(Cross posted from Pine Treeconomics)
THE recent flap over the Michaud bogus jobs claim has exposed both some questions about the motives of the campaign in making the claim, and, more importantly, a misunderstanding about the jobs data. The former will likely blow over, and the latter, sadly, will likely never get noticed despite being the more important issue. Jobs data is crucial to any comprehensive economic policy discussion, and as a result the public should demand accuracy both in that data itself but in the interpretation of it by policymakers. Without proper interpretation and use of jobs data, bad policies will result. It’s disconcerting to see so much confusion about the data, particularly when policy discourse is largely dominated by the economy, jobs, wages, etc.
Numerous articles about the Michaud jobs claim highlight the confusion about the data at issue, and it is my hope to resolve the confusion in this post. First, I am going to explain what data the Michaud campaign relied on and where the campaign’s claim went awry. Next, I will examine some of the critiques of the Michaud campaign’s claim, specifically looking at the data used in those critiques. Subsequently I will explain the differences in the jobs data sets used to highlight the confusion found in those critiques.
So, here’s where it went all wrong for the Michaud campaign. As reported by Steve Robinson at the Maine Wire, Mike Michaud put out an ad in which me made this somewhat ambiguous claim:
If you look at the Maine being one of the three states that’s actually lost jobs, uh, over the last year, the uncertainty, uh, is real.
I say ambiguous because he does not cite a source or give any specifics of his claim, but the likely genesis of the Michaud claims comes from this study by Pew, released on June 14 of this year, that examine data released by the Bureau of Labor Statistics. According to the data as of June of this year, from April, 2012, to April, 2013, Maine was one of three states to lose jobs. This is the infamous graphic associated with that study:
The above graphic is confirmed by the release of the BLS April jobs data, published on May 17th of this year. Here is a graphic of the relevant portion of that release which shows that in April, 2012, Maine had 598,900 payroll jobs, and a year later in April, 2013 Maine had 597,400:
This comports with the 1,500 jobs lost the Pew graphic shows, as wells as a May 21 press release from the BLS stating the same statistics as the pew study.
Now, when data is initially released by the BLS, it is typically referred to as preliminary data as that data is subject to subsequent revisions. From the BLS website:
Preliminary-to-Final Estimates
Initial monthly estimates are calculated from an incomplete sample and are subject to revision in the subsequent month when more sample data are available. Revisions at the total nonfarm levels for preliminary statewide employment are generally small. For more info on preliminary-to-final revisions, please visit www.bls.gov/sae/saerevtxt.htm.
Final-to-Benchmark Estimates
“Final” estimates are subject to annual benchmarks of universe counts of employment derived from the unemployment insurance (UI) reports from employers. The average absolute benchmark revision at the state total nonfarm level was 0.7 percent in March 2012. The average absolute revision from 2007 to 2011 was 0.5 percent. The range of the percentage revision for the States at the total nonfarm level was from -1.5 to 2.2 percent in March 2012. (The direction of the revisions indicates whether the March 2012 benchmarks are greater or less than the sample-based estimates.)
Without delving into the technicalities of data revisions, the point is that when data is initially released, such data is met and interpreted with skepticism because of the likelihood that it will be revised either up or down. And this is what happened with the Michaud campaign’s use of the Pew graphic; it was premised on preliminary data that was subsequently revised up. As I noted in a previous post:
Assuming that the Michaud claim is based on this report, which the Maine media picked up on, then Robinson is right, Michaud is wrong, the numbers were revised, and that depends. When the BLS released its April jobs data (see Table 5), it showed Maine lost 1,500 jobs, a 0.3% decline, from 598,900 to 597,400. However, that latter number has since been revised up to 599,100, meaning Maine gained 200 jobs–weak, but a change of +1,700 from the original data nonetheless.
And here is the updated data in picture form:
The campaign’s use of erroneous data resulted in several criticisms for that claim, the political substance of which I have no interest delving into. What I am interested in is the data they use in their critiques. First, there was this post from Phil Harriman:
After hearing Eliot Cutler and Mike Michaud proclaim that Maine has lost jobs since Gov. Paul LePage took office, I did some relatively easy research and proved their claims are false. A Pew Charitable Trusts research study released in June that shows Maine was one of three states to lose jobs in the last year is built upon jobs data that was later corrected.
