An official website of the United States government
Author: Graham Jones
This paper was published as part of the Consumer Expenditure Surveys Program Report Series.
Periodically, the Bureau of Labor Statistics Consumer Expenditure Surveys (CE) program includes additional questions in the fourth interview of the CE Interview Survey for program research purposes. From April 2020 to March of 2021, such a question was added to the CE Interview Survey fielding procedures, requesting that respondents provide full date of birth information for members of their consumer unit aged 14 or older.[1] The main goal of this research question was to gauge the willingness of CUs to disclose their members’ full birthdate information. This willingness to disclose birthdate data is of particular interest to the BLS, as this information is a helpful tool for facilitating improved administrative data linkages for member level data within each CU.
Administrative data linkages allow the BLS to work with other government agencies that already possess administrative data, collected for their own programmatic and regulatory purposes, so they may use that shared data to improve statistical products and further support the initiative for evidence-based policy making. These potential improvements to data quality would have clear positive spillovers for CE data users, as administrative data linkages in other household surveys have been shown to improve researchers’ ability to track the effectiveness of government programs, especially as it relates to difficult to reach populations (Harron, K., et al 2017) and for poverty analysis (Medalia, C et al 2019). On the other side of the survey-methods spectrum, administrative data hold value in their potential to cut costs associated with CE data collection, as the data in question already exist as a byproduct of the day-to-day business operations of another government entity (Harris-Kojetin, B. A., Groves, R. M, 2017).
Previous internal program research has shown the utility of administrative data for improving CE data, with past studies highlighting the potential to reduce measurement error and improve income imputation (Brummet, Q. et al 2017). However, recent efforts have been somewhat limited by insufficient data needed for linking CU members to administrative data sources, as currently the Interview Survey only includes an “age as of today” question during the CU rostering process. Research at the US Census Bureau on administrative data linking showed that using full date of birth information in the American Community Survey (ACS), as opposed to current age, resulted in an increase of roughly 10 percent in the matching rate.[2] This means that the additional information used in the creation of Protected Identification Keys (PIKs) resulted in a reduction of false matches, and an increase in the overall value of the linked data (Brummet et al, 2018).[3] This encouraging research and general trend toward the use of administrative data to improve Federal statistics motivated the BLS to consider the use of a full birthdate question.
In the past, the BLS fielded a full birthdate question in the CE Interview Survey, but it was taken out of circulation in 2009 due to a lack of business need for that level of Personally Identifiable Information (PII).[4] Since that time, the need for administrative data linkages increased for a variety of reasons (e.g., more available administrative data, decreasing survey response rates, rising data collection costs, etc.) leading to the experimental inclusion of the full birthdate question. The addition of questions to the already lengthy CE Interview Survey can have unintended consequences for respondent burden, and with outside research indicating that household survey respondents have had mixed responses to data linkage consent in Federal surveys, a preliminary test of respondent cooperation was necessary despite the clear business need for the birthdate question (Singer, E. et al 2011).[5] Also embedded within the instructions to implement the birthdate research question was a test of question framing.[6] This test, along with question placement and overall research design, will be covered in the following methodology and research question design section.
The addition of the birthdate question into the CE Interview Survey, as is the case for any new question, required forethought and considerations beyond the business need for the data. Along with literature on the use of administrative data, there is also a considerable amount of research on the placement of questions related to data linkage consent and their framing, which the BLS considered during the planning phases of the birthdate research question. The goals of the studies referenced were largely to find ways to optimize linkage consent rates while minimizing linkage consent bias, to the extent possible. While the BLS is not adding a consent question specifically, these studies on wording and framing of questions help inform BLS researcher’s understanding of the request for CU member birthdate data.
As it pertains to question placement, multiple contemporary studies have concluded that asking questions related to linkage consent were more effective when asked earlier in the survey (Sakshaug, J.W. et al. 2019) (Sakshaug, J. W., & Vicari, B. J, 2018). This has even been shown to be the case in the fourth wave of interview surveys, which is a more applicable finding for the CE program (Sala, E. et al. 2014). Research questions posed in the CE Interview Survey, like the birthdate question studied in this paper, are confined to the fourth interview research section, which occurs at the very end of the twenty-two-section survey.[7] With the previously cited findings in mind, one would assume that the placement of the birthdate question at the end of the CE Interview Survey was suboptimal, but this was not a factor that the researchers were able manipulate. While the placement of the birthdate question was not ideal, the fact that the question was posed to all respondents at the same point in the interview reduced the risk of noise when conducting the test of question framing.
