Monday, November 4, 2013

An Actual Data "Manipulation"

As a rule, all accusations of government economic data manipulation made in a political setting are without basis.  Discussions of the CPI, or manipulated unemployment numbers, or Federal Reserve Bank data are politically motivated and have never panned out (and are never followed up on as time goes on).

In September 2013, the BLS will begin to incorporate a large non-economic code change and "artificially" increase employment.  Specifically, 469,000 people currently employed in private households (such as housekeepers and gardeners) will begin to be included as employees. This change will be "wedged" back into past estimates come the February 2014 report of January 2014 employment data.

This is an improvement in data, but it is an artificial inflation of job numbers compared to past history.  No doubt it will be used by some to hyperbolically account for the 7 million job increase over the last 3 and a half years, but this will be in error.


  1. You don't get it. The scientific data of economics are the self published numbers of government agencies. There is no peer review or enforced verification of any kind. Therefor even if the government published openly fraudulent statistics, there would be no scientific proof of this because the data is whatever the state vanity press says it is.

  2. The sort of thinking that anonymous has put forward seems all too common. It is in error. First, not all economic data comes from the government: consider the ADP employment report: self-reported numbers of large payroll processors as noisy signals of the same employment data the government generates. While it does not track government data completely, it is an unbiased predictor. Or consider the Case-Shiller price index: produced by a private company. Or consider the Billion Prices project, which coincidentally matches up with the CPI to a respectable degree. Similarly with stock market prices.

    Not only do private corporations often replicate governmental efforts, but so too do states. States as well as cities keep track of their local unemployment claims and local employment. We have never seen a report that diverges enough to cause interest.

    There is one data source that diverges enough to be of interest: the BLS has four different employment surveys: the CES, in which it asks employers how many employees they have, the CPS and the ACS, in which they ask households, and the QCEW, a mixture of Federal and State. Interestingly,the CPS and the CES have diverged a little over the last few years, a cause for study of many smart people.

    For all of these to agree so remarkably would be a feat of strength of even the quality statisticians working for the BLS and BEA. Remember: they produce microdata that must match up across dozens of different surveys, private and public, and that means all the conditional statistics must match as well.

    It is the privilege of people who do not work with the data to misunderstand how difficult it would be to fake or manipulate. Ignoring the conditional statistics, private data, and multiple separate branches of government largely agreeing with one another is an error.

    Conspiracy theorists that make unverifiable and untestable claims that various fields are unscientific are the most painful of all the public claims any scientist must hear. Can you make a testable statement based on your fraudulent government data?

    Our test would be the count of the number of government statisticians, many of whom are extremely competent Ph.D.'s who could get a job anywhere, who come forward and attack the data. Corrections can list off a few who will write papers concerned about this aspect or the data or that, but none who have ever claimed fraud. It is a telling Bayesian update that it is only those conspiracy theorists who are politically motivated that do so.