According to the Maine Department of Labor, here are Maine’s actual jobs data:
July 2012 — 653,600 employed
July 2013 — 660,300 employed
Since LePage took office, more than 10,000 new jobs have been created.
And this recent one from Matt Gagnon:
The labor statistics, you see, were actually revised as part of the state’s regular processing of updating economic data, and showed that Maine had in fact gained jobs under LePage.
Indeed, according to the Maine Department of Labor, the actual number of employed workers went from 653,600 in July of 2012, to 660,300 in July of 2013.
In fact, more than 10,000 new jobs have been created during LePage’s tenure in office.
And from Steve Robinson:
Mistler rightly warns against confusing BLS numbers with those of MDOL. While the corrected federal numbers report a year-over-year increase of just 200 jobs for the most recent data available, closer to home, MDOL pegs the increase at 2,900 jobs. The estimates are produced through two different methods, but both point toward the same conclusion: Michaud’s claim is a folksy bowl of malarkey.
See the problem? No? Okay. The problem is not, as Robinson suggests, conflating data between the Maine Department of Labor and BLS; the problem is conflating two sets of data derived from two different surveys. The MDOL puts out jobs data through the Center for Workforce Research and Information, which:
The Center for Workforce Research and Information (formerly Labor Market Information Services) develops and disseminates state and area labor market information to employers, job seekers, and other users; provides measurements of labor market outcomes to assist local and state officials, employers, educators, trainers, and the public in making decisions that promote economic opportunity and efficient use of state labor resources; and supports the Department with management and actuarial analyses for program planning and delivery.
The CWRI, like all other state level entities, works in cooperation with the BLS to publish state and regional employment data. From the CWRI’s website:
Data is derived from the Current Employment Statistics (CES) program. CES is a federal-state cooperative statistical program to estimate current employment in a state or sub-state area. A statistically valid sample of employers is surveyed to supply total number of employees, total number of female employees, total number of production workers, total number of hours worked, and total wages earned. This survey is the basis of current estimates of these characteristics. It is used in the calculation of the monthly estimates of employed, unemployed, and the unemployment rate.
And:
Data is derived from the Local Area Unemployment Statistics (LAUS) program. LAUS is a Federal-State cooperative program that develops monthly estimates of the labor force, employment, unemployment, and unemployment rates for the state, counties, labor market areas, metropolitan and micropolitan statistical areas, and cities and towns (minor civil divisions).
So problem is not that data varies depending on whether the source comes from a state level entity or from the BLS, a federal level entity. However, you probably noticed that in the two above paragraphs mention two different data programs; the CES and the LAUS. The CES is a type of survey employed by the BLS, known as the establishment survey. The latter is a program that uses a different type of survey, the Current Population Situation (CPS), also known as the household survey. The following are brief descriptions of the two types of surveys from a 2006 Labor Review article published by the BLS:
With a sample of about 60,000 eligible households, the CPS covers the U.S. civilian noninstitutional population aged 16 years and older. The survey results are collected for the Bureau of Labor Statistics by the U.S. Census Bureau. Interviewers from the Census Bureau contact households and ask questions regarding the labor force status of members of the household during the calendar week that includes the 12th day of the month. The broad coverage of the CPS encompasses not only wage and salary workers, but also the self-employed, farm workers, unpaid family workers, persons employed by private households, and workers temporarily absent from work without pay. The CPS provides information on labor force status by detailed demographic characteristics, such as sex, age, race, and ethnicity. Data are collected not only on the employed, but also on the unemployed and on those persons who are not in the labor force.
The CES survey is a Federal-State cooperative program in which the Bureau of Labor Statistics works with State employment security agencies to collect data each month on employment, hours, and earnings from a sample of nonfarm establishments, including government. The CES survey sample, which includes about 160,000 U.S. firms of all sizes, covers about 400,000 worksites. The survey counts nonfarm payroll jobs only—with no age restriction on the employed—and does not include the self-employed. Businesses report the number of persons on their payrolls who received pay during the pay period that includes the 12th day of the month. Workers who did notreceive pay during the pay period are not counted. CES data are available by detailed industry and geographic area. In addition to collecting data on employment, the CES program provides data on average hours paid and average hourly and weekly earnings for private-sector production or nonsupervisory workers.
