sample
selection, and estimation methodologies used to produce the number and frequency (incidence rates) of nonfatal
workplace injuries and illnesses.

Key Words: stratified sample, target estimation industry
1. Background of the Survey
The Bureau of Labor Statistics Survey of Occupational Injuries and Illnesses (SOII) gathers information on the number
and frequency of injuries and illnesses experienced by America’s workforce.
The current survey evolved from annual BLS surveys first conducted in the 1940’s. For the next three decades, those
nationwide surveys proved useful but with two major limitations: (1) represented employers recorded and reported
worker injuries on a voluntary basis; (2) injuries were limited to death, permanent impairment, or temporary disability.
Congress addressed these limitations and passed a landmark piece of legislation, The Occupational Safety and Health
Act of 1970, implementing regulations requiring most private industry employers to maintain records (logs) and prepare
reports on work-related injuries and illnesses. Participation in the survey was no longer voluntary.
Since then, the survey has undergone various changes to meet the record keeping regulations set by the Occupation
Safety and Health Administration (OSHA), U.S. Department of Labor. Details of these regulations, both old and new,
are available from the OSHA internet site (http://www.osha.gov/recordkeeping/index.html).
The SOII is a Federal/State cooperative program, in which the Federal government and participating States share the
costs of participating State data collection activities. State participation in the survey may vary by year. Sample sizes
are determined by the participating States based on budget constraints and independent samples are selected for each
State annually. Data are collected by BLS regional offices for non-participating States. For 2006 survey data released
in calendar year 2007, private sector injury and illness data were tabulated separately for 42 States and the District of
Columbia participating in the program. Estimates for State and local government were produced for 26 of these States.
There were eight states which did not participate in the program. For these States, a smaller sample is selected to
provide data for national estimates only. Additionally, estimates are tabulated for three U.S. territories- Guam, Puerto
Rico, and the Virgin Islands- but data from these territories are not included in the tabulation of national estimates.
Establishments responding to the SOII provide two sets of information – annual counts of injury and illness cases and
detailed case and demographic data for individual cases involving one or more days away from work. The latter data
include information on the worker injured, the nature of the disabling condition, and the event and source producing
that condition. BLS publishes selected national estimates for private industry in two news releases: (1) a summary of
counts and incidence rates in the private industry; (2) a more detailed release on the characteristics of injury and illness
cases that involved days away from work.

A third release from the BLS reports on workplace fatalities in the Census of Fatal Occupational Injuries. However,
given that this is a census, no discussion of its methodology will be included here. For more information on workplace
injuries and illnesses, please visit http://www.bls.gov/iif/.

2. Frame Development
Because the SOII is a Federal-State cooperative program and data must meet the needs of participating State agencies,
an independent sample is selected for each State. The survey covers the entire private sector except self-employed
persons, private households, and small farms. Establishments with as few as one employee are sampled in all
industries except agriculture production, where the minimum size is eleven employees. Mining and Railroad industries
are not covered as part of the sampling process. These data are furnished directly from the Mine Safety and Health
Administration and the Federal Railroad Administration, respectively, and used to produce State and national level
estimates. The United States Postal Service and federal government employees are out-of-scope for the survey.
Although State and local government agencies are not currently surveyed for national estimates, several States have
legislation which enables them to collect these data and publish State-level estimates. Beginning with the 2008 survey,
to be published in calendar year 2009, State and local government estimates will be published at the national level and
for most States.
The main source for the SOII sampling frame is the BLS Quarterly Census of Employment and Wages (QCEW). The
QCEW is a near quarterly census of employers collecting employment and wages by ownership, county, and six-digit
North American Industry Classification System (NAICS) code. States who survey State and local government units
have an option to either use the QCEW or supply public sector sampling frames.
Data are collected on injuries and illnesses for the entire calendar year. Samples are allocated and selected from the
latest available frame information from the QCEW. For example, the sample for survey year 2009 was selected in July
2008 using second quarter 2007 data from the QCEW. Once the sample is selected, BLS and State staff ensure address
information is correct. In December 2008, selected establishments are notified that they are in the 2009 sample and
asked to keep records.

