Confirmation Bias, Illustrated

Scott Sumner writes,

And I just want to make sure readers are not getting lost in the weeds here. This is not one of those “he said, she said” where reasonable people can disagree on whether the PCE or CPI is a better price index. This is a pay/productivity gap being invented by using the slowing moving price index (NDP, which is similar to the PCE) to make worker productivity look better, and the faster moving price index (CPI) to make real wages look lower. That’s not kosher. You need to use the same type of index for both lines on the graph.

Brad DeLong writes,

I score this for Larry Mishel….

Pointer from Mark Thoma.
DeLong and Sumner are on opposite sides, and each is certain.

Fundamentally, I think that the reason that this happens is that economic propositions are not falsifiable. They only hold “other things equal,” and other things are never equal. A measure of worker productivity that is reasonable according to one person’s framework is not reasonable according to another person’s framework.

If you continue to insist that economics is a science in spite of the non-falsifiability of propositions, you end up deciding that those who disagree with you are evil and anti-science. As I say in the Book of Arnold, you end up wallowing in confirmation bias.

On this particular, by the way, my inclination is to agree with Sumner. That may be political bias on my part. But also, I think you would be seeing a lot of other dramatic things happening if productivity were outstripping wage growth. Very high demand for labor. A big improvement in international competitiveness, leading to large trade surpluses. etc. At the minimum, it seems to me that Mishel and DeLong owe us some comment on why these other developments do not seem to have taken place.

Contrary to what you see in online debates “this one chart” is never a debate-settler. You don’t make a convincing case with one chart. You need lots of disparate, corroborating evidence.

10 thoughts on “Confirmation Bias, Illustrated

  1. We can compare productivity and real labor income back to 1948.

    From 1948 to 1975 the two moved in lock step but after 1974 the two diverged significantly.

    I would think that for either argument to win it should have to explain why and how the divergence emerged.

    The PCE deflator does not include shelter — almost a third of the CPI — and this accounts for about two-thirds of the difference between the CPI and PCE deflator.
    Shelter is counted elsewhere in the GDP accounts so it is included in the gross business output deflator used in calculating productivity.

    If you are measuring standards of living I would think you need to account for shelter.

    • The two series are deflated using different inflation measures (CPI-U for wages, Implicit Price Deflator for productivity). You can think of it as the price indices for what workers buy vs. what they make. CPI-U rose 2.6%/yr from 1947Q1 to 1973Q1 whereas IPD rose 2.8%/yr. So even with flat “real” productivity growth, workers could buy 0.2% more each year based on improving terms of trade. From 1973Q1 to 2015Q1, CPI grew 4.1%/yr but IPD grew just 3.5%/yr, meaning that if productivity didn’t grow, workers would be able to buy 0.6% less each year on average during this time period. To give a better sense of the impact, the ratio of IPD/CPI was 0.571 in 1947Q1, 0.591 in 1973Q1 and 0.463 in 2015Q1.

      The ratios of labor compensation and net domestic product (net, since depreciation has grown faster than GDP and you can’t eat depreciation) for the three dates are as follows: 60.0% for 1947Q1, 65.1% for 1973Q1, and 63.2% for 2015Q1. Since 1973 was a business cycle peak and 2015 is mid-cycle, here are two additional dates for more context: 1965Q1 was 62.1% and 2007Q1 was 65.2%.

      In short, labor compensation tracks productivity very well.

  2. Productivity is a measure of production. Wages are a measure of both production and consumption, but mostly important for consumption, so this is fitting. Compare Yglesias for example, who removes both to show a large fall. Labor demand is a measure of production, not productivity. It is irrelevant to productivity, but it is relevant to wages. An increase in international competitiveness would show up in labor demand and wages as well as productivity. Profits at all time highs, interest rates at bottom, slack labor demand, tells a different story.

    • This is incorrect. Productivity zoomed in the late 90’s and never really has come back down to the pre-trend.

      I am not seeing the labor slack at all, if anything, that slack is all about gone.

