≡ Menu

Bellante and Picone on Card and Krueger

Until last night, I thought that I’d shared here at Cafe Hayek the excellent Fall 1999 Journal of Labor Research paper by Don Bellante and Gabriel Picone on David Card’s and Alan Krueger’s famous empirical research on minimum-wage legislation – research that is purported to show that hiking minimum wages at least sometimes has little or no negative, and perhaps even a positive, impact on the employment opportunities of low-skilled workers. But I discover that I’ve not yet shared the Bellante-Picone paper, the title of which is “Fast Food and Unnatural Experiments: Another Perspective on the New Jersey Minimum Wage,” so here it is. Unfortunately, it’s behind a high pay wall. Below the fold I excerpt some key passages. (Long-time patrons of Cafe Hayek might recall that the labor economist Don Bellante is the author also of the insightful 2007 paper “The Non Sequitur in the Revival of Monopsony Theory.”)

Fast Food and Unnatural Experiments: Another Perspective on the New Jersey Minimum Wage – Don Bellante and Gabriel Picone.

Card and Krueger have changed all of that [strong consensus among economists that minimum-wage legislation harms many of the low-skilled workers who are ostensibly meant to be helped], and many pages of print are now devoted to the question of whether or not disemployment effects exist and even to whether or not minimum wages increase employment. It has surely been a source of amusement or frustration to economists to witness the degree of uncritical acceptance given to this work’s finding (of a positive employment effect) by the mass media, the Secretary of Labor, and others when a mountain of previous work has failed to find a positive effect. Granted, Card and Krueger have argued that publication bias combined with specification search produced those prior results. But most all reporting for the mass media is by journalists who are blithely unattuned to this and all other such technical issues in methodology, so it’s unlikely that they were influenced by the specification search argument. The response of the mass media was more likely an example of what psychologists call cognitive dissonance, wherein evidence inconsistent with one’s priors doesn’t seem to make sense and is filtered out of permanent memory.
Our purpose, however, is not to defend the large number of empirical studies that have supported a disemployment effect of minimum wages. In fact we briefly argue below that no studies that have been done or could have been done with existing, generally available data can fully capture the dynamics of minimum wage consequences. Nor is our purpose to analyze or give a description of the running debate over Card and Krueger’s New Jersey studies, though we will summarize some of the criticisms and rebuttals that appear elsewhere. Our main purpose is to criticize the suggestion that the general method employed by Card and Krueger is a superior one to judge minimum wage effects. We argue that however valid the criticisms that have been offered against the data and data collection methods of Card and Krueger, improved data that are exam- ined using the same general approach to the problem do not deserve much confidence. Even if confidence could be placed in results obtained for the fast-food industry, it is highly questionable whether these results can be generalized. In short, the New Jersey minimum wage offers nothing like the “natural experiment” that Card and Krueger have purported it to be.

The running debate on the New Jersey minimum wage can be quickly summarized regarding its history and current state. First, Card and Krueger publish their finding in the American Economic Review showing that the New Jersey minimum wage resulted in an increase in employment in the fast-food industry. Next, the study generates substantial criticism for, among other reasons, the fact that their finding is based on a poorly-worded telephone survey of managers or assistant managers of franchised fast-food establishments throughout New Jersey and in seven noncontiguous, mostly rural counties of eastern Pennsylvania. Neumark and Wascher (1995a, revised in 1997), using actual payroll data partly collected by the Employment Policies Institute, but otherwise following a method similar to Card and Krueger, obtain conclusions that conform to the standard expectation of a disemployment effect. Skipping all of the intermediate broad- sides, Card and Krueger (1998) have sought to counter the various criticisms of their data (but not of their method) by essentially redoing their New Jersey study using data from the Bureau of Labor Statistics Establishment Survey. As with their initial study, they find no support for a “large” disemployment effect immediately following the rise in the New Jersey minimum. But the weight of their various measurements no longer supports their contention of a significant positive effect on employment. In short, the best interpretation that can be given to the most recent Card and Krueger effort is that it really doesn’t show anything. Given their method, that is probably what should have been expected all along.