3. Stratification Criteria
The SOII utilizes a stratified sample design with varying degrees of industry stratification levels within each State.
This is desirable because some industries are more prevalent in some States compared to others. Also, some industries
can be relatively small in employment but have high injury and illness rates which make them likely to be designated
for estimation. Thus, States determine which industries are most important in terms of publication and the extent of
industry stratification is set independently within each State. This was deemed a very important part of the SOII
design because the industry structure can vary widely from State to State. The term Target Estimation Industries (TEIs)
is used to describe the specific industries that States request for publication. Thus, a TEI is a NAICS industry, or group
of NAICS industries, for which a State plans to produce an estimate. A State may set TEIs at different levels than other
States. The TEIs are used to set the level of stratification within the States. A State’s ability to actually publish these
estimates depends on the reliability of the estimates and confidentiality requirements.
There are two types of TEIs: Publishable TEIs as mentioned above and Residual TEIs. Residual TEIs are groupings of
industries for which the State does not wish to publish estimates but will be grouped with other TEIs for aggregate
State estimates.
One distinct rule for setting TEIs is that each NAICS industry must be accounted for in one and only one TEI. As an
example, within NAICS 11 (Agriculture, Forestry, Fishing and Hunting), a State may want to publish estimates for
NAICS 111 (Crop Production), and NAICS 112 (Animal Production) but not for the remaining NAICS 113 (Forestry
and Logging), 114 (Fishing, Hunting and Trapping), and 115 (Support Activities for Agriculture and Forestry). In this
case a separate stratum would be formed to represent all of the units in NAICS 113, 114, and 115. The final product of
the TEI setting process is that specific industries are identified by each State they desire to publish separately.

3.1 Additional Rules for Setting TEIs
There are additional rules in place for setting TEIs in order to produce desirable industry level estimates at both the
National and State levels. National level TEIs (NTEIs) are set at various NAICS levels that target a desirable industry
level estimate. Two additional constraints are that States must select a minimum number of TEIs and cannot exceed a
maximum number of TEIs. The minimum constraint is required to support national estimates to which State data
contribute. In general, minimum TEIs are set at the two-digit industry level for all major industries. As an example, at
minimum, a State would need to set a TEI at the NAICS 11 industry level in order to support the national estimate at
the same level.
States cannot exceed a predetermined maximum number of TEIs for private industry. This maximum limit is
determined using the State’s sample size. Establishing a maximum number of TEIs helps to ensure that the sample will
be sufficient to provide reliable estimates for each TEI. The maximum threshold includes both publishable TEIs and
residual TEIs.

3.2 Importance of TEI selection
Targeting appropriate State industry levels for publication is a vital step in producing accurate and reliable estimates.
Setting State TEIs too broadly may not provide the necessary industry detail a data user is looking for or needs. Setting
TEIs at a more detailed level in one industry that doesn’t support publication will subtract from sample in other
industries where more detailed levels could be published.
Some considerations that States take into account when setting TEI levels are: (1) What industries are important to their
State? (2) Who are the main data users and what level of detail is required to meet their needs? (3) Is one industry more
prone to injuries and illnesses than other industries? States also look at what TEI levels have been published in
previous years. If detailed levels have not been published due to reliability or confidentiality issues in the past, setting
the TEI at a less detailed level may allow the estimate to be published. It also improves the efficiency of the sample
design. Decreasing the number of TEIs increases the number of sample units in the remaining TEIs. A higher
sampling rate within a TEI results in more reliable estimates. States are encouraged to review TEI settings each year to
improve reliability and maximize publication of their estimates.
TEI selection also impacts the survey sample allocation. TEIs are the cornerstone to the allocation process as the
sampling cells are based on the TEIs set by the States. A sampling cell is defined by ownership x TEI x employment
size class. Ownership is made up of three divisions: State Government, Local Government and private industry. Size
classes are based on an establishment’s average annual employment and defined as follows:
Size Class