The article also provides this quick reference chart highlighting the differences between the CPS and CES:
Perhaps the most important distinction between the two surveys as far as type of data collected is that the CPS counts people (the number of people employed), while the CES counts payrolls (the number of jobs). To understand this subtle but important distinction, think of an economy with two people but with three jobs, where one person works two jobs, and the other works one job. The CPS counts each jobholder, meaning its final tally would be two, whereas the CES counts payroll jobs, and would have a final tally of three. This distinction becomes more and more important the larger and more complex an economy becomes where more jobholders are taking two or more jobs. Assume that an economy has 100 workers, and 150 jobs, but expands to 200 jobs but remains at 100 workers. Under the CPS, employment remains the same, but under the CES, 50 jobs were created. In other words, depending on the survey used, one could claim either that the labor market improved or stayed the same.*
There are other important differences between the two surveys** and the data sets they create, but for now it’s important to understand the basic difference between the two; the CPS counts persons employed, the CES counts payroll jobs. So, when speaking of “job growth”, one is typically referring to the CES data which measures actual gains in payroll jobs. Conversely, when speaking of employment or unemployment, one is typically referring to CPS data which measures persons employed and unemployed.
Now, back to the Pew study and the Michaud critics. As noted above, the Pew study found that the BLS data initially reported Maine lost 1,500 jobs from April, 2012 to April, 2013. This comports with the table provided above, and reproduced here:
That table is from the April jobs report’s CES data set, meaning Pew used the CES (establishment data), The CES data is produced on both the BLS and MDOL websites. The first graphic is from the BLS, the second is from the CWRI website:
As you can see from the box in the lower left of the graphic from the CWRI (the second graphic), the numbers produced by the BLS and MDOL websites are the same data, so the problem is not that BLS and MDOL use different methods.
The numbers in the above graphics, remember, are nonfarm payroll numbers that are collected from the CES, and that data shows that Maine’s employment for July, 2013 was 603,100, up from 597,200 12 months prior, and the year-over-year change from April, 2012 to April, 2013 at 200. However, both Harriman and Gagnon state that as of July 2013, Maine employment stood at 660,300, up from 653,200 12 months prior. So why the difference in numbers, and what about Robinson’s claim that Maine’s employment from April, 2012 to April, 2013 increased 2,900 and not 200?
Remember, there are two sets of data produced by the BLS and MDOL–the CES (establishment data that measures payroll jobs) and the CPS (household data that measures employed persons). Below are graphics from the BLS and CWRI showing the CPS data for Maine (data circled in red is related to the Harriman and Gagnon numbers, data circled in green is related to Robinson’s claim):
As you can see, the CPS numbers comport with the totals given by Harriman and Gagnon, and the April, 2012 to April, 2013 gains mentioned by Robinson (657,854 – 654,988 = 2,866, or roughly 2,900). Moreover, both Harriman and Gagnon correctly use the term “employed” when mentioning the July, 2012 and July, 2013 numbers because those numbers come from the CPS which, again, measures employed persons. However, they both claim that Maine has gained 10,000 “jobs” from the start of LePage’s Administration. Jobs are measured by the CES which shows that Maine payrolls numbered 593,200 in January, 2011, and the preliminary August, 2013 number is 602,600–a gain of 9,400 jobs (of course, we will have to wait until after the revisions are made next month to know the official number). Yes, using the CPS data would show that Maine employment grew by 12,056 (from 647,003 to 659,059), but they did not refer to employment growth during the current Administration, but specifically mentioned job growth. As a result, their claim of Maine job growth during the LePage Administration of 10,000+ is false.
Yes, the Michaud campaign should have more carefully looked at the data (which had been out well before the ad was made), but some of the criticisms of the campaign highlight the somewhat cryptic nature of jobs data.
* Using one metric is a minimalist way of examining the labor market. Both the CES and CPS and all data produced therefrom, as well as other metrics, are extremely important in order to properly evaluate the labor market. Of course, pundits, members of the press, and others often like to have succinct numbers at their disposal, and so often cite job creation, unemployment, or other similar data because it is concise. In those instances, however, it is important to use the appropriate data.
** Namely that the CPS is subject to more monthly volatility than the CES.