First and foremost, there is nothing in the usual neoclassical analysis of the supply and demand for labor suggesting that an exogenous rise in the wage will lead each and every industry affected by the rise to lower employment. It is true that the usual text- book representation of the consequences of a minimum wage increase typically reasons from the perspective of a representative firm. The use of a representative firm (or industry) is heuristically appropriate in many applications. For example, one frequently sees analysis at the aggregate level that concludes that increases in labor productivity raise the demand for labor and hence the average real wage. Or stated otherwise, over the long run only such factors as tend to increase output per worker tend to increase income per worker.

But no serious economist would suggest that productivity growth within a specific industry always raises the demand for labor and wages in that industry, nor that wage increases will in each industry be proportionate to productivity increases, and certainly not that there is any correlation between gains in productivity growth and employment growth by industry. Factors like relative elasticity of product demand affect the employment outcomes of specific industries with productivity growth, but can safely be ignored when looking at the aggregate employment affects of productivity growth. Hence we find nothing puzzling about the fact that very rapid productivity growth in agricultural industries has shrunk employment there, while productivity growth in the electronics industry is associated with very strong employment growth.

A similar statement can be made about the minimum wage. Taking the universe of firms where at least the new minimum is binding, there is every reason to expect that in the representative firm or industry, and in the aggregate, employment will shrink. But without violating either the laws of demand and supply or the usual ceteris paribus conditions, one might expect one or more industries to expand employment perhaps even in the long run. To see this, think of food away from home as being obtainable from fast-food restaurants or from, for lack of a better term, table-service restaurants. While the fast-food sector is very capital intensive, the table-service food sector is relatively much more labor intensive. An effective rise in the minimum wage will cause the costs of production to rise in both sectors, but more so in the relatively labor-intensive table-service food sector. Hence if product price were to remain constant, the demand for labor in the fast-food sector would be less elastic than in the table-service food sector.

However there will eventually be product price rises, and logically they would be higher, proportionately, in the table-service food sector where average costs have risen relatively more. We would expect that consumers will reduce their consumption of food away from home in general, but within the food-away-from-home industry, to substitute away from the table-service and toward the fast-food sector. Without this shift in relative consumption, we would expect both sectors to reduce employment with the greater reduction being in the table-service sector. This effect would be reinforced if the demand for table-service food is more elastic than that for fast food. Such relative elasticities seem highly likely, given that table-service food is closer than fast food to the luxury end of a necessity-luxury scale. But with the shift in relative consumption, the overall effect could very well be to increase absolute employment in the fast-food sector, even as overall employment in the industry is reduced. Moreover, to the extent that fast food is an inferior good relative to table-service food, the effect described here will be reinforced by the income effect of the change in the absolute prices of both fast- and table-service food. However, it is hard to believe that this type of income effect could by itself be substantial. It is also unlikely that the effect of substitution in consumption discussed here would materialize quickly, even if, as is unlikely, the price changes that would induce them were to quickly take place.

By focusing strictly on the fast-food sector, Card and Krueger or others looking at that sector might simply confirm part of the possibility described here. Nonetheless, such shifts in relative employment still would be the product of a distorted allocation of labor that is Pareto-inferior to the one that would otherwise take place. But having made the argument that a minimum wage rise might raise employment in some sectors, we are not inclined to view the Card and Krueger methodology as capable of providing evidence in support of that possibility.

Card and Krueger used what they labeled a “difference in differences” method of measuring the growth of fast-food employment (relative to overall employment growth) in New Jersey as compared to the relative growth of fast-food employment in Pennsylvania. This raising of New Jersey’s state minimum, it has been argued, presented a “natural experiment,” but, if so, it is much inferior to a lab experiment. The so-called natural experiment did nothing to control for most of the myriad of variables other than the minimum wage that might affect changes in relative fast-food employment growth in the two states. Simply dividing by overall employment doesn’t begin to do the job. The conclusion drawn from this method is the cross-sectional equivalent of the post hoc ergo propter hoc fallacy. This criticism, though, is a comparatively minor one. For further elaboration on the point, see Hamermesh (1995). More substantial is the issue of the timing of the supposed measurement of the effects of the New Jersey minimum wage rise.