Question framing, unlike question placement, was less restricted by the current Interview Survey format and provided more opportunity for experimentation. Past research on the effects of question framing suggests that framing questions in the context of the benefits provided, or “gain framing”, showed a positive impact on responses (Bates, N. et al. 2012) and (Sakshaug, J. and F. Kreuter, 2014). In a similar effort, the BLS created two versions of the birthdate research question, one for each half of the one-year survey fielding period analyzed, to measure the impact of question framing on the willingness of respondents to provide their household members full birthdate information. For the first two quarters of the survey fielding period, 2020q2 and 2020q3, respondents received an introduction explaining the purpose of the question, while respondents in the last two quarters, 2020q4 and 2021q1, were simply told that the BLS would like to collect some more demographic information at the end of the interview. The Computer Assisted Personal Interviewing (CAPI) instrument used for the CE Interview Survey was programmed to show the designated birthdate question to the respondent based on the quarter that the interview was occurring in.
In quarters 2020q2 and 2020q3, all respondents at the end of their fourth interview received the “explanatory wording” as an introduction to the Interview Survey research section.
“Thank you for the responses you have provided so far. The information you have provided is valuable, especially when we can refer to other records. This process is most accurate when we have dates of birth."
For those CUs whose fourth interview occurred in quarter 2020q4 or 2021q1, the respondent received the “generic wording” version of the research section introduction:
“Before we wrap up, we’d like to collect some additional demographic information on <you/your household> and find out what you thought about the interviews.”
In addition to analyzing differences in response due to the introduction wording, the inclusion of the birthdate research question allowed the BLS to evaluate the effect of the birthdate research question on respondent burden and explore measurement error by comparing collected ages asked in the first interview against ages calculated using member date of birth information. The remainder of this paper will cover analytical findings from the implementation of the birthdate research question, as well as recommendations for future research.
Before moving into the Results section, it is important to note that the inclusion of the birthdate question from 2020q2 to 2021q1 coincided with the onset of the COVID-19 pandemic, which led the Census Bureau to suspend in-person interviewing for all surveys it conducts, including CE, until July 2020.[8] The Census Bureau reinstated in-person interviews in certain localities in July of 2020, which resulted in the majority of the birthdate data being collected via telephone interview, with a slight increase in in-person interviews in 2020q4 and 2021q1.
The data utilized for this study contained birthdate information on the members of CUs who participated in the CE Interview Survey research question section from April 2020 to March 2021. The main variable of interest from the data was the calculated age of participating CU members. These calculated age values were generated using information gathered from the research questions, in which CUs were asked to provide the day, month, and year of birth year for each of their reported members. The presence of a calculated age for a given CU member signified that complete birthdate information was provided. Table 1.1 below shows the overall response rate to the birthdate research question, using the calculated age as the response indicator. The data show that of the 10,488 eligible CU members for which birthdate information was requested, complete information was provided by 8,464 CUs.[9] This left 2,024 members without a calculated age and equates to a total response rate of 80.7 percent at the member level for the birthdate questions.
Full Birthdate Info Provided | Percent |
---|---|
Yes |
80.7 |
No |
19.3 |
Overall |
100 |
When viewing the answers provided for CU respondents, rather than all CU members in the household, the data in Table 1.2 below shows that of the CU respondents who were asked the birthdate question, 83.3 percent provided complete information. This indicates that respondents were slightly more willing and/or able to provide their own birthdate information than for all CU members.
Full Birthdate Info Provided | Percent |
---|---|
Yes |
83.3 |
No |
16.7 |
Overall |
100 |
For the cases in which CU members’ ages were not able to be calculated due to inadequate birthdate information, we analyzed which pieces of information that respondents were willing to provide (i.e., birthday, birth month, and birth year). Figure 1.3 below shows the level of detail associated with birthdate refusal cases.
The data show that the majority of these cases were associated with a full refusal, with just over 61 percent providing no birthdate information. In roughly 29 percent of cases, only a birth year was provided, while a little under 8 percent provided only their birth month and year. For a few other cases, ‘month and day’, ‘month only’, and ‘day and year’ were provided, each representing, at most just over 1 percent of non-consent cases. In terms of improvements to data linkage efforts, in only 9.8 percent of non-calculated age cases, was birthdate information provided beyond what could already be gathered from the “age as of today” question.
