In a book of this size (over 400 pages) one would like to see a chapter tracing the history of economists’ attempts to measure the effects of minimum wages. Instead, the book contains only casual and sometimes misleading references. For example,

[Here he quotes Card and Krueger] The idea of using natural experiments is hardly new in economics. Indeed, the earliest research, by Richard Lester (1946) and others, used that approach. Nevertheless, it is controversial-perhaps because studies based on the natural experiment approach often seem to overturn the “conventional wisdom.” (p. 21)

This begs [sic] the question of how the wisdom became conventional in the first place. Moreover, the contribution of Lester (to whom the book is dedicated) is overstated. A more representative list of natural experiment studies would include work by Marie L. Obenauer and Bertha von der Nienburg (1915), John F. Maloney (1942), and John M. Peterson (1957, 1959), and these studies did not overturn the conventional wisdom (perhaps that is why Card and Krueger did not mention them).

This is from John Kennan, “The Elusive Effects of Minimum Wages,” Journal of Economic Literature, Vol. 33 (December 1995), pp. 1949-1965. It’s a comprehensive review of David Card and Alan Krueger’s famous 1995 book, Myth and Measurement: The New Economics of the Minimum Wage. The book made a real splash in the mid-1990s. My favorite review of Card and Krueger had been my own; now I think I’ve found a contender for first place.

Yesterday I posted on a 1915 study of the effect of minimum wages on women in Oregon. Labor economist Jonathan Meer, of Texas A&M University, contacted me and made me aware of this review by Kennan that actually digs into the 1915 study.

Here’s part of Kennan’s review of the 1915 study:

Obenauer and Nienburg pointed out that the overall decline in employment was due to a general recession: for example, total sales in these Portland stores fell by 8.6 percent over this period. Moreover, they noted that the jobs held by men were less vulnerable to this decline than women, so that the difference in differences estimate does not capture a pure minimum wage effect. They concluded that “Little, if any, of the loss of employment among women as a group can be related to the minimum-wage determinations” (p. 12). Nevertheless there was unmistakable evidence, confirmed in the interviews, that experienced women in the least-skilled positions (such as errand girls) lost their jobs in favor of girls and “apprentices” (women with less than one year’s experience) who could be paid $6 instead of $9.25 per week. This effect was explained as follows:

[Quoting Obenauer and Nienburg] department-store men do not consider an ordinary bundle wrapper or a stock girl, whatever her experience, to be worth $9.25. To earn $9.25, in the judgment of the employer, she must be put at work requiring more skill.Twenty-three of the women making changes in occupation had . . . gone into better positions. . . . There are some women, however, who have an aversion for certain occupations and others who can not perform more skilled duties…. Under the present conditions they will not be retained more than their [one-year] apprenticeship period. (p. 72)

Back to Kennan on Card and Krueger:

Figure 3 is more sobering. The seasonally adjusted employment rate for teenagers (both sexes, 16 or 17 years old) displays large cyclical swings and a high degree of serial correlation. Among the 17 minimum wage changes shown in Figure 2, the median increase was 12 percent, which would produce at most a three percent reduction in the teenage employment rate, if the conventional summary is accepted. That means we are looking for employment rate changes of about one percentage point, and such changes happen all the time, even from one month to the next. In short, we are looking for a needle in a haystack.

Here’s Kennan’s great drug-test analogy:

The difference in differences estimate in Table 3 says that the minimum wage increase [in New Jersey] caused a significant increase in employment. Employment did not change in New Jersey, while employment fell in Pennsylvania. Should we conclude that the same fall in employment would have been seen in New Jersey if the minimum wage had not been increased? The answer depends on conjectures about what caused the employment changes in Pennsylvania. My own reaction is that the same difference in differences would have been more persuasive if employment had risen by 2.7 in New Jersey, with no change in Pennsylvania. The result in Table 3, on the other hand, is like having a drug trial in which the drug has no effect but the placebo makes people sick.

Kennan’s penultimate evaluation of Card and Krueger:

Myth and Measurement is a serious, well-written book, well worth reading (despite its misleading title). It is at its best when using standard methods to look at new data, and at its worst when pretending to use new economics to explode myths. A high standard is maintained in the formal argument, but there is an ambivalence about interpretation that is disconcerting. Sometimes the empirical results are cautiously stated: “The weight of this evidence makes it very unlikely that the minimum wage has a large, negative employment effect” (p. 390). Few would disagree with this, but it is hardly worth 400 pages. Elsewhere, a more controversial tone appears:

[Quoting Card and Krueger] Under close scrutiny, the bulk of the empirical evidence on the employment effects of the minimum wage is shown to be consistent with our findings in chapter 2-4, which suggest that increases in the minimum wage have had, if anything, a small, positive effect on employment, rather than an adverse effect. In our opinion, the conventional view that increases in the minimum wage necessarily have an adverse effect on employment has very weak empirical foundations. At a minimum, we believe that our reanalysis of the literature should encourage economists to keep an open mind about the effect of a minimum wage. (p. 236)

Here the authors seem intrigued by the novelty of a positive employment effect, and can not quite resist it even though the result is surely as fragile as the competing results that they criticize. Is the purpose of the book to lay claim to a revolutionary new finding, just in case it turns out to be right in the end? This might explain the argumentative style that runs through the book, distracting attention from the results. The findings are tilted toward the view that the minimum wage does not reduce employment: results that are favorable to this view are accepted at face value, but unfavorable results are exposed to “close scrutiny” and found wanting.

The finding of “no large or systematic effects on employment” in California was quoted above. That quote was taken from the summary at the end of the chapter on California. The summary at the beginning of the same chapter says the same thing, more or less:

[Quoting Card and Krueger] Nevertheless, we find no indication that these wage gains led to employment losses for teenagers or other low-wage workers. To the contrary, we find the rise in the minimum wage actually may have increased both wages and employment rates of teenagers in the state. Even in the retail trade industry, we find little evidence that the increase led to significant employment losses. (p. 79)

Here the congenial results for teenagers are highlighted, and the uncongenial retail trade results are minimized. A more balanced assessment is that the minimum wage may have a small effect on employment, but it is hard to detect in noisy data. Why not just say that?

And Kennan’s conclusion:

Myth and Measurement’s lasting contribution may well be to show that we just don’t know how many jobs would be lost if the minimum wage were increased to $5.15, and that we are unlikely to find out by using more sophisticated methods of inference on the existing body of data. What is needed is more sophisticated data. The fast food data for Texas and New Jersey show the potential benefits, but these data were collected with minimal resources. Given the resources available for data collection in government statistical agencies, much more could be done, as was shown 80 years ago by Obenauer and Nienburg (1915).