The New Jersey minimum went into effect in April of 1992. However, the legislation raising the minimum was passed in 1990, and the likelihood of passage may have been known well before its actual passage. Despite this, Card and Krueger conducted their survey in February 1992 to gauge New Jersey employment “before” the rise. Their gauging of the potential disemployment effects uses November 1992 employment, seven months after the effective date of the legislated rise. It is of course entirely conceivable that establishments in New Jersey, or some of them, would have made their employment adjustments much in advance of February. If so, it is not entirely clear what the “before” measurements are actually measuring, but it won’t be the case that they measure what New Jersey employment would have been if no rise in the mini- mum wage had been legislated.

If it is argued that finns will not begin adjustments until the actual date that the rise goes into effect, then surely just seven months after the legislated rise, only very short-run employment adjustments will have been made. Under realistic circumstances, those short-run firm level adjustments could be fairly trivial relative to long-run general equilibrium adjustments. The usual neoclassical conception views substitution between capital and labor as a long-run phenomenon wherein a firm chooses from a menu of production techniques such that any given level of output can be produced with various combinations of labor and capital. In the limiting case, the production isoquant approaches a continuous trade-off between capital and labor, but in any event each different possible combination of capital and labor entails a somewhat different production technique. In the short run, with capital fixed, the firm’s only adjustment to a change in the relative price of labor is to vary the proportion in which labor and capital are combined. This variable-proportions model is pedagogically useful for a number of purposes, not the least of which is to illustrate why the firm’s short-run response to a rise in the wage rate can never be as strong as will be its long-run reduction.

But just how variable proportions are in the short run, and so just how much adjustment to the employment of labor will be made, is a complex matter. According to [Alfred] Marshall’s four principles of factor demand elasticity (as explained in Bellante and Jackson, 1983, pp. 102-5), the elasticity of labor demand in an industry is more elastic: (1) the more elastic the demand for the product; (2) the greater the ratio of labor costs to total costs; (3) the greater the ease of substitution between labor and capital; and (4) the more elastic the supply of capital. As for point (1), we have argued above that the elasticity of demand for fast food is likely to be much less than that for food in the table-service sector, particularly if fast food is an inferior item to table-service food. As for point (2), note that the fast-food sector is highly capital intensive, and most of the capital is highly fixed to its specific location. Regarding point (3), the short-run possibility of substituting capital for labor in the fast-food sector is nil. Concerning point (4), the elasticity of supply of fixed capital is going to be highly inelastic in any industry in the short run, but in light of the impossibility of substituting capital for labor in the short run in the fast-food sector, the short-run elasticity of supply of capital is a moot point. All in all then, we should expect that the short-run elasticity of demand for labor in the fast-food sector will be quite low compared to other sectors, and particularly compared to its own long-run elasticity.

