Trust Government Statistics, Not Government

“Expert failure” is clearly having a moment. Pollsters, Wall Street analysts, tech futurists… all are facing demands to reckon with getting it wrong. Economics, though, seems to be getting special attention. Lately, this has metastasized into Orweillian skepticism of government data itself. It’s one thing to argue that economists have misread numbers. It’s quite another to claim that the numbers themselves are lies.

Believe me, I understand the reflex. If it’s true that the government fails at so many things it sets out to do, why trust its statistics? But this cynicism commits a category error: confusing the government’s inability to solve economic problems with its capacity to solve technical problems. Understanding this distinction explains why we can simultaneously distrust economic planning efforts while also trusting, e.g., the Bureau of Labor Statistics to provide employment figures.

Briefly, economic problems involve mutually exclusive ends and trade-offs. Should we use titanium to build railroad tracks or golf clubs? Should corn become ethanol or be used to feed cattle? Markets solve these through prices, profits, and losses. Governments, as F.A. Hayek demonstrated, are fundamentally incapable of evaluating the trade-offs involved. Technical problems, by contrast, have a singular goal in mind. Build the railroad tracks, feed the cattle, and count the total number of jobs in the US. No trade-offs are involved in these problems, it’s just a matter of execution.

Market participants can obviously solve technical problems, but so too can governments. The Soviet Union, for example, beat America to space but couldn’t stock the shelves at the grocery store. This wasn’t a coincidence. Technical problems have clear and specific endpoints. Economic problems require the evaluation of infinite trade-offs that market prices make understandable. 

Note that there is nothing here about the cost-effectiveness of the government’s solutions, nor does it suggest that solving the problem was even worthwhile to begin with. Getting to space was an impressive feat in 1961. A more impressive feat, though, is feeding your people. As it turns out, the Soviet Union did the former, but did not accomplish the latter. The result: collapse.

What does this have to do with government statistics? In a word: everything. Collecting and analyzing data is a technical problem with a clear, singular objective: accurate measurement. There are no trade-offs for e.g., the BLS to evaluate, no resource allocation problem to solve, and no need for price signals.

Consider the BLS’s track record, specifically. Unlike, to take another example, China’s National Bureau of Statistics, which answers directly to the State Council and is more accurately described as a “propaganda arm,” the BLS operates with statutory independence. The much-maligned 911,000 downward revision in total non-farm jobs growth means that the Bureau was still well over 99% accurate—there are over 150 million non-farm employees in the US right now. The 2020 Census was estimated to be off by as many as 782,000 people. With over 330 million people in the US, the Census Bureau was accurate to within 0.25%..

Does this mean that the data collected perfectly matches reality? Of course not. There are serious and legitimate debates about what should count toward GDP, how to adjust the CPI for aspects of things like changes in quality, what the threshold should be before someone is considered “unemployed,” and plenty of other measures. These debates are all about what to measure, not whether the measurements themselves are accurate or technically competent.

This distinction matters for classical liberals. We rightly distrust the government’s ability to pick winners, allocate resources, and plan economies. But dismissing government statistics as inaccurate writ large conflates technical competence with economic planning. Could the private sector collect this same data in a more efficient way? Maybe, but keep in mind that Bloomberg Terminals, which cost upwards of $24,000 per user per year, use government data.

Should we trust governments to plan economies? Absolutely not. But should we trust government statistics, at least those in the US? The evidence suggests that we should. We should trust them not because governments are virtuous, but because measurement is fundamentally different from deciding what to do with those measurements.