Average Annual Employment

1

10 or less

2

11 – 49

3

50 – 249

4

250 – 999

5

1,000 or greater

4. Sample Allocation and Selection
The sample allocation and selection system was completely overhauled in 2003 along with the conversion from
Standard Industrial Classification (SIC) coding to NAICS coding based estimates. Three major goals were achieved
with the new system: (1) moving from the mainframe to Unix environment; (2) greater flexibility to handle sample
design changes where States may move from a non-participating State to a participating State and vice-versa; and (3)
simpler and more efficient allocation module.
The former system used a complex allocation method which took into account many variables including lost workday
cases (LWDC) data, target sampling errors, and frame sizes. A Neyman’s formula for a fixed variance was used to

calculate sample sizes for each sampling cell. Although this worked fairly well, it created inefficiently large samples in
low hazard industries.
The need to move from SIC to NAICS based estimates provided an opportunity to improve the allocation module.
While the new process was being developed, the old processing system was modified to handle NAICS-based strata for
the 2003 and 2004 samples. Initial NAICS-based SOII estimates were first available in October 2004 from the 2003
survey. Thus, the implementation of the allocation module was delayed until NAICS-based data was available for
application to the survey design. The new allocation module was introduced beginning with the 2006 survey which
was selected in October 2005.
Because historical NAICS data was unavailable, an optimal allocation procedure was proposed which attempts to
distribute the sample to the industries in a manner that minimizes the variance of the estimates. This method provides a
smaller sample size in cells where units have similar incident rates and a larger sample size in cells where units have
more variable rates.
Research was done to determine what measure of size was most appropriate for the allocation module. With the trend
of establishments in the SOII going to more restricted work activity for employees that are injured on the job, the
choices were narrowed down to the following: (1) Days Away from Work cases (DAFW); (2) Days of job transfer or
restriction (DJTR); and (3) Total Recordable Cases (TRC). Rates from the 2003 SOII were studied for all 1251 TEIs
for each of the above three categories. The average case rate, standard deviation, and coefficient of variation (CV) for
each set of rates were calculated. The CV is the standard deviation divided by the estimate and is commonly used to
compare estimates in relative terms. The CV for the TRC was lowest which led to the recommendation of using the
TRC rate by size class for the measure of size input for the allocation module. Because TRC includes both DAFW and
DJTR cases, it is the most prevalent estimate per establishment.
The important feature of the sample design is its use of stratified random sampling with a Neyman allocation. Because
industry (TEI) and employment (size class) groups is highly correlated with an establishment’s number and rate of
reported injuries and illnesses (TRC), stratified sampling provides greater precision and results in a smaller sample size
than simple random sampling would require. With Neyman allocation, the overall variance of the TRC rate is
minimized for a fixed total sample size.

4.1 Allocation Process
The optimum allocation method used is an iterative process. Certainty cells are removed from the calculation after each
iteration. Any cell allocated more sample than there are units in the cell are designated as certainty cells. Certainty
cells can also occur as a result of ensuring that an adequate number of units are sampled to produce accurate and
reliable estimates for the cell. The methodology also ensures that each sampling cell has at least two units selected
(where there are at least two units in the cell) and that the maximum weight of any sampled unit is less than 250. The
end product is a probability sample where, for a fixed overall sample size, the sample is distributed across the strata
such that the sampling variance of the TRC rate is minimized.
The allocation module is a multi-step process applied to each State’s sampling frame created as described above. The
highlights of the major steps in the process are as follows:
- The Measure of Size (MOS) for sampling cell h is derived using the following formula:
MOS

p 1

p

E

Where E h = employment in stratum h and ph = TRC rate / 100
- Certainty and minimum sample criteria are applied. The number of units in certainty cells and the minimum number
of units allocated to non-certainty cells are excluded from the allocation process.
- Sum the MOSh of the non-certainty cells. The first iteration of optimum allocation is performed by computing the
sample size for each remaining non-certainty cell:

n

MOS
∑
MOS

n

Where MOSh = measure of size for sampling cell h not assigned a sample size
k = number of sampling cells not assigned a sample size
n = total sample size for the State minus sample size already fixed
nh = sample size in sampling cell h rounded to the nearest integer
- Check for new certainties, that is, where the stratum sample size exceeds the number of frame units and recalculate
the sum of the MOS and n’.
- Re-allocate n’ to the remaining non-certainty sampling cells and check for new certainty cells. If certainty cells exist,
then complete another iteration of allocation. If no new certainty cells exist, then nh is the sample size for the stratum.
Finally, apply minimum sample sizes where there are at least two frame units in a cell. Also check to see if frame units
divided by the allocated sample size is less than 250. If not, divide the frame units by 250 and round up to the nearest
integer to determine the minimum sample size. These additions are made to the final sample. No further adjustments
to other sampling cells are made. Thus, the final sample may be slightly larger due to these adjustments.