All of the above arguments are made with the idea of describing change in a firm’s short-run equilibrium position. But in a world of uncertainty and imperfect information, even adjustments to short-run equilibria will be slow to unfold. It can be argued that the tendency to adjust to a changed short-run equilibrium is particularly slow in the fast-food sector. Card and Krueger argued that the fast-food sector was ideal for testing minimum wage effects, because rapid turnover and low training costs make adjustment costs of less consequence than they would be elsewhere. But there are other considerations that make the fast-food sector quite remote from ideal when looking for evidence of short-run adjustment to disequilibria. To understand this point it should be understood that when we talk of the fast-food sector, we are mainly talking about franchised or chain-operated establishments. Indeed these are the only types of establishments considered by Card and Krueger and by Neumark and Wascher. Wimmer (1966) points out that the brand-name capital associated with franchise enables the franchisee to receive a premium, but the brand name capital must be protected by the maintenance of a contractually provided uniform standard of performance. Level-of-service maintenance is achieved by a simple rule of thumb for staffing in the short-run: Staffing decisions are made on the basis of volume of sales experience. The result is a fairly stable sales-to-employment ratio. Wimmer further suggests that since capital-labor substitution is not possible, the only short-run adjustments available to individual franchisees are to raise prices or accept residually lower profits. Only when the higher prices result in a visible pattern of lower sales will staffing hours (with a lag) be reduced. And if a reduction in profits is the result, the ultimate long-run employment adjustment will be at the industry level in terms of slower expansion in the number of franchises, etc., rather than at the establishment level. With sufficient time in a fluid capital market, differential profits generate differential capital flows. No short-run “difference in differences” measurement ever will be able to gauge this effect.
VI. Conclusion

We have argued that the difference-in-differences method of examining the effects of the New Jersey minimum wage, especially when the differences in differences are measured shortly after the change, is questionable. That being the case, if we nonetheless take this method seriously, and if following Card and Krueger we look at the data shortly before and after the New Jersey minimum was imposed, then the only statistically significant evidence we can find supports the conclusion that the New Jersey minimum reduced employment of young workers.

For what it is worth, we still would be inclined to attach a much greater confidence in the weight of evidence provided in time-series studies. These at least can have the virtue of incorporating important aspects of realism not available in a simple two- state, single-point-in-time measure, one example being distributed lags in the adjustment to changes in the minimum wage. Even so, these are themselves subject to methodological criticisms. The undesirable effects of a legislated minimum are likely to be far more complex than can ever be gauged by a study limited to short-run employment effects. For example, Neumark and Wascher (1995b) have provided evidence that affected firms respond to increases in the wage minimum to some degree by replacing younger teens with older teens, and Black teens with White teens. And in even more subtle ways, access to working middle-class status via entry level jobs may be blocked to individuals in ways that will not show up in employment numbers (Bellante and Porter, 1990, pp. 668-69). However, even to the task of measuring such employment effects as might take place, time-series studies at best must be very approximate and tentative as to their quantitative findings since model specification is of necessity incomplete. That is to say, surely expectations figure into employment decision making. When the legislated minimum rises, do employers expect it to be the highest real minimum to exist over the next several years, or does the legislated rise lead to expectations of further legislated rises? The two possible answers would likely lead to very different employment responses, but how can these expectations be specified in a model that uses the types of data available? Moreover, even if such expectations could be modeled and incorporated into empirical tests, there are other reasons why time-series efforts are problematic. As long as minimum wage changes come at uncertain intervals and magnitudes and adjustment to them is both delayed and distributed, it will be hard to empirically separate responses to the last legislated change from responses to the previous one. Moreover, the maintained but implicit assumption of all time series estimates is structural stability of the relationship under study. We have already mentioned the evidence of structural instability in the minimum wage-employment relationship.
What we would most want to emphasize is as follows: When data are imperfect, or the specifications of models that use the data are imperfect, it could just as well be argued that one is testing the data with theory as it could be argued that one is testing the theory with the data. Of course most economists think of themselves as always doing the latter. But when well-established theory is purported to conflict with the empirical evidence of a marginal study, the hesitancy of the economics profession to reject the core theory has served the profession well. In fact, it is the exceptions to this tendency that have proved most costly for the profession and the society at large. Perhaps the best example is when the profession accepted the “measurement without theory” in the original idea of the Phillips Curve — a permanent trade-off between inflation and unemployment that was exploitable by policy makers. Two decades of very costly policy errors were the result. And so with the minimum wage, it would be most unfortunate if a few studies by a small circle of economists–studies in conflict not only with received theory but most other empirical studies — were to become the justification of another set of damaging policy conclusions.

Next post:

Previous post: