Book Review: Myth and Measurement and the Theory and Practice of Labor Economics

Credible Design-Driven Empirical Research

By the mid-1980s, researchers in the field widely acknowledged that empirical economics was in trouble. At the start of the decade, Hendry (1980) and Leamer (1983) had drawn attention to the problems of specification searching, which informed observers knew undermined many empirical studies. ⁶ Dewald, Thursby, and Anderson (1986) showed that a distressing fraction of recently published empirical papers could not be replicated at all, and others contained serious errors. Closer to home, H. Gregg Lewis (1986), the highly respected authority on union wage effects, concluded that estimates from the body of research that tried to correct for the nonrandom selection of workers into the union sector were not reliable enough to be taken seriously. And in an extremely influential paper on the impact of training, Lalonde (1986) showed that alternative econometric methods gave estimates that deviated substantially from corresponding experimentally based estimates.

This discontent led to a search for research methods that were more transparent and less likely to be influenced by differences in the choice of specification or sample. Our colleague Orley Ashenfelter, who was a forceful advocate of this new focus, called these “hands-above-the-table” methods. The pursuit of more credible methods, together with the facts that minimum wages sometimes change in one state and not in another and that the bite of the federal minimum wage varies widely across states, culminated in the research on employment effects of the minimum wage that was summarized in Myth and Measurement.

Some commentators were troubled by our presentation of a wide collage of evidence without explicit reference to economic theory. To us, this criticism has always been puzzling. Standard theories of how the minimum wage will affect a given firm (or a given labor market) are well known and well understood. The primary goal of the analyses in Myth and Measurement was to assess the magnitude of the causal effects of a rise in the minimum wage—not to squeeze the findings into a particular box. A more serious criticism is that our New Jersey–Pennsylvania study failed to collect data on some of the outcomes that are central to alternative theories, such as quit rates and recruitment flows. Fortunately, more recent studies have been able to address this shortcoming (Giuliano 2013; Dube, Lester, and Reich 2016; Gittings and Schmutte 2016).

The role of empirical evidence in published economic research has been greatly elevated in the past two decades. In the early 1980s, only 38% of articles in top economics journals contained any empirical analysis (Hamermesh 2013). By 2011 that figure had reached 72%. Moreover, purely empirical contributions that seek to measure the causal effects from interesting and/or important programs, regulations, and institutions are routinely published in the top journals in the field.

A large fraction of the evidence in recent studies is based on rigorous research designs including controlled randomized experiments, regression discontinuities, and the simple difference-in-differences that are the backbone of Myth and Measurement. Applied economists have adopted a basic vocabulary for discussing counterfactuals and developed techniques for assessing the validity of proposed designs. They also routinely present descriptive analyses of their data, robustness checks, and tests of identifying assumptions, and they use graphical methods to convey the main features of their analyses. As Lemieux notes in his companion piece in this symposium, such “modern” practices were not so commonly used 20 years ago. These shifts in the research community’s understanding of research design principles and our ability to quickly communicate strengths and weaknesses of a given study have greatly facilitated the acceptance of rigorous empirical evidence in the field.

Wage-Setting Models

Although most economists agree that sellers of breakfast cereal and ketchup set prices that are somewhat above their marginal costs, the idea that firms have any wage-setting power has been very slow to take hold in labor economics. Many of the harshest critics of Myth and Measurement expressed the view that low-wage labor markets cannot be characterized as monopsonistic because many potential employers are in the market and the firm-specific supply function is therefore perfectly elastic.

In the past two decades, however, a much clearer understanding of market power in the presence of market frictions has emerged from the search literature. Indeed, search models have become a standard tool in macroeconomics, and two labor economists—Dale Mortensen and Christopher Pissarides—shared the Nobel Prize with Peter Diamond for their fundamental work on frictional labor market models. A growing body of evidence, much of it building on the contribution by Abowd, Kramarz, and Margolis (1999), also points to the importance of firm-specific wage setting in overall inequality and the wage gaps between different groups (see, e.g., Card, Cardoso, Heining, and Kline 2017). ⁷ As Lemieux notes in this symposium, “the traditional competitive model of the labor market is no longer the workhorse of empirical labor economics it used to be when Myth and Measurement was written.” Addison and Dube, also part of this symposium, both make similar observations.

Manning (2003) pioneered the use of the type of wage-posting models developed by Burdett and Mortensen (1998) for labor market policy analysis. These models (which were briefly discussed in Myth and Measurement) describe the equilibrium distribution of wages that arises as employers consider the trade-off between their position on the wage ladder and their turnover rate. ⁸ As shown by Manning (2003) and Engbom and Moser (2016), these models can be readily used to study the market-wide effects of minimum wages, particularly the spillovers that can affect higher-paid workers when the minimum is raised.

Flinn (2006, 2010) considered an alternative class of search and matching models (developed by Pissarides 1985 and Mortensen and Pissarides 1994) that emphasized bargaining over the idiosyncratic value of worker–firm matches. These models provided a simple explanation for the spike in the observed distribution of wages at the minimum wage—a longstanding puzzle for traditional labor market models. ⁹ They can also be readily used to analyze firm-level and market-level responses to minimum wages.

A third class of models, formally identical to the models used in industrial organization to model differentiated products, assumes that workers know the wages offered by different employers but have idiosyncratic valuations for the non-wage features of each job, such as the location (Bhaskar, Manning, and To 2002; Card et al. 2017). These models map directly to the traditional monopsony model described by Robinson (1933). Yet, as in the product market setting, each firm can have market power even when the number of firms is large (Berry and Pakes 2007). All three types of models emphasize the supply-side impacts of minimum wages in the presence of monopsonistic wage markdowns, and predict that a rise in the minimum wage has an ambiguous effect on the level of employment at a given firm, and in the market as a whole.

The Weight of the Evidence Post-1995

One of the main conclusions of our work was that moderate increases in the minimum wage have little or no effect on employment. Though highly controversial at the time, the weight of the evidence over the past two decades has supported this view. Our original conclusions were validated in our follow-up study (Card and Krueger 2000), which used administrative data on employment counts from the Bureau of Labor Statistics’ (BLS) ES-202 database. Using a difference-in-differences specification with controls for different chains, our original survey data revealed that total employment rose by 2.49 workers (standard error = 1.32) at New Jersey restaurants relative to those in seven counties of Pennsylvania between February and November 1992. The BLS data showed a relative increase of 1.05 (standard error = 1.32). An even larger relative change was registered in the BLS data from March 1992 to March 1993 (2.35, standard error = 1.68)—a comparison that controls for any differences in seasonal factors in the two states. Other comparisons, including those using a larger 14-county sample of Pennsylvania restaurants, and others within New Jersey, showed a range of estimates centered above zero but in all cases insignificantly different from zero, exactly as in our original study.

Our conclusion has also been confirmed by many other studies of minimum wage increases in the United States. A recent meta-analysis of more than 400 estimates from 23 minimum wage studies published between 2000 and 2013 by Wolfson and Belman (2014) concluded that the median elasticity of employment or hours with respect to the minimum wages is −0.05; the precision-weighted median, which takes account of the sampling errors of the individual estimates, is −0.03. (A similar conclusion is reached by Doucouliagos and Stanley [2009].) Recent studies are about equally as likely to find positive as negative employment effects of the minimum wage, with the typical estimate very close to zero.

Some of the most innovative recent studies build on our New Jersey–Pennsylvania study and look at the effects of state-wide minimum wages using comparisons between counties on opposite sides of a state border (Dube, Lester, and Reich 2010, 2016; Addison, Blackburn, and Cotti 2012, 2014). By combining multiple pairs of counties and pooling data for multiple years, these studies are able to provide credible and relatively precise estimates of the effect of minimum wage increases. The estimated employment effects of minimum wages from this approach cluster around zero. Other compelling studies focus on the experiences at specific retail-sector firms that operate similar stores in different states (e.g., Giuliano 2013; Hirsch, Kaufman, and Zelenska 2015). Again, the employment effects estimated by these studies are typically small and sometimes positive rather than negative.

Another set of important conclusions from our work concerned the equalizing effect of higher minimum wages. We concluded that increases in minimum wages lead to reductions in wage inequality and increases in earnings and incomes for lower-income workers and families. Again, both conclusions were controversial. Many economists in the 1970s and 1980s believed that the elasticity of demand for workers affected by minimum wages was less than −1, so any increase in the minimum wage would reduce the total earnings received by these workers, causing an increase in earnings inequality. In view of the evidence now available, this appears to be extremely unlikely. Indeed, most of the subsequent literature found substantially larger effects of the minimum wage on reducing inequality than we found (see Chapter 9 of Myth and Measurement).

Two decades ago many economists also believed that any earnings gains arising from a minimum wage increase would accrue to secondary earners in wealthy families. As we noted in Myth and Measurement, this was not true in the early 1990s, and today even larger shares of low-wage workers live in low-income families and the positive distributional effects of a rise in the minimum wage are clearer (CBO 2014).

Relative to the literature on the employment effects of minimum wages, there are fewer recent studies of the distributional impacts. DiNardo, Fortin, and Lemieux (1996) concluded that minimum wages had a very strong effect on measures of wage inequality for female workers. Lee (1999) reached a similar conclusion for workers as a whole, showing there were potentially large spillover effects on higher-wage workers. Autor, Manning, and Smith (2016) argued that Lee’s methodology overstates the effect of the minimum wage, but they still found an important effect of minimum wages on wage inequality and evidence of spillover effects. Exploiting the remarkable history of minimum wage legislation in the United Kingdom, Dickens et al. (2012) concluded that the introduction of the national minimum wage had a strong effect on the lower tail of British wages, pushing up the wages of workers as high as the 35th percentile in the overall wage distribution.

There has been somewhat more recent research on the effects of minimum wages on family earnings, incomes, and poverty. Dube (2014) summarized 12 studies and found an average elasticity of the group-specific poverty rates to the minimum wage of −0.15. Although the poverty rate is arguably a fuzzy yardstick for assessing the distributional effects of the minimum wage, recent studies confirm that minimum wages have a poverty-reducing effect.

Minimum Wage as a Policy Tool

The federal minimum wage has not increased since 2009, and it seems unlikely that any increase will occur soon. Nevertheless, minimum wages remain highly popular with voters (see Bartels 2009), and their use as a policy tool appears to be more widely accepted than it was two decades ago, when even many liberal economists supported lowering or even eliminating the minimum wage.

As was true in the late 1980s, when Congress lags, states become increasingly likely to fill the void by raising their own minimums. As of this writing, a total of 30 states and the District of Columbia have set a minimum wage above the federal level. The processes by which this occurred are quite diverse. In some states—including several traditionally “red” states—higher minimum wages have arisen through referendums, demonstrating the greater popularity of minimum wage regulations among the general public than among many politicians. In the case of New Jersey, the governor proposed a referendum after vetoing legislation to raise the minimum. The referendum that was eventually passed, however, established a higher minimum wage in the state constitution that is permanently indexed to inflation. Another interesting variant is the minimum wage law in California, which grants the governor the power to temporarily delay the increases scheduled over the next six years in the event of poor economic conditions in the state.

An unprecedented number of cities (currently more than 30) have also enacted ordinances to raise local minimum wages above the federal level (NELP 2016). Many of the city minimums, including those in Los Angeles, San Francisco, and Seattle, are slated to rise as high as $15 an hour over the next few years. These state and city minimum wage increases will provide natural experiments for economists to analyze in years to come.

Although the political economy of the minimum wage has not changed much in the United States over the past two decades, outside the United States the situation is very different. In 1993 the U.K. government abolished the system of wage councils, which had set industry-level minimum wages for 80 years. Just a few years later the Blair government created an independent Low Pay Commission to advise the government on the minimum wage and to conduct research on its effects. In 1999 the commission’s recommendations for a national minimum wage went into effect. The new policy was strongly opposed by business interest groups and the Conservative Party but has become much less controversial over time, as its effects were discovered to be far more benign than critics had claimed.

The experiences with wage floors in the United Kingdom have generated much research. The rapid transition from sector-specific minimums set by the Wage Councils to no minimum at all, and then to a nationwide floor of £6.50 in 2014 (the equivalent of $9.86 an hour in U.S. dollars at the current exchange rate) provides a unique natural laboratory for studying the effects of the minimum wage. Research by the Low Pay Commission and by others has generally found that the national minimum wage reduced inequality in the bottom half of the income distribution, narrowed the gender pay gap, and did not adversely affect employment (e.g., Metcalf 2008).

Germany provides another example of an advanced economy that went from no minimum wage to a national wage floor set at a relatively high level. Although the idea of a minimum wage was long opposed by Chancellor Angela Merkel, she ultimately agreed to a minimum wage of 8.50 euros, effective January 2015 (about $9.50 an hour in U.S. dollars at the current exchange rate). The law provided a two-year grace period for some employers and certain categories of workers, but immediately raised the minimum wage well above the level in the United States for most employers. We expect the experiences in Germany, where rich administrative data on the universe of private-sector jobs are available to analysts, to provide another important setting for minimum wage research in the years to come.