  3. My interpretation of the data, informed by several papers I link to at the end:
    1. Output per worker going up due to reduced costs of automation and outsourcing of labor intensive jobs. This is also consistent with the fact that inflation in nontradables and services has been well above that of tradable goods. A related, more Kling like, interpretation is that manufacturing is more of a market than education and healthcare, which are controlled by elite insiders who set their own prices and use rationed elite degrees to maintain the cartel.

    2. Since in an ideal world wage=marginal productivity, and there is no reason why output/worker increases imply marginal productivity increases when automation and outsourcing are the likely candidates. This explains the “pay/productivity” gap, even though the marginal product of labor has not grown with output.

    3. Since the decreasing costs of automation and outsourcing mean that the cost of a given level of corporate investment has gone down, rich people (whose savings rate is very high and who own a very disproportionate share ) have been in a “search for yield” and therefore are willing to lend to poor people (and also have been less constrained by financial regulation) to finance consumption at a much more attractive interest rate than previously available. The Chinese have also been “recycling sweatdolllars” into financing American consumption for possibly similar reasons. This explains why consumption has kept up with output even while wage income has note.
    4. The result of this is the bottom 90% of Americans have been living well beyond their means, financed primarily by rich Americans and Chinese who have nothing better to do with their money. This resulted in a falling real interest rate and eventually a financial crisis when they were not able to pay back their debts. The fact that more and more debt was needed to finance a given consumption/output ratio meant that this financial fragility could build up without a meaningful increase in inflation.

    Ignoring the business cycle/financial crisis aspects of this, this is fairly similar to what is sometimes called “Engel’s Pause” in the early industrial revolution, where wages did not move with output per worker (which is not the same as the marginal product of labor) for several decades until labor became scarce relative to capital. Eventually, there will be so many robots making goodies per person, and wages in China will be so high, that American labor income will get moving again.

    relevant papers that have informed my view on these issues:
    http://faculty.chicagobooth.edu/loukas.karabarbounis/research/labor_share.pdf
    http://economics.mit.edu/files/6613
    http://economics.mit.edu/files/5571
    http://gabriel-zucman.eu/files/SaezZucman2015.pdf
    http://www.economics.ox.ac.uk/materials/working_papers/paper315.pdf
    https://www.aeaweb.org/articles.php?doi=10.1257/aer.101.5.2132
    http://www.nber.org/papers/w20935
    https://ideas.repec.org/p/fip/fedfwp/2014-23.html

  4. Its funny that basically all the comments are about one side verses the other with some modest amounts of evidence, and this post is about how you shouldn’t be sure of your position on an economic question without large amounts of evidence.

    • I don’t really see the meta gotcha, but I’ll raise you a meta meta gotcha. Who is the Arnold Kling of the other side?

    • I disagree about confirmation bias.

      I am a conservative and used to accept the notion that wages and productivity had decoupled based on the popular charts comparing the two. After diving into the data, it’s clear that the traditional position of the economics profession holds – productivity drives income and labor receives a relatively constant share of income over the long haul, and you can get at this conclusion from multiple directions. In fact, the main breakdown is that labor’s share of income actually increased somewhat from the 1945-1965 period to the 1966-2014 period, with only the slightest negative trend from the late 60s until today.

      Compensation of employees was a higher share of net domestic product than in 1966 than *every single year* after 1966. The 2001-2014 average for this ratio was higher than the 1955-1973 average. Unless you want to make the claim that 1966 was a particularly bad year for American labor, or that 1955-1973 taken as a whole was a bad period for American labor, the idea that labor compensation isn’t keeping up with overall economic growth is simply false.

      One might not like the distribution of compensation among workers, but that’s a different issue, and differences in productivity growth by sector seems to be a driving factor.

      Again, I fail to see how this is confirmation bias. The theoretical economic result and a close look at the empirical data agree, and there is simply no convincing counterargument. I’m fully happy to be shown where my analysis has gone astray. If I argue from a position in error, then my argument is unconvincing. It doesn’t matter to me one way or another whether or not labor compensation tracks overall production growth except for the fact that it’s indisputably true given the evidence we have.

      http://bit.ly/1Oi2owD

Comments are closed.