As described in the methods section, the birthdate research question was programmed into the CAPI instrument to provide different question wording between the first two and the last two quarters of the birthdate questions fielding. In the first two quarters interviewers read the explanatory wording to respondents, and in last two quarters; they read the generic wording as an introduction to the research question section. This design served as an embedded test of question framing. To evaluate the differences in response, the response rate to the birthdate research question was analyzed for each question framing group and compared. Figure 2.1 below shows the rate at which complete birthdate information was provided for CU members by the birthdate question framing the CU received.
Both groups of CUs provided birthdate information for their members at about the same rate (roughly 81 percent). This would lead us to believe that the wording of the research question introduction likely had little to no influence on the overall rate at which CUs provided complete birthdate information for their members. These results would also imply that protocols for in-person interviews, which changed over the birthdate research question fielding period, also did not impact compliance with the birthdate question.
After analyzing the rate at which birthdate information was provided for CU members, the calculated ages were further examined to verify the accuracy of the birthdate information collected. To do so, calculated ages from the birthdate research question data were compared to the ages provided for CU members during the household rostering and demographic information section at the beginning of the interview process. Table 3.1 below shows this rate of agreement for the 8,464 CU members for which an age was calculated (i.e., the rate at which the two age variables matched for a given CU member).
Status | Percent Agreement |
---|---|
Exact Match |
76.6 |
Within 1 Year |
19.3 |
Not Matched |
4.1 |
Overall |
100 |
The percentage of exact matches between calculated and reported CU member ages was 76.6 percent. On its face this does not look like an encouraging rate, but when considering that reported member age is collected in the first interview and calculated age was based on a fourth interview research question, it leaves open the possibility that a reported age could be a year off from the calculated age for a CU member. When factoring in differences of only one year, we see the rate of agreement jump up to 95.9 percent.
Next, the demographic characteristics of those who were asked the birthdate research question were compared to those who were not asked. The goal was to understand potential biases that may arise from including a birthdate question. Eligibility for the birthdate question was determined by two criteria, first the CU had to be participating in the fourth wave interview, and second the CU member in question had to be at least 14 years of age. Since this question is only asked in the fourth wave it is important to note that CU respondents participating through the fourth wave tend to have different demographic characteristics than the rest of the sample. Table 4.1.1 illustrates the demographic composition of CUs interviewed during the birthdate research question fielding period (2020q2 to 2021q1), as well as the differences between CUs that were asked the birthdate research question and those who were not. All demographic variables in the comparison below were analyzed at the respondent level.
Demographic category | CU Respondents | Birthdate Asked | Birthdate Not Asked (i.e., Interviews 1-3) | Diff (DOB Not Asked minus DOB Asked) |
---|---|---|---|---|
No. of Respondents |
20,158 | 5,438 | 14,720 | |
Percent Distribution of Demographic Variables |
||||
Race of Respondent |
||||
White |
81.5 | 82.0 | 81.4 | 0.6 |
Black |
