Quartz recently reported that the $33 billion deal made at the Africa Summit last week might not go as far as its investors would like, largely due to the “terrible state of statistical reporting in most of Africa.” The article cites Nigeria’s sudden leap in GDP, which allowed the country to replace South Africa as the continent’s largest economy, as evidence of “sketchy” data reporting. A recent working paper from the Center for Global Development explains the two incentive structures that may cause aid recipients to report data incorrectly. Front-line service providers sometimes overstate their progress to governments, as in the case of primary education programs funded with per-pupil grants. Similarly, governments may themselves also misrepresent findings to foreign donors, for example when continued funding is proportional to reported vaccination rates.
Particularly during budget negotiations, suspicion of waste and corruption in foreign assistance programs is a popular theme in the United States, but this country has long had a broader problem with statistical error reporting that may affect all departments and agencies. In a draft of his upcoming paper, Professor Charles Manski addresses the lack of progress in agencies’ statistical reporting accuracy since the incendiary 1963 work by Oskar Morgenstern, the economist most remembered for contributing to the development of game theory. Manski demonstrates that federal data is overwhelmingly represented by point estimates, while nonsampling error remains unquantified in federal employment and income data, significantly reducing its utility to researchers and interested citizens. A paper in the Journal of Official Statistics agrees: “data quality information that users need is not always made available and is not consistently reported across agencies.”
The possibility of error does not necessarily go unmentioned: Manski identifies several examples of acknowledgment, including addendums to an employment report from the Bureau of Labor Statistics (BLS), and to the Census Bureau’s annual Current Population Report. Critically, the magnitudes of error are not provided.
Economists have established statistical principles for dealing with sampling error, but there is currently no universal standard for measuring the various forms of nonsampling error. Manski explains that representing unemployment, for instance, as a quarterly point estimate allows policy analysts to make conclusions with “incredible certitude” where it simply does not exist. When agencies continue to omit magnitudes of error, some users “may naively assume that errors are small and inconsequential,” while those who “understand that statistics are subject to error must fend for themselves and conjecture the error magnitudes.” Echoing the example of Nigeria’s artificial GDP explosion, Manski writes that failure to report error in this country could allow a “central bank monitoring statistics on GDP growth, inflation, and employment [to] misevaluate the status of the economy and consequently set inappropriate monetary policy.”
It appears that not only can the government not play moneyball due to a scarcity of data, but that those who try may be misinformed by the data that does exist. Morgenstern may have said it best more than fifty years ago, when he called for the President's Council of Economic Advisors, Federal Reserve Board, and other public and private agencies to stop presenting to the public “economic statistics as if these were free from fault...it is for the economists to reject and criticize such statements which are devoid of all scientific value, but it is even more important for them not to participate in their fabrication."