Non-Substantive Change to the 2015 Optimizing Self-Reponse Test

Request for a Non-Substantive Change to the 2015 Optimizing Self-Response Test

OMB No. 0607-0981

Purpose

The Census Bureau is requesting approval to conduct a reinterview of 5,000 housing units who responded to the 2015 Optimizing Self-Response Test conducted in the Savannah, Georgia media market. The addition of the reinterview will allow us to learn about our ability to validate responses provided without a unique Census identification code.

Background

The 2015 OSR Test was designed to increase internet self-response by testing multiple contact and notification strategies. The results from this test are critical inputs to the Census Bureau’s design decision in the fall of 2015 on how major cost savings may be realized for the 2020 Decennial Census.

The foundation of the 2015 OSR design is a contact approach termed “internet push.” This refers to a series of four mailed contacts to each housing unit, each designed to promote self-response via the internet questionnaire. Contacts include (1) a letter of invitation to complete the questionnaire online; (2) a postcard reminder with the questionnaire URL; (3) a second postcard prompt also including the questionnaire URL; and (4) a hard copy questionnaire for mailing back.

This internet push strategy was supplemented with an outreach and communications campaign targeting the entire local media market. Non-sampled households who did not receive any of the internet push mailings but were exposed to the advertising materials were also invited to participate in the 2015 OSR test. Online test that were completed without a Census ID are referred to as “Non-ID” cases. There are two sources of Non-ID cases in this test. The first is a panel for which respondents were not provided with the existing Census ID. The second is cases that were not in sample for this test but were potentially reached through the communications campaign in the site.

Proposed Reinterview of Test Responses

Prior to the 2015 OSR Test, Non-ID submissions had never been tested in the context of large-scale advertising. The types of questionnaires that are classified as Non-ID have been utilized in previous censuses, and functionality of the process, in particular on the internet, had been previously tested, but its impact on community-level response rates was unknown. Not having tested this approach in “real life” conditions we were unable to anticipate the full scope of requirements. Based on the results we are seeing in production, we are proposing to supplement the original design with the addition of a reinterview of a sample of the test respondents focusing primarily on the Non-ID responses.

The proposed reinterview falls directly within the scope and intent of the original test design to identify duplicate or potentially fraudulent Non-ID responses. Conducting the reinterview will contribute to our initial measure of our ability to validate respondent identity.

The proposed reinterview will involve a sample of 5000 of the original responses selected from the households received with a Census identification code, the households who chose to participate in the panel of the Census test without a Census identification code, and the households who chose to participate in the 2015 Optimizing Self-Response Test without a Census identification code who learned of the test through the communications campaign. Re-contacting a sample of the original responses and comparing these responses to the original responses will provide data points, in addition to those provided through administrative records, from which we can validate a respondent’s identity.

Reinterview Design

The proposed reinterview will leverage data collection modes authorized under the clearance established for 0607-0981. Of the 5000 sample cases, 4000 of the cases will be contacted by telephone using the telephone questionnaire assistance instrument in an outbound call capacity. The remaining 1000 cases will be contacted through an in-field data collection using the automated mobile instrument being used in the 2015 Census Test in Maricopa County, Arizona.

A sample size of 5000 was selected based on margins of error calculations. The confidence interval depends on the proportion of the population that can verify their initial response. If that proportion is 90%, the overall confidence interval is expected to be 0.9 +/- 0.0152 for the non-ID panel. Alternatively, if we want to consider a characteristic that represents 20% of the population, an estimate of 90% would have a confidence interval of 0.9 +/- .0339 for the Non-ID panel. Calculations were performed at various levels, and 5000 was seen as the reasonable balance between minimizing sample size and detecting differences across the interviews. Details of these calculations are contained in a supplemental spreadsheet.

Schedule

The 2015 OSR Test will end all advertising and internet data collection on May 31, 2015. The proposed reinterview data collection would occur from July 15 – August 7, 2015.

Burden

Estimating a reinterview questionnaire completion time at 10 minutes, the burden to the public is approximately 835 hours.

2