10.4 | 10.2 | 10.4 | -0.2 |
Other (incl. Asian, multi, other) |
8.1 | 7.8 | 8.2 | -0.4 |
Hispanic Origin Status |
||||
Hispanic |
14.3 | 14.1 | 12.6 | 1.5 |
Non-Hispanic |
87.0 | 85.9 | 87.4 | -1.5 |
Area Type |
||||
Urban |
82.5 | 82.5 | 82.5 | 0 |
Non-Urban |
17.5 | 17.5 | 17.5 | 0 |
Age of Respondent |
||||
Under 25 years |
3.5 | 3.4 | 3.5 | -0.1 |
25-34 years |
13.6 | 13.0 | 13.8 | -0.8 |
35-49 years |
24.4 | 24.6 | 24.4 | 0.2 |
50-64 years |
29.2 | 28.9 | 29.3 | -0.4 |
65 years and older |
29.3 | 30.1 | 29.1 | 1.0 |
Education of Respondent |
||||
Less than high school |
+ | + | + | + |
High school graduate |
29.1 | 30.0 | 28.7 | 1.3 |
Some college |
29.8 | 30.2 | 29.6 | 0.6 |
College graduate |
41.2 | 39.8 | 41.6 | -1.8 |
CU Size |
||||
Single person |
30.9 | 30.8 | 30.9 | -0.1 |
2-3 persons |
48.1 | 48.1 | 48.1 | 0 |
4+ persons |
21.0 | 21.2 | 21.0 | 0.2 |
Housing Tenure |
||||
Non-Owner |
32.9 | 32.7 | 33.0 | -0.3 |
Owner |
67.1 | 67.3 | 67.0 | 0.3 |
+Indicator that value is positive, but less than one-half of one percent. |
||||
Demographic category | CU Respondents | Birthdate Asked | Birthdate Not Asked (i.e., Interviews 1-3) | Diff (DOB Not Asked minus DOB Asked) |
---|---|---|---|---|
No. of Respondents |
20,158 | 5,438 | 14,720 | |
Mean Income Before Tax |
$86,776 | $83,429 | $88,013 | ($4,584) |
Median Income Before Tax |
$60,326 | $58,256 | $61,535 | ($3,280) |
Expenditure Total |
$14,856 | $14,395 | $15,026 | ($631) |
Table 4.1.1 shows little variation in CU respondent characteristics whether they were or weren’t asked the birthdate research question. Both the percent distribution of respondents identifying as Hispanic and those with a college education varied between the two groups. The data in Table 4.1.1 show a higher rate of respondents who identified as Hispanic among those asked the birthdate question (14.1 percent) relative to those not asked (12.6 percent). Table 4.1.1 also shows that 39.8 percent of those asked the birthdate question were college-educated, compared to 41.6 percent of those who were not.
Table 4.1.2 shows a similar comparison, but with a focus on income and expenditures rather than the traditional demographic variables shown in Table 4.1.1. The data in Table 4.1.2 show that CUs asked the birthdate research question had mean and median imputed income values of $4,584 and $3,280 lower than those not asked the question.[10] Regarding average total expenditures, the data show a difference of $631 between those CUs asked the birthdate question and those who weren’t. Overall, the relatively low level of demographic variation between the two groups is a positive finding, as it would imply a low likelihood of bias in the results from the fielding of the birthdate question.
Tables 4.2.1 and 4.2.2 show the demographic breakdown of CU respondents by their answer to the birthdate research question. The second column from the left shows a demographic breakdown of CU respondents who were asked the birthdate research question. The third column shows a breakdown of CU respondents who provided full birthdate information for themselves (i.e., birthday, birth month, and birth year). The fourth shows a breakdown of CU respondents who did not provide full birthdate information. In these cases, at least one piece of birthdate information (i.e., birthday, birth month, and birth year) was not provided by the respondent. For this demographic breakdown, variables were analyzed at the respondent level, except for age which is shown in Table 4.3 at the. Age was analyzed for all CU members due to the potential interaction between a CU member’s age, and the respondent’s willingness to provide full birthdate information on their behalf. The data show that full compliance with the birthdate research question varied by respondent race, respondent Hispanic origin status, CU area type, CU size, and CU member age.