4.2 Sample Selection
The sample selection is done using the SAS survey select procedure. The procedure uses two inputs files: (1) the final
State frame file and (2) the final allocation file. A systematic selection with equal probability is used to select a sample
from each sampling cell (stratum). As mentioned earlier, a sampling cell is defined as State/ownership/TEI/size class.
Prior to sample selection, units within a sampling cell are sorted by employment and then by LDB number (unique
identifier assigned to each reporting unit on the QCEW) to ensure a consistent representation of all employments in
each stratum. The output from the sample selection includes a sample weight assigned to each sample member.
Sampling weights are calculated by dividing the number of frame units in the sampling cell by the number of sample
units in that cell.
After sample selection is complete and prior to releasing the sample to the State, case subsampling codes are added to
help reduce respondent burden for establishments where a large number of cases requiring days away from work are
expected. As mentioned earlier, more detailed data are gathered for these cases to support the Case and Demographics
estimates. During estimation, case subsampling factors are calculated to reduce the number of cases submitted from
the expected number of cases recorded by the establishment.

4.3 Final Summary Weight Calculation
Nonfatal workplace injury and illness data collected for the SOII are used to tabulate estimates for two separate data
series: (1) summary, or industry-level, estimates and (2) more detailed case and demographic estimates for cases that
involve days away from work (DAWF). By means of a weighting procedure, sample units represent all units from the
universe or sampling frame. A final summary weight (FSW) is calculated for usable units to represent all units in their
size class. This final summary weight is a product of the original sample weight and four adjustment factors: (1) unit
nonresponse or Nonresponse Adjustment Factor (NRAF); (2) reaggregation factor (REAG); (3) Outlier Adjustment
Factor (OAF); and (4) benchmarking factor (BMF).
FSW

Original sample weight x NRAF x REAG x OAF x BMF

The unit nonresponse (NRAF) factor adjusts for the small proportion (less than 9%) of establishments that do not
respond to the survey. This factor is calculated by dividing the sum of the weighted employment of viable units in a
sampling cell by the sum of the weighted employment of usable units in that sampling cell. Viable units are units
deemed to be in-scope during collection (i.e., collected units and units from which no response was obtained).

NRAF

∑ weighted viable employment
∑ weighted usable employment

The reaggregation factor (REAG) is applied to adjust for those instances where a sample unit may be unable to report
data for the unit that was sampled. For example, company XYZ reports to the sampling frame as several single
locations. One single location is sampled and data are requested for only that location but the company maintains one
log for all locations and cannot separate the data for the sampled unit. The reaggregation factor is simply a ratio of the
assigned employment to the reported employment.
REAG

assigned employment
reported employment

The outlier adjustment factor (OAF) is applied to units where an unusually extreme case count or hours worked has an
undue influence on the estimates. For example, a unit in the health care industry reports an unusually high number of
illness cases. The documentation noted a severe scabies outbreak that led to the high number of reported cases.
Possible outliers are identified on a report generated by the estimation system after preliminary estimates are run. This
report is first reviewed to verify that the reported data are correct (i.e. no data entry errors). State offices review this
report and identify units they feel should be considered an outlier. The National Office uses several review tools to
determine if the outlier request will be granted. An outlier adjustment factor is calculated to make the unit self
representing, effectively changing its final weight to one before applying the benchmark factor. The formula for the
OAF is:
OAF

1
original sample weight x NRAF x REAG

The remaining units in a sampling cell that contains an outlier unit need to have an OAF calculated that accounts for
the outlier now being self representing. This factor essentially equally distributes the remaining weighted employment
of the outlier unit to the other usable units in the sampling cell.
The final factor to be calculated to produce the final summary weight is the benchmark factor (BMF). Benchmarking
adjusts the reported summary data for an industry to account for employment changes in the universe between the time
the sample was selected and the reference period of the collected injury and illness data. The sample for a given current
survey is selected from QCEW data that is approximately two years old. During this two year lag, establishments may
close, start up, or change employment size. The reported employment is provided by the responding establishment.
The TEI target employment is obtained from the most recent employment data from the QCEW. The SOII uses this
employment as a benchmark to adjust injury and illness estimates. Benchmarking is performed at the lowest estimating
level, the individual TEI level.

