Policy Hub: Macroblog provides concise commentary and analysis on economic topics including monetary policy, macroeconomic developments, inflation, labor economics, and financial issues for a broad audience.
Authors for Policy Hub: Macroblog are Dave Altig, John Robertson, and other Atlanta Fed economists and researchers.
Comments are moderated and will not appear until the moderator has approved them.
Please submit appropriate comments. Inappropriate comments include content that is abusive, harassing, or threatening; obscene, vulgar, or profane; an attack of a personal nature; or overtly political.
In addition, no off-topic remarks or spam is permitted.
February 9, 2023
Population Control Adjustment's Impact on Labor Force Data: The 2023 Edition
Regular readers of Policy Hub: Macroblog will recall my description last year of smoothed labor force data, which reflect the latest population control adjustments by the US Bureau of Labor Statistics (BLS). I'm writing this post to let you know that I have updated those smoothed labor force data to incorporate the latest adjustments. You can find these smoothed series in this spreadsheet. As you may recall, each January, the BLS incorporates updated population estimates from the US Census Bureau into the data from the household survey used to construct important statistics such as the labor force participation (LFP) rate and unemployment rate. The BLS noted that the majority of the overall population change incorporated into the latest adjustment reflected recent increased international migration, following a period of subdued international migration due to the pandemic, along with various methodological improvements.
The BLS does not revise historical data to incorporate the population control adjustments, a fact that could make comparisons of labor force data over time a bit misleading. However, the BLS does show the impact of the population adjustments for a selection of labor force series and population characteristics for December of the preceding year. To construct historical labor force series that are more comparable over time, I use those estimated impacts to implement a simple smoothing method the BLS used previously to account for annual population control adjustments (described here ). This method essentially distributes the level shifts that result from the population control adjustments back over the relevant historical period for each series.
More specifically, I first smoothed data for each year from 2012 to 2020 (2012 being the year when the 2010 census estimates were first incorporated). Then, I smoothed the data for 2012 to 2021 to account for the effects of the 2020 census population control adjustment introduced in January 2022. Finally, I smoothed the data for 2022 to reflect the latest population control adjustment introduced in January 2023. I applied this method separately to the statistics for which the BLS provides population adjustment impact estimates, and I adjusted the data using (where available) published seasonal adjustment factors. The linked spreadsheet contains smoothed estimates for the population, labor force, employment, labor force participation rate, and employment-to-population rate for selected population characteristics for 2012 through 2022.
As the following table shows, the effect of the latest population-control adjustment on broad age-group labor force participation rates for December 2022 was generally smaller than the impact of last year's adjustment on the December 2021 participation rates.
For example, the BLS estimated that the population adjustment impact on the December 2021 LFP rate for the population aged 55 and older group was 0.7 percentage points, and this adjustment contributed to a 0.3 increase in the overall LFP rate relative to the published estimate. The BLS also estimated that the population-adjustment impact on the December 2021 LFP rate for the population aged 16–24 was −0.3 percentage points.
For December 2022, the BLS estimates that the LFP rate for the overall 55-and-older population would have been 0.1 percentage points lower than the published data indicated. Looking at the underlying population data, it appears that this adjustment resulted from an increase in the estimated size of the population aged 75 and older (up 1.5 percent, an unusually large amount, in the published population estimate between December 2022 and January 2023) and a decline in the population aged 65–69 (down 1.4 percent between December and January). At the other end of the age spectrum, the LFP rate for the 16–24 population would have been 0.5 percentage points higher than the published estimate. This adjustment seems to be the result of an increase in the estimated size of the size of the population aged 20–24 (the published estimate of the population aged 20–24 rose 5 percent between December and January, whereas the size of population aged 16–19 was essentially unchanged).
Chart 1, which compares the smoothed and published LFP rate for the population aged 16 and older, depicts the impact of smoothing on the historical data.
