Unemployment Number "Conspiracy" Unraveled

By Ryan Snefsky

The other day, a friend of mine asked me on Facebook, "Why are the jobless numbers always 'unexpectedly higher' and [seemingly] every month 'revised upward from the previous report?' Is the algorithm they use wrong or do they just lie to the public to keep confidence up knowing most people won't remember 'revised upward' the next month?"

This is a great question. When the numbers are "revised" a month or two down the road in a way that shows the economy isn't doing as well as the Government originally reported, it's easy to become suspicious of an unemployment number "conspiracy."

While there is temptation to suspect a possible conspiracy, the truth of the matter is that these suspicions are caused by yet a different problem: the Government's reporting of nationwide estimates based on incomplete data. It's basically like calling an election with a specific total of votes before all the votes are counted.

In the case of unemployment data, the collection period of the data is different for each survey used to produce the data. That's why some of the data is seemingly revised more than others. Let's look at a breakdown below.

The Employment Situation Monthly Report

In the first week of every month, the United States Department of Labor's Bureau of Labor Statistics ("the Bureau") publishes a monthly Employment Situation Report. This report summarizes data from two different surveys:

1. The Household Survey (aka the Current Population Survey) – Each month, the Bureau has a staff that surveys 60,000 households, representing approximately 110,000 individuals. This 60,000 household sample is designed to represent the entire U.S. population. As each surveyor completes a survey and enters the data into a laptop, the data is automatically sent to the Bureau in Washington, D.C. and compiled.

This survey is completed in one month and tallied in something close to real-time, so the statistics produced by this report do not usually get revised.

The statistics produced from this survey are estimates that are projected from the 60,000 household sample and are believed to be not less than 90% accurate.

Some of the key statistics produced from this survey include the unemployment rate, a breakdown of unemployed persons by race and gender, the number of long-term unemployed (persons considered unemployed for 27 or more weeks), and the number of "discouraged" workers (more on the strange implications of discouraged workers later.)

For answers to Frequently Asked Questions about this survey, click here.

2. The Establishment Survey (aka the Current Employment Statistics or CES Survey) – This is a voluntary survey of approximately 140,000 businesses and government agencies representing approximately 410,000 worksites throughout the United States. The survey data is used to project estimates of key statistics relating to the nation's unemployment data.

Data from this survey is sent to the bureau directly by participants via telephone, fax, mail, email, and the World Wide Web. Because participants send survey data in on a voluntary basis, rather than it being collected by interviews conducted by the Bureau's staff, survey data for a particular survey period trickles in a month or more past the initial publication of the estimates.

The Bureau publishes up to two revisions of the estimates based on surveys that arrived after the initial publication of estimates. The first revision is published when the next month's estimates are initially published. The second revision is published a month after the first revision when yet another month's estimates are initially published.

Now you can see why I make the analogy to calling an election with specific estimated vote counts before all of the votes are counted. Why doesn't the bureau just wait to publish the estimates until all of the surveys are submitted? I don't know. But as long as this is the way they publish the data from this survey, it is quite reasonable to expect revisions.

Some of the key and most widely followed statistics produced in this survey include nonfarm payrolls, nonfarm private payrolls, hourly earnings, and average number of hours worked in a work week.

For answers to Frequently Asked Questions about this survey, click here.

The Unemployment Insurance Weekly Claims Report

Every Thursday morning, the United States Department of Labor’s Employment and Training Administration (“the Administration”) publishes the Unemployment Insurance Weekly Claims Report.

Among other data, the report publishes an advance (estimated) figure of initial unemployment claims for the prior week and a revised (more complete) figure that reflects initial claims made two weeks prior. The data used to create this report is provided by individual states and is generally revised not more than once.

Other Factors That Can Create Perceptions of a Conspiracy

After taking a look at where different pieces of unemployment data come from and how they are collected, you can more easily understand why data revisions occur. But what other factors help create the perception that something more sinister is going on?

1. Media headlines are not necessarily complete – When unemployment data makes its way into the media, the headlines are sometimes incomplete. Let's look at some real examples. On June 3rd, 2011, the following headlines were produced:

"Jobs Jolt: U.S. Payrolls Add Just 54K In May" – Forbes

"U.S. Nonfarm Payrolls Increase by 54K; Safe Haven Currencies Rally" – Daily FX

"Job growth slows to 54,000, rate up to 9.1%" – Market Watch

All three of the above headlines suggest that 54,000 nonfarm payrolls were added. However, none of them suggest the fact that this number is an estimate, based on incomplete data, and is possibly going to be revised up to two times as more data becomes available.

The portion of the Market Watch headline that says, "rate up to 9.1%," is complete in that the published unemployment number does not usually get revised as we learned above.

When people read these headlines, it is easy to see why they would take them as certain facts that are not going to change, unless they know specifically how each survey collects data.

If the Government then revises the nonfarm payroll number down the road, people can incorrectly conclude, "Well, they got the number wrong, so there was obviously either an error that needed corrected or they wanted to hide the real number until later."

2. The number of "discouraged workers" can cause the published unemployment rate to paint a rosier picture of unemployment than what exists in reality – The Market Watch headline "Job growth slows to 54,000, rate up to 9.1%" seems awfully suspicious without an understanding of how discouraged workers influence the published unemployment rate. Think about it. If the number of jobs is increasing, how in the world can the unemployment rate be increasing at the same time?

The published unemployment rate is basically estimated by dividing the number of unemployed persons in the household survey by the labor force. Discouraged workers, which the bureau defines as individuals "not currently looking for work specifically because they believed no jobs were available for them or there were none for which they would qualify," are not, however, included in the labor force.

Simply stated, unemployed individuals that "give up" looking for a job are neither counted as being part of the labor force nor counted as being unemployed as far as the published unemployment rate is concerned.

As the economy picks up steam and jobs start to become available again, individuals that had previously given up looking for a job can start to become encouraged and start job searching again. As their search begins, they are again counted in the labor force number.

When the labor force number increases proportionally faster than the amount of new jobs, the unemployment number will increase at the same time the number of jobs increases.

When the seasonally adjusted unemployment rate peaked in October of 2010 at 10.1%, many economists estimated that the real unemployment rate was somewhere between 13% and 17%. Those estimates included persons not normally factored into the published unemployment rate, such as discouraged workers.

What can we conclude?

To put things in terms of my friend's original question, we can now see that revisions to the jobless numbers occur because of the government's release of incomplete estimates, rather than changes to data that were considered to be complete and accurate in the first place.

Is it possible that the Government is conspiring to lie to the public in an effort to keep confidence high for as long as possible? Maybe. However, I do not see any evidence suggesting that such a conspiracy is occurring. Even if we knew as an absolute certainty that there was no conspiracy, that does not change the fact that revisions to some of the data are likely to happen unless the Government waits to release their estimates until all survey data becomes available.