A financial trader reacts as he monitors data on computer screens at the Frankfurt Stock Exchange in Frankfurt, Germany (Photographer: Martin Leissl/Bloomberg)  

How Econ Numbers Can Lead You (and Me) Astray

(Bloomberg Opinion) -- The other day, in a column discussing the short-term impacts of President Donald Trump’s tax cut, I showed data indicating that the tax cut hadn’t yet led to any appreciable rise in wages. One of the things I used was a graph from the website PayScale, which collects online survey data about compensation from millions of its users. PayScale’s data showed real wages dropping precipitously in the second quarter of 2018 -- an annualized rate of almost 7 percent. Social media being what it is, the PayScale wage graph was rapidly turned into a widely shared meme. Claims that real wages were falling off a cliff as a result of Trump’s tax policies proliferated.

That probably isn’t true. Official measures of real wages show little to no change in the second quarter. The “wages and salaries” portion of the Employment Cost Index, released by the Bureau of Labor Statistics 13 days after my piece, showed nominal wages increasing by 0.5 percent in the second quarter of 2018. That’s exactly the same as the rate of inflation for that quarter, indicating that real wages neither rose nor fell. The BLS reports that when benefits are included, total real compensation actually rose by 0.1 percent in the second quarter -- an anemic rise to be sure, but nothing like the plunge reported by PayScale.

So why does PayScale show such a big drop? For one thing, its data is very volatile. PayScale reported large swings in real wage growth in 2013 and 2014, when the official government data only moved within a narrow range. Additionally, PayScale’s numbers have a long-term negative trend -- the website reports real wages having decreased since 2006, while official data show real average hourly earnings increasing by about 7 percent during that period.

In a recent post, PayScale defended its data, saying that it measures things the government numbers fail to capture. The website measures somewhat different regions and types of cash compensation than the government does. PayScale also contends that its data is quicker to account for the creation of new types of jobs, and with shifting shares of employment within occupational categories.

PayScale has a point. Although its data are much more volatile and probably contain many more statistical biases than the government’s, the two also are different; PayScale probably captures at least a bit of information that the official numbers miss. It’s best not to take the PayScale data too seriously -- and I probably shouldn’t have put it in graphical form, since that gave it an imprimatur of rigor and numerical precision that it doesn’t merit. Instead, it should be regarded as simply one more indicator that might contribute to a general picture of the direction wages are headed.

This episode shows why short-term macroeconomic data interpretation is so difficult, and so perilous. Every time a new macro data point comes out, commentators rush to interpret it. The monthly jobs and unemployment numbers, and the quarterly economic growth numbers, are other examples. Using these data points to do on-the-spot policy analysis is fraught with dangers.

For one thing, while government economic data are the best we have, they’re not perfect -- they depend on surveys that may contain biases, noise, and incorrect assumptions. Government statisticians are always working on improving their data collection and definitions, and they should be supported with greater resources. But there will never be one true, perfect measure of economic growth, or wages, or unemployment.

Second, government data are subject to revisions, sometimes years after they come out. Over the long term -- years or decades -- these revisions make little difference. But they can entirely invalidate the stories that commentators tell based on monthly or quarterly data points.

Third, a few short-term data points are often inadequate for analyzing the effects of an economic policy like the tax cuts. Sometimes an economic policy can have a dramatic and rapid effect, but sometimes the effect takes place slowly, over the course of several years. In the case of Trump’s tax cut, it might take businesses several quarters to start handing out raises in response. Also, at short time scales, swings in real wages or GDP are heavily affected by swings in gasoline prices that have little to do with policy.

Were it not for the advent of Trump’s trade war -- whose impact could swamp that of the tax cuts -- it would make sense to wait several more quarters before analyzing the tax cuts’ results at all. But if you’re going to do it, as I attempted to do, it seems best to look at as many different data sources as possible, while not putting much weight on any single number. And it’s best to draw only the most tentative of conclusions from the tea leaves.

This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.

Noah Smith is a Bloomberg Opinion columnist. He was an assistant professor of finance at Stony Brook University, and he blogs at Noahpinion.

©2018 Bloomberg L.P.