Demographic category | Birthdate Asked | Full Birthdate Given | Birthdate Information Incomplete | Diff (DOB Given minus Incomplete) |
---|---|---|---|---|
No. of Respondents |
5,438 | 4,528 | 910 | |
Percent Distribution of Demographic Variables |
||||
Race of Respondent |
||||
White |
82.0 | 83.2 | 76.0 | 7.2 |
Black |
10.2 | 9.8 | 11.9 | -2.1 |
Other (incl. Asian, multi, other) |
7.8 | 7.0 | 12.1 | -5.1 |
Hispanic Origin Status |
||||
Hispanic |
14.1 | 13.4 | 17.5 | -4.1 |
Non-Hispanic |
85.9 | 86.6 | 82.5 | 4.1 |
Area Type |
||||
Urban |
82.5 | 81.8 | 86.0 | -4.2 |
Non-Urban |
17.5 | 18.2 | 14.0 | 4.2 |
Education of Respondent |
||||
Less than high school |
+ | + | + | + |
High school graduate |
30.0 | 29.9 | 30.6 | -0.7 |
Some college |
30.2 | 30.4 | 29.1 | 1.3 |
College graduate |
39.8 | 39.7 | 40.3 | -0.6 |
CU Size |
||||
Single person |
30.8 | 31.5 | 26.9 | 4.6 |
2-3 persons |
48.1 | 47.9 | 49.2 | -1.3 |
4+ persons |
21.2 | 20.6 | 23.9 | -3.3 |
Mean Income Before Tax |
$83,429 | $83,382 | $83,660 | ($278) |
Median Income Before Tax |
$58,256 | $57,729 | $59,903 | ($2,174) |
Expenditure Total |
$14,395 | $14,533 | 13,707 | $826 |
Housing Tenure |
||||
Non-Owner |
32.7 | 32.8 | 32.2 | 0.6 |
Owner |
67.3 | 67.2 | 67.8 | -0.6 |
+Indicator that value is positive, but less than one-half of one percent. |
||||
Demographic category | Birthdate Asked | Full Birthdate Given | Birthdate Information Incomplete | Diff (DOB Given minus Incomplete) |
---|---|---|---|---|
No. of Member Level Cases |
10,488 | 8,464 | 2,024 | |
Age of Member |
||||
14 - 25 years |
12.3 | 11.7 | 15.0 | -3.3 |
25-34 years |
14.3 | 14.2 | 14.3 | -0.1 |
35-49 years |
23.6 | 23.1 | 25.4 | -2.3 |
50-64 years |
25.5 | 25.5 | 25.3 | 0.2 |
65 years and older |
24.4 | 25.4 | 20.0 | 5.4 |
Beginning with CU respondent race, the analysis shows that there was a higher proportion of respondents whose race was identified as white among cases where complete birthdate information was given (83.2 percent) relative to those that did not provide complete information (76 percent). The data also show that there was a lower proportion of respondents who identified as a member of a race other than white or black and provided full birthdate information (7 percent) compared to incomplete birthdate cases (12.1 percent). Similarly, respondents who identified as being of Hispanic origin made up a lesser share of complete birthdate cases (13.4 percent) relative to those who provided incomplete birthdate information (17.5 percent). Regarding CU area type, urban CUs accounted for 86 percent of cases where incomplete birthdate information was given, compared to 81.8 percent of cases where complete information was given. For CU size, Table 4.2 shows that single person CUs made up a greater proportion of cases where complete information was provided (31.5 percent) compared to incomplete cases (26.9 percent). The data also show that CUs with a size of four or more members made up a higher proportion of incomplete cases (23.9 percent) compared to complete cases (20.6 percent).
When examining discrepancies in birthdate question answers by age group at the member level, the data in Table 4.3 show that responses did vary by age group, with the greatest differences between the youngest and oldest groups. CU members under 25 years of age accounted for a higher rate of incomplete birthdate cases (15 percent) compared to complete cases (11.7 percent). The opposite was true for the 65 year and older group, in which they accounted for a higher rate of cases where complete information was provided (25.4 percent) compared to incomplete cases (20 percent).
Following the birthdate question, respondents in their fourth interview were asked a set of standard research questions meant to gauge survey burden, difficulty, sensitivity, and length from their perspective. The results from this set of questions are routinely used by the BLS to assess overall levels of survey burden over time. This section will examine respondents’ perceived burden and compare the results from the birthdate dataset to previous quarters without the birthdate question. A comparative analysis of the perceived burden between the birthdate question framing groups is also included in this section. Before proceeding into the results, it should be noted, that the timing of the birthdate research question fielding period (2020q2-2021q1) coincided with the halt on in-person interviews due to the COVID-19 pandemic. This may have affected how long or burdensome respondents perceived the interview.
5.1 Burden Comparison to Previous CE Interview Data
Figure 5.1.1 below shows the standard metrics for assessing CE Interview Survey burden, over eight quarters of data collection. The first four quarters of burden data, 2019q2 through 2020q1, cover the year of data collection that preceded the inclusion of the birthdate research question, and the last four cover the birthdate question fielding period. Burden varied little, with the exception of a slight reduction in 2019Q4, over the eight quarters of data shown, but rose slightly in the final quarter, 2021q1.
The five standard burden options typically given to respondents were collapsed into three categories to better examine the relationship between the birthdate question and perceived burden.[11] The standard burden response options were collapsed to: ‘burdened’, ‘unburdened’, and ‘neutral’, while missing responses were ignored. The same was done for the responses to the research questions on difficulty, sensitivity, and length of the survey to observe reported sentiments more accurately.
