Myth and Measurement: The New Economics of the Minimum Wage.

This book analyzes data to answer the question: what are the effects of a "modest" minimum wage increase? Contrary to the answer of the standard textbook model, the authors find that such an increase does not reduce employment of low wage workers. Nor does it much reduce profits of firms that employ such workers. Instead, consumers of products of firms that employ such workers experience price increases. Card and Krueger also find that wages of low wage workers increase with a "modest" minimum wage increase, and that such workers are disproportionately from households with low earnings. They conclude that the gap between the earnings of those in the lower and upper tails of the household earnings distribution lessens with a "modest" minimum wage increase. Finally, they explain their findings with a set of models that reject the notion that employers take wages and prices as given.

The book's new empirical evidence on the employment effects of a minimum wage hike is from seven "natural experiments." These find the proportional change in employment after and before a minimum wage hike for each unit of observation, (employer or state), and compare these changes for units with different "exposures" to the hike. They compare a group of low wage employers from a state where a state minimum wage hike occurred, with a similar group from a state where no such hike occurred. Or they compare a group of high with a group of low wage employers when a federal minimum wage hike occurred. Or they compare states with differing proportions of low wage workers when a federal minimum wage hike occurred. The authors find that employment increases with a minimum wage hike in all seven experiments; two of these increases are statistically significant [p. 338].

The "difference in differences" method of Card and Krueger controls for causes extraneous to the minimum wage hike better than did earlier work on employment effects of a minimum wage. A "natural experiment," however, lacks the social experiment's random assignment that implies that measured group mean differences are unbiased estimates of mean effects of causes. Also, the authors' Texas and New Jersey-Pennsylvania experiments phone surveyed managers of national fast food franchises. Such franchises may not represent low wage industries well, and phone survey data uncorroborated by employment records are suspect. Finally, the 38% response rate for "difference in differences" in the Texas experiment survey [pp. 58-59], calls into question whether its statistically significant 20% employment increase is an unbiased estimate of a mean effect of the federal minimum wage hike.

The authors argue that earlier aggregate time series studies of the U.S. economy rely on the questionable Kaitz index, and probably suffer from publication, specification search, and structural change biases. Also, "if one estimates exactly the same time-series models that have been estimated in the past but includes more recent data, then the minimum wage has a numerically smaller and statistically insignificant effect on employment" [p. 205]. This is perhaps their most telling evidence. They also reassess cross-section and panel-data studies of the U.S. economy. They conclude that "the bulk of the empirical evidence on the employment effects of the minimum wage is . . . consistent with our findings . . . which suggest that increases in the minimum wage have had, if anything, a small positive effect on employment" [p. 236]. Finally, they reanalyze studies of Britain, Canada, and Puerto Rico and decide that these do "not provide unambiguous support for the textbook model of the minimum wage" [p. 271].

The book shares weaknesses with the work that it criticizes. For example, its statistically significant finding of a 3% employment increase from a cross state restaurant industry regression may suffer from specification search bias. Or when "[w]e limit our analysis to 15 [time series] studies that used quarterly data" [p. 189] of 21 considered, selection bias may enter their finding of possible publication bias. The book's extensions and reanalyses of older time-series with old models still rely on the questionable Kaitz index. Card and Krueger cautiously conclude: "[t]he consistent finding of weak or negligible employment effects in both the United States and elsewhere suggests that the problem may lie with the textbook model, rather than with the evidence" [p. 272]. Is an even more cautious conclusion appropriate? The available empirical evidence on the magnitude and direction of the employment effect of a minimum wage hike remains somewhat ambiguous and methodologically problematic.