The latest population adjustments don't significantly affect the basic story of the overall LFP rate. This rate changed little over the course of 2022 and is still lower than it was pre-COVID, and the pre-COVID LFP rate was probably higher than the published data suggest. The biggest factor influencing the recent behavior of the overall LFP rate has been the lower participation by the population aged 55 and older, which reflects the combination of an aging population and a greater propensity of the older population to be retired than they were pre-COVID (see, for example, this recent study ). This drop in participation by the older population is evident in chart 2, which compares the smoothed and published LFP rates for the population aged 55 and older.
Similar to last year, the population control adjustment didn't affect the LFP rate for the overall 25–54 (prime) age group. As chart 3 shows, there is no difference between the smoothed and published prime-age LFP rates, and they have been fluctuating during the last year at close to their pre-COVID levels.
Later this year, the Census Bureau will publish updated monthly population estimates and projections for 2022 and 2023 for individual ages that will allow more careful adjustments to LFP rates for finer age groups than the BLS provided. In the meantime, I hope you find the smoothed labor force series useful.
January 9, 2023
The Wage Growth Tracker with Rounded Wage Data: The Final Plan
On December 15, 2022, the US Census Bureau released its final plan for improving disclosure avoidance procedures for the Current Population Survey Public Use Files (CPS PUF), and that plan is available here. As you may recall, we here at the Atlanta Fed have been keenly interested in the proposed changes because we actively use the public use files to produce statistics such as the Wage Growth Tracker.
Part of the plan to avoid disclosure of individuals in the CPS-PUF is to round the CPS PUF earnings data. Previously, I have written about how the initial proposed rounding rules for hourly and weekly wages would have harmed the reliability of the Wage Growth Tracker (see here and here). In this Policy Hub: Macroblog post, I take a look at how the final plan for rounding wages would affect the Wage Growth Tracker.
The following table summarizes the Census Bureau's final rounding rules for hourly and weekly wages in the CPS PUF, with the prior proposed values shown in parentheses if they differ from the final values:
As you can see in the table, the final rounding rules are less restrictive than the prior proposal released in July 2022. In particular, the Census Bureau modified its proposal by raising the upper boundaries of the rounding for hourly wages. It also updated the weekly rounding to better align with the hourly wage rounding rules, assuming a traditional 40-hour work week, along the lines I had suggested here.
The following chart shows the published Wage Growth Tracker based on unrounded data (orange line), and what it would have been if the final rounding plan had been in place (blue line).
If you have difficulty seeing any difference between the two lines, it's because they differ very little. The rounded wage data would have had very little impact on the Wage Growth Tracker statistic. I believe this outcome is a win for the collaborative process that the Census Bureau employed when developing this final plan, which included sharing information about the proposals and gathering suggestions for revision from the user community.
The Census Bureau plan includes one other change that will also directly affect the Wage Growth Tracker data. The Wage Growth Tracker excludes wage observations that have been topcoded. (Topcoding helps preserve the anonymity of the highest wage earners in the sample under study by replacing their actual wage with a topcode value.) The Census Bureau is introducing a dynamic topcode that will apply to the top 3 percent of earnings reported each month. This method will replace the current one, which applies fixed-dollar topcode thresholds to the wage data. For weekly earnings, the static threshold is currently $2,884.61 ($150,000 a year) and results in the potential exclusion of about 5.5 percent of the wage data that could have gone into the Wage Growth Tracker statistics. The new dynamic topcode will result in fewer cases being topcoded and thereby modestly expand the sample size used to compute the Wage Growth Tracker. However, if the highest wages are mostly people with relatively low wage growth (because, for example, they are late in their careers), then the calculated median wage growth could be a bit lower than it would have been. For that reason, at least initially, we plan to maintain a parallel set of Wage Growth Tracker data that continue to implement the static topcoding to see if we note any systematic differences arising from the dynamic topcode.
The changes to the CPS PUF will be implemented with the release of the January 2023 data in early February. I will report here on what we learn about the impact of the switch to dynamic topcoding, but users of the Wage Growth Tracker data can be confident that the switch to the rounding of the underlying wage data will have minimal impact.
July 13, 2022
Rounded Wage Data and the Wage Growth Tracker: An Update
In an earlier Policy Hub: Macroblog post, I noted that the US Census Bureau had announced that it planned to make changes to the Current Population Survey Public Use File (CPS PUF). Those changes, part of the Enhanced Disclosure Protection program, included the rounding of the reported wage data in a way that would have a dramatic impact on the usefulness of the Atlanta Fed's Wage Growth Tracker.