This analysis revealed that responses to the perceived burden question varied little between respondents in and before the birthdate fielding period. Further analysis of the responses to questions on sensitivity and survey length questions showed some variance between the groups. These results should be interpreted with some caution, due to other variables for which this study was not able to control: Specifically, the onset of the COVID-19 pandemic and the resulting interviewing protocol changes on respondent behavior and perception of survey questions. Figure 5.1.2 below illustrates the differing levels of perceived survey question sensitivity, comparing the quarters of data collection when the birthdate question was fielded to those of the CE Interview Survey production data collected one year earlier.
The data show that respondents from the sample that received the birthdate research question perceived the survey questions to be sensitive at about the same rate as those from the previous year (17.7 percent compared to 18.1 percent) but said the questions were not sensitive at a lower rate (57.6 percent compared to 59.9 percent). This difference is mostly accounted for in the increase in the share of “neutral” feelings (22.4 percent to 24.3 percent) about question sensitivity from those that received the birthdate research question.
Figure 5.1.3 below shows the differences between the two groups in terms of perceived survey length. In this case, there was a starker difference between the sample groups responses, with 50.4 percent of those that received the birthdate research question perceiving the survey as “Long” compared to 41.9 percent in the production sample from the previous year. The previous year’s production sample also perceived the survey as “Not Long” at a higher rate than those that received the birthdate research question, with 18.4 percent compared to 13.2 percent, respectively.
This increase in perceived survey length was accompanied by an increase of about 2 minutes for fourth wave interviews in the Interview Survey from 2020Q1 to 2020Q2, but this was likely due to the change in interviewing protocols over that time period as opposed to the introduction of the birthdate question.[12]
5.2 Burden Comparison by Birthdate Question Framing
Next, differences in perceived burden between the explanatory wording and generic wording groups were analyzed (i.e., 2020q2-2020q3 versus 2020q4-2021q1). In this case, perceived burden was found to vary slightly between the two groups. Figure 5.2.1 below illustrates these differences. The data show that a higher percentage of the generic wording group (2020q4 through 2021q1) reported being burdened, with 15.4 percent compared to 13.1 percent in the explanatory wording group (2020q2 through 2020q3).
While it is unlikely that the shortened version of the birthdate research question given to respondents between 2020q4 and 2021q1 had an impact on perceived burden between the two groups, this outcome did correspond to the overall uptick in burden shown in Figure 5.1.1 where burden can be seen rising in 2021q1.
As was briefly noted in the Methodology & Research Question Design section, there were two factors that hindered the interpretation of results from this study. The coincidence of the birthdate research question’s fielding with the suspension of in-person interviewing for the CE Surveys, was the first limiting factor for this study. This halt in normal interviewing procedures, due to public health measures in response to the COVID-19 pandemic, necessitated a switch to phone interviews until restrictions began to lift in some localities in July of 2020. The external shock to how interviews were conducted brought about data quality fluctuations. Parsing out these mode effects went beyond the scope of this original study. Furthermore, this does not even begin to address the complexity of controlling for the social and psychological impact of the pandemic on respondent behavior. Thus, making it even more difficult to make meaningful comparisons of the birthdate data to past production data.
Beyond the external complications of the COVID-19 pandemic, there was also the limiting factor of question placement. As previously discussed, the birthdate question was only asked to fourth interview respondents near the very end of the interview. This placement may have inflated the rate of full compliance with the birthdate question, as fourth wave respondents are thought to be relatively more cooperative and responsive than those from other waves. Recent internal research has shown, however, that there is little evidence of wave-related bias in the CE Interview Survey, and that bias is much more of a concern when sub-setting fourth wave respondents further to only those with previous complete interview outcomes. As research questions are asked of all fourth wave respondents, not just those with previous complete interviews, this should mitigate the potential bias associated with the birthdate question being asked in the fourth wave.
The author of this paper recommends that in future research, the full birthdate question be asked earlier in the CE Interview Survey, which would be in line with the previously cited survey methods literature on using early question placement to maximize compliance (Sakshaug, J.W. et al. 2019). The results showed that a high percentage of respondents provided birthdate information on behalf of their CU members beyond what is currently asked, but this could potentially be improved by more optimal question placement. For the question framing, the results of the test did not indicate that either wording resulted in higher rates of compliance, so the author would recommend following the established literature on asking questions in a “gain framing” context to show the added value this question provides to all the survey data being collected.