Card and Krueger find the proportional change in stock market share prices of each firm before and after a political event related to passage of a change in federal minimum wage law, and compare changes for all firms with those for firms that employ primarily low wage workers. They find no "difference in differences" for 1991-1995 events, and ambiguous results for events just prior to the 1991 federal minimum wage hike. They also concede that "the conclusions that we draw from this analysis must be viewed as tentative" [p. 347]. Their strongest, and probably least controversial, findings are that statistically significant increases in the earnings of low wage workers, whose average magnitudes range from 5% to 11%, follow a minimum wage increase in all seven "natural experiments" [p. 388].

To explain their employment increase finding, the authors show that there are partial equilibrium models in which employers take wages and prices as given that imply employment decreases with a minimum wage hike. They then show that there are partial equilibrium models in which employers set wages or prices that imply employment increases with a minimum wage hike. Citing four other of their findings, Card and Krueger argue:

Considered individually, each feature can be explained by a suitably modified version of the textbook model. It is more difficult to develop a unified explanation for their coexistence. Perhaps as importantly, the textbook model assumes that equally skilled workers are paid the same wage at all employers. It is difficult (although not impossible) to reconcile this assumption with the range of wages that exist in the labor market for seemingly identical workers [p. 390].

Card and Krueger conclude by favoring a theory that supposes that employers set wages or prices over one that supposes that employers take wages and prices as given.

The reasoning with which the authors reject a theory that employers take wages and prices as given is problematic. Neither their explanatory models in which employers take wages and prices as given, nor their explanatory models in which employers set wages or prices, are consistent with the econometric models that generated their employment increase findings. Also, there are partial equilibrium models in which employers take wages and prices as given that do imply an employment increase with a minimum wage hike. Similarly, there are partial equilibrium models in which employers set wages and prices that do not imply an employment increase with a minimum wage hike.

The authors' partial equilibrium perspective is too restrictive. They show that when a firm takes wages and prices as given, "L = h(y, w, r)" [p. 356] with (L, y, w, r) = (labor hours, output, wage, capital price). If y is also firm share of market output demand, and (p, Y) = (output price, income distribution), then y = y(p, Y). So if incomes of low wage workers increase with a minimum wage hike, then output may increase, e.g., low wage workers may buy more fast food with their additional incomes, and hence labor hours may increase. Similar general equilibrium reasoning with models in which employers take wages and prices as given unifies and explains all of the authors' findings.

Finally, since wages that the authors observe vary across at least a state, they are not likely set in the same labor market. In addition, since each worker is a unique bundle of skills, judging two workers identical is difficult. Also, employers don't buy skills alone; they buy skills and motivation. So the assumption that observed wages are set in a single labor market and the standard "human capital" metaphor, both of which underpin empirical labor economic research, seem inadequate. These inadequacies explain the variation in wages that the authors observe as well as does some flaw in the assumption that employers in a given market pay the same wages for the same labor.

The authors argue that a modest minimum wage increase will result in no loss of economic efficiency, i.e., employment, and so favor one on the distributional grounds that it will reduce slightly the gaps between the upper and lower tails of both the household earnings and income distributions. Suppose that one weakens their first premise to: empirical econometric studies of whether a modest minimum wage increase would cause an employment decrease are all methodologically problematic and differ in their conclusions. Then a modest minimum wage increase still remains a good idea because most voters favor one as a way to assure that all who wish to can work their way out of poverty. When science is inconclusive on the efficiency effects of a measure for modest redistribution toward the poor that most voters favor, democracy and the findings of no efficiency loss ought to trump the efficiency loss findings.

Card and Krueger cite Will Rogers: "It's not what you don't know that's the problem. It's what you think you know that ain't so" [p. 208]. Before Myth and Measurement, economists thought that an increase in the minimum wage caused a decline in employment. Now we know that we do not yet know the employment effect of a minimum wage increase. Nor have we succeeded in getting a theory that assumes prices and wages are given, one that assumes employers set prices or wages, and empirical observations of labor markets to confront each other in ways that rule out one of the theories. The book's principal unintended lesson is that labor economics has not quite got its scientific footing yet. No book could teach a more important lesson.

Leland G. Neuberg New York University