Starting in January 2023, data users may note some changes to the Current Population Survey (CPS) public use files (PUF) designed to provide added confidentiality protection to these data. For an overview and background on these changes, please see https://www.census.gov/programs-surveys/cps/technical-documentation/user-notes/2023-cps-puf-changes.html.
This User Note provides additional detail about these changes and addresses considerations that data users may have when using the data.
The changes will be phased in on a monthly basis. This means that in January 2023 only Month-in-Sample (MIS) 1 will reflect the changes. This phase-in will continue for 16 months until April 2024, at which time all cases on the CPS microdata files will reflect the revisions listed below. This phase-in procedure will allow users to continue to conduct longitudinal analyses without having a break in series.
Starting in January 2023, all MIS 1 cases will be assigned Household Identification Numbers (HRHHID1) using an algorithm different than the one used in the current files. As with the current files, these numbers will remain constant over the time that a household remains in sample such that users will be able to use the same matching identification for the life of the case. This, along with HRHHID2, will uniquely identify a household. This is for your information and the data user will not have to do anything different than in previous data use.
Section A: Geographic Synthesis
All Geography with population between 100,000 and 249,999 will go through a geographic synthesis for privacy protection. Items for consideration include the following:
a. GTCBSA
b. GTCBSAST
c. GTCBSASZ
d. GTCO
e. GTINDVPC
f. GTMETSTA
g. GTCSA
Determining which cases are eligible for synthesis
The following are key things to consider when looking at data that maybe eligible for geographic synthesis:
For example, the Salisbury CBSA is contained within the states of Maryland and Delaware, with each state-specific portion of the CBSA having a population of 100,000 to 249,999. Thus, each state portion will be subject to the synthesis. Conversely, the Wisconsin portion of the Minneapolis CBSA is subject to synthesis, but not the Minnesota portion since its population is over 249,999. Additionally, there are some CBSAs only partially in sample. For these areas, the population for the portion of the area in sample will be used to determine whether it requires synthesis.
What follows below are examples based on user questions and comments that arose as the Census Bureau vetted these changes to the CPS PUF over the course of the last year. We hope that these examples help to understand the scope of these changes more clearly.
Analysis
Analysts at the Census Bureau reviewed 3 non-consecutive months of basic monthly public use data files to examine potential issues with the synthesized geographies. (As always, the Census Bureau urges caution when analyzing small sample sizes or subpopulations. For example, Census guidance recommends pooling three years of CPS ASEC to produce state-level estimates.)
Analysts examined demographics (age, marital status, sex, race, ethnicity, cohabitation status, household type) and key migration variables for differences in distributions between public use basic monthly files with and without geographic synthesis techniques applied. Additionally, analysts examined how geographic synthesis may affect the relationship between full-time employment status at the household level and household-level regressors, such as family income, race and ethnicity, education, average age, household tenure, employment status, disability, citizenship, marital status, having children, and veteran status. Analysts found no meaningful differences in distributions across the two sets of files, nor was there considerable evidence that the geographic synthesis changed the relationship between employment status and the examined variables.
One area of potential concern raised by external data users was the impact on the Supplemental Poverty Measure (SPM), which has geographically-adjusted thresholds. Census publications will continue to utilize the internal CPS ASEC files and therefore will be unaffected by changes in the CPS PUF. Nonetheless, estimates derived from the public use files were examined to see potential impacts on external data users. For the available subset of the 2022 CPS ASEC, SPM thresholds were re-calculated using synthesized geographies and used to re-estimate SPM rates overall and for detailed demographic subgroups (age, race and Hispanic origin, and CBSA status). The overall SPM rate was not statistically different when using geographically synthesized thresholds. However, significant differences were found with some demographic subgroups and regions, indicating the need for additional research once the geographic synthesis has been fully implemented. Differences in the distribution of people by income-to-poverty thresholds did not significantly change when estimating supplemental poverty using the new synthesized geographies. This remained true when looking at subsamples by CBSA status.
As part of commitment to data users, Census will review the impacts on the current production data and release occasional reports on any possible impacts on data quality. Census is constantly reviewing new methods for privacy protections and will refine the model should a better method be found that may lessen any impacts.
Section B – Changes to Earnings
Note: All changes in this section will not appear until April 2023 when MIS 4 cases are first phased in.
Rounding of the Usual Weekly Earnings Data
Provided below are the rounding algorithms that will be introduced for the implementation of rounding for the CPS files. Table 1 is the rounding rules for the following hourly earning variables: PTERNHLY, PTERNH1O, PTERNH2, and PTERNH1C. Table 2 is the rounding for the weekly earnings variables: PTERNWA, PTERN, and PTERN2. Both variables still have the two implied decimals.
Hourly Rounding Examples
For example, a value of $4.31/hour rounds down to a value of $4.30. A value of $4.78/hour rounds up to a value of $4.80.
A value of $8.22/hour rounds down to $8.20, while a value of $8.28/hour rounds up to $8.30.
A value of $32.07/hour rounds down to $32.00 while a value of $32.64/hour rounds up to $32.75.
Weekly Rounding Examples
For example, a value of $217/week rounds up $218, while a value of $228.34/week rounds down to $228.00.
A value of $3,622.50/week rounds down to $3,620.00, while a value $3,653.00/week rounds up to $3,660.00.
Topcoding of Usual Weekly Earnings
Prior to this point in time, the CPS used a static topcode for these items that did not change month to month. Currently, that value has been topcoding about 4 -5 percent of the values in these fields, while the Census Bureau’s requirement is the top 3 percent. A dynamic topcode of 3 percent will be set each month moving forward.
As part of the new topcoding, the topcode will be calculated as a weighted average for all cases subject to the topcode. By making this change, users will now have the ability to calculate accurate estimates for the aggregate combined earnings for all the records in sample. As topcodes may vary from month-to-month, care should be used when estimating average aggregate values over two or more months. It is suggested that a user works with the lowest topcode for the months under review to guarantee that all values greater than the smallest topcode are accounted for.
New Flag – PRERNMIN
In order to identify wages that were originally reported below the Federal Minimum wage even with rounding on the PUF, all records originally reported below 725 ($7.25) hourly will be flagged. PRERNMIN is set to 1 if below minimum wage was reported. PRERMIN can be found on the PUF in position 536-537.
For questions on technical issues involving these revisions, contact the CPS staff at: 301-763-3806 or email at DSD.CPS@census.gov