The results covered in the report also confirmed that the inclusion of the birthdate research question did not have a negative impact on the various metrics of respondent burden, The birthdate questions apparent associations with perceived respondent burden and survey length were likely caused by pandemic restrictions put in place on in-person interview surveys in March of 2020, which happened to coincide with the start of the birthdate question fielding period. While the study was limited to fourth wave participants, the results imply that it is unlikely the inclusion of a full birthdate question would lead to a significant increase to respondent burden.
With respect to future research, there are currently no business needs or proposals that would require full birthdate information from CE respondents, but should that need arise, the findings from this paper will inform additional analyses that would facilitate incorporating a birthdate question into the questionnaire. In this event, the next step would likely be to test the impact of birthdate question placement. The information gathered could then be used to conduct a CE-based PIK exercise, with exact birthdate information from CU members, to test matching rates and confirm or deny whether they improved from past research efforts. This would also open the door for research on CU member-level linkages to administrative records like IRS income data, HUD public housing data, social security data and other governmental organizations.
[1] A consumer unit (CU) is the measurement unit used by the BLS to refer to a group of individuals living together within the same living quarters and having financial interdependence. All individuals related by blood, marriage, adoption, or other legal arrangements living together are considered a CU. Unrelated individual living together may each constitute separate CUs if they are financially independent. Financial independence is determined by the three major expense categories: Housing, food, and other living expenses. To be considered financially independent, at least two of the three major expense categories have to be provided entirely, or in part, by the respondent. For a complete definition, see the CE glossary at https://www.bls.gov/cex/csxgloss.htm
[2] When household survey data is linked with external administrative data containing additional information on the individuals sampled, the datasets are merged using a key variable know as a Protected Identification Key (PIK). The matching rate refers to the rate at which individuals sampled in a household survey, are successfully linked to their own information in an administrative dataset. This rate can be used to assess the fitness for use of a PIK.
[3] A Protected Identification Key (PIK) is a key variable used to merge household survey data with outside administrative data sources. The PIK variable value is often a piece of personally identifiable information that is collected about individuals for household survey purposes, that is also collected by another administrative entity with data of interest to researchers.
[4] In 2011, the CE stopped asking respondents to provide information about the date of birth for each CU member. This was replaced with the question, “As of today, how old are you?” At the time, this decision was made to reduce any potential privacy concerns from respondents, as there was no business need for exact date of birth.
[5] Data linkage consent refers to a request for respondent permission to link data collected from a survey to administrative data sources outside of the survey program, for research purposes. As administrative data are not expressly collected for research purposes, consent is required for research use in most cases.
[6] Question framing refers to the context provided to respondents on the purpose of a question. Framing is often intended to educate survey respondents and encourage response.
[7] The CE Interview Survey is divided into twenty-two distinct sections, not including the process of CU member identification. More information on the sections that comprise the survey can be found on the official BLS site.
[8] The BLS contracts with the U.S. Census Bureau to perform data collection services for the CE Surveys. These services include, but are not limited to, conducting interviews in-person or over the phone. In 2020, the Census Bureau suspended in-person interviews due to health and safety concerns related to the COVID-19 pandemic.
[9] It should be noted that there were four CU members for which the birthdate question was asked, but they were not eligible for due to age.
[10] The estimates of mean and median income in Table 4.1.2 were based on data that include imputed income values, and it is worth noting that there may be a relationship between imputation rates and response to the birthdate question (e.g., a respondent who is uncomfortable providing income information may also be uncomfortable providing birthdate information). That stated, while the relationship of propensity to provide birthdate information and probability of requiring imputation of income (or vice versa) is an interesting question, it is beyond the scope of this report to investigate. If you are interested in more information about the income imputation process used in the CE Surveys, please see "User's Guide to Income Imputation in the CE".
[11] The following standard survey burden response options are given to CE respondents; ‘extremely burdensome,' 'somewhat burdensome,' 'neutral,' 'slightly burdensome,' and 'not at all burdensome'.
[12] Median interview time data in the CE Interview Survey in measured by the CE Data Quality Profile team. The metric reports of which are published on a biannual basis on the BLS webpage.