The Census Bureau subsequently revised its plans and has proposed a different rounding method described here, to be introduced in February 2023 . This Macroblog post looks at the new method's potential impact on the Wage Growth Tracker. It also considers another of the Census Bureau's other proposed changes to the CPS PUF.
So, for example, $19.99 an hour would become $20, whereas $19.95 an hour would be unchanged. Also, $999 a week would be rounded to $1,000, while $995 would be unchanged.
How much impact would this revised scheme have had on the Wage Growth Tracker if it had been used in the past? The following chart plots three versions of the Wage Growth Tracker time series. The blue line is the published Wage Growth Tracker using unrounded data. The gray line is the Tracker based on the original proposal and described in the earlier Macroblog post. The orange line is the Tracker based on the revised rounding rules. The following table summarizes the current proposed rounding rules:
The chart makes clear that the impact on the Wage Growth Tracker under the current proposed method for rounding is much smaller than the original proposal. While the revised method holds some impact, the basic time series properties of the historical Wage Growth Tracker remain largely intact. The largest difference between the Wage Growth Tracker based on the current proposal and the Tracker computed using unrounded wage data is 0.13 percentage points, the average difference is −0.002 percentage points, and the mean absolute difference is 0.03 percentage points.
No approach is perfect, though, and one quibble I have with the current proposal is that the rounding schemes for reported hourly and weekly wages are not very consistent. For example, for someone who usually works 40 hours a week (the most commonly reported workweek), rounding an hourly wage less than $20 to the nearest $0.05 should be the same as rounding a weekly wage less than $800 to the nearest $2. But the current proposal rounds a weekly wage less than $800 to the nearest $5 instead. For someone reporting a wage of between $20 and $39.99 an hour, the proposed rounding to the nearest $0.25 equates to rounding a 40-hour weekly wage between $800 and $1,599 to the nearest $10. However, the current proposal rounds a weekly wage between $800 and $1,000 to the nearest $5, and a weekly wage above $1000 to the nearest $25. Finally, for someone reporting an hourly wage of $40 or more, the proposed rounding to the nearest $0.50 equates to a 40-hour weekly wage of $1,600 or more rounded to the nearest $20. But the proposal rounds a weekly wage of $1,600 or more to the nearest $25.
The preceding analysis suggests that a more consistent method would be to round a weekly wage less than $800 to the nearest $2, a weekly wage between $800 and $1,599 to the nearest $10, and a weekly wage above $1,600 to the nearest $20.
This alternative rounding method reduces the impact on the Wage Growth Tracker series relative to the current proposal by about one-third. Specifically, the mean absolute difference between the unrounded Tracker series and the series based on the currently proposed rounding scheme is 0.03 percentage points, versus 0.02 percentage points using my alternative. The largest difference is 0.09 percentage points, and the average difference is 0.002 percentage points.
For the CPS PUF, the current proposal has another aspect relevant to the Wage Growth Tracker: the future computation of topcoded earnings data. Currently, a threshold hourly wage that varies with hours worked is used to determine if an hourly wage is topcoded. For weekly earnings the threshold is $2884.61 ($150,000 a year). However, these threshold values have not changed since 1998, and because of generally rising nominal wages over time this has led to the topcoding of more wage observations each year (see here for more discussion of this issue). The Wage Growth Tracker's calculations exclude topcoded wage values because their inclusion would be computed as zero wage change—artificially pulling median wage growth lower.
The current proposal would instead compute a dynamic topcode value that varies in a way that would result in the top-coding of only the highest 3 percent of earnings each month. Although that change means more observations to use to compute the Tracker, those observations will come from a part of the wage distribution that might exhibit quite distinct wage growth properties. For example, wage growth tends to be lower for people at the end of their careers than at the start, and if the highest wages are mostly from people with relatively low wage growth, median wage growth could be pulled lower. Unfortunately, without access to the historical wage data that are not topcoded, constructing a counterfactual to explore the impact of this proposed change is simply not possible. Perhaps someone at the Census Bureau will explore the impact this change has on the properties of the wage growth distribution.