BMF

TEI target emploment
∑ TEI weighted reported employment

Industry benchmark factor ratios are produced and reviewed by the States. The BMF ratio for aggregate industry levels
higher than the TEI level is simply the ratio of the sum of the weighted employments of the next lower levels to the
target employment of the aggregate TEI level. Industry out-of-range BMFs have been defined and are usually caused
by a change in size class of an establishment. That is, there is a change in the assigned employment versus the reported
employment that causes a unit to change size classes. Estimates with BMFs out of range are not published unless a
waiver is requested by the State and approved by the national office.
For the more detailed case and demographic estimates for DAWF cases, the final summary weight applied to each case
is adjusted by additional factors to ensure that the number of usable cases that have been submitted represent the total
DAWF cases reported by the establishment used in the tabulation of summary estimates. More information on the case
and demographic estimates can be found in Chapter 9 of the BLS Handbook of Methods.

5. Estimation

The final summary weight is used to weight the cases of individual establishments to produce counts of injuries and
illnesses by various characteristics as ownership, industry and size class. For example, to produce an estimate for the
manufacturing industry in the private sector, weighted cases are summed for establishments with private ownership
within NAICS codes 310000 to 339999.
In addition to injury and illness counts, the SOII also reports the frequency (incidence rate) of such cases in terms of the
number of injuries and illnesses per 100 full-time workers. Incidence rates permit comparison among industries and
establishments of varying sizes. They express various measures of injuries and illnesses in terms of a constant
reflecting exposure hours in the work environment. The formula used for calculating incidence rates across injuries
and illnesses and for injury cases only is:
Incidence Rate

Sum of characteristic reported
200,000
Sum of number of hours worked

where the 200,000 represents 40 hours per week x 50 weeks x 100 full-time employees within the calendar year.
In this way, a firm with five cases recorded for 70 employees can compare its injury and illness experience to that of an
entire industry with 150,000 employees and 12,000 cases. To view all the estimates produced by the SOII, please visit
http://www.bls.gov/iif/.
Incidence rates for illnesses and for case and worker characteristic categories are published per 10,000 fulltime
employees. The above incidence rate calculation would then use 20,000,000 hours instead of the 200,000 hours to
represent the 10,000 full-time employees working 40 hours per week, 50 weeks per year.

6. Reliability of Estimates
Since a probability sample was used to produce estimates for occupational injuries and illnesses, these estimates
probably differ from estimates that would be obtained from a census. All estimates derived from a sample are subject
to sampling and non-sampling error. Sampling errors occur because observations (estimates) are made on a sample,
not the entire population (census). Standard errors are calculated to determine precision or error for each estimate in
the survey. The survey does not adequately report some long-term latent illnesses. The inability to obtain information
about all cases in the sample, mistakes in recording or coding the data, and definition difficulties are other examples of
non-sampling error in the survey. Beginning with the 2006 SOII, a quality assurance program was implemented to
address these potential non-sampling errors. This program is evaluated annually and modifications are made when
deemed necessary by program management.
The SOII uses a Taylor series linearization methodology to calculate estimates of standard errors for published
estimates. This method is flexible with the survey design and is relatively easy to program. Due to the number of
estimates produced for individual State and national estimates, other variance estimators are time consuming and
require ample computer storage.
Standard errors are used to determine if estimates meet publishable criteria defined by the OSHS program office.
Standard errors are also used in validation of statistical comparisons made within a publication. Relative standard
errors, standard errors divided by the estimate, are calculated for each estimate and are available on the BLS website
mentioned above.

References
BLS Handbook of Methods, Chapter 9, September 2008,
http://www.bls.gov/opub/hom/pdf/homch9.pdf
Cochran, Willam G. (1953), Sampling Techniques, New York: John Wiley & Sons, Inc.
Wolter, Kirk M. (1985), Introduction to Variance Estimation, New York: Springer-Verlag, Inc.

The Occupational Safety and Health Act of 1970,
www.osha.gov
Any opinions expressed in this paper are those of the authors and do not constitute policy of the Bureau of Labor
Statistics.