The Census Bureau is seeking comments on the Enhanced Disclosure Protection proposals through July 15, 2022. If you have any suggestions on any aspect of the proposal, send an email to ADDP.CPS.PUF.List@census.gov. I will be sending them a copy of this post for their consideration.
July 15, 2019
Making Analysis of the Current Population Survey Easier
Speaking from experience, research projects often require many grueling hours of deciphering obtuse data dictionaries, recoding variable definitions to be consistent, and checking for data errors. Inevitably, you miss something, and you can only hope that it does not change your results when it's time to publish the results. It would be far less difficult if data sets came prebuilt with time-consistent variable definitions and a guidebook that makes the data relatively easy to use. Not only would research projects be more efficient, but also the research would be easier to replicate and extend.
To this end, we have worked closely with our friends at the Kansas City Fed's Center for the Advancement of Data and Research in Economics (CADRE) to produce what we call a harmonized variable and longitudinally matched (HVLM) data set. This particular data set uses the basic monthly Current Population Survey (CPS) data published by the U.S. Census Bureau and the Bureau of Labor Statistics. The HVLM data set underlies products such as the Atlanta Fed's Wage Growth Tracker and the various tools on the Atlanta Fed's Labor Force Participation Dynamics web page.
You may be wondering how this data set is different from the basic monthly CPS data available at IPUMS. Like the IPUMS-CPS data, the HVLM-CPS data set uses consistent variable names and includes identifiers for longitudinally linking individuals and households over time. Unlike the IPUMS-CPS data, the HVLM-CPS also has time-consistent variable definitions. For example, the top-coded values for the age variable in the IPUMS-CPS is not the same in all years, whereas the HVLM-CPS age variable is consistently coded by using the most restrictive age top-code. As another example, the number of race categories is not the same in every year in the IPUMS-CPS (having increased from 3 to 26), while the race variable in the HLVM-CPS data set is consistently coded by using the original three race categories. Applying these types of restrictions means that the resulting data set can be more readily used to make comparisons over time.
The screenshot below shows how accessible the HVLM-CPS data are. For a visual of each variable over time, click on Charts at the top to see a PDF file of time-series charts. Code Book is an Excel file containing the details of how each variable has been coded. You can see in the screenshot how each variable ends with two numbers. These two numbers correspond to the first year that variable is available. For example, mlr76 is coded with consistent values (1 = employed, 2 = unemployed and 3 = not in labor force) from 1976 until today. The Data File is a Stata (.dta) format file with variable labels already attached. For users wishing to use the panel structure of the CPS survey, lags of many variables are provided on the data set already—for example, mlr76_tm12 is an individual's labor force status from 12 months ago).
Clicking on the c icon under Code Book opens a screen with the values of the corresponding variable. The screenshot shows lfdetail94 and nlfdetail94 as examples. The first variable, lfdetail94, contains a large amount of detail on those engaged in the labor market, while nlfdetail94 contains detailed categories for those not engaged in the labor market.
The HVLM-CPS data set is freely available to download and is updated within hours of when the CPS microdata are published, thanks to sophistical coding techniques and the fast processors at the Kansas City Fed. To access the data, go to the CADRE page (using Chrome or Firefox). At the top right, select Sign in, then Google Login. Then, under schema, select Harmonized Variable and Longitudinally Matched [Atlanta Federal Reserve] (1976–Present).
- Business Cycles
- Business Inflation Expectations
- Capital and Investment
- Capital Markets
- Data Releases
- Economic conditions
- Economic Growth and Development
- Exchange Rates and the Dollar
- Fed Funds Futures
- Federal Debt and Deficits
- Federal Reserve and Monetary Policy
- Financial System
- Fiscal Policy
- Health Care
- Inflation Expectations
- Interest Rates
- Labor Markets
- Latin AmericaSouth America
- Monetary Policy
- Money Markets
- Real Estate
- Saving Capital and Investment
- Small Business
- Social Security
- This That and the Other
- Trade Deficit
- Wage Growth