Pomen replikacije znanstvenih raziskav: primer minimalne plače

Zgodba je naslednja: Leta 1995 sta David Card in Alan Krueger v American Economic Review objavila članek o vplivu dviga minimalne plače v New Jerseyu na zaposlenost v restavracijah hitre prehrane. Ugotovila sta, da dvig plače ni imel negativnega učinka na zaposlenost. Leta 2000 sta David Neumark and William Wascher objavila kontra študijo oziroma komentar (na podlagi študije, narejene sicer že leta 1995), ki naj bi na drugačnih podatkih pokazala nasprotno: da je dvig minimalne plače v New Jerseyu vplival na relativno zmanjšanje zaposlenosti v restavracijah hitre prehrane v New Jerseyu glede na Pensilvanijo. No, Card in Krueger sta istega leta (2000) nato objavila odgovor, v katerem sta primerjala obe vrsti podatkov ter pokazala, da so podatki Neumarka in Wascherja nereprezentativni in da sta naredila metodološko napako pri merjenju zaposlenosti (zaradi drugačne periode pri poročanju). Ko sta Neumark-Wascherjeve podatke ustrezno metodološko obdelala, sta še enkrat potrdila ugotovitve svoje prve študije: da dvig minimalne plače ni imel niti absolutnega niti relativnega negativnega učinka na zaposlenost v New Jerseyu.

Primer kaže na pomen možnosti replikacije znanstvenih raziskav, kar seveda zahteva dostopnost tako originalnih podatkov kot tudi statistične/ekonometrične kode. Spodaj je kratek izsek iz pojasnil Carda in Kruegerja, če koga zanima.

Replication and reanalysis are important endeavors in economics, especially when new findings run counter to conventional wisdom. In their Comment on our 1994 American Economic Review article, David Neumark and William Wascher (2000) challenge our conclusion that the April 1992 increase in the New Jersey minimum wage led to no loss of employment in the fast-food industry. Using data drawn from payroll records for a set of restaurants initially assembled by Richard Berman of the Employment Policies Institute (EPI) and later supplemented by their own datacollection efforts, Neumark and Wascher (hereafter, NW) conclude that “… the New Jersey minimum-wage increase led to a relative decline in fast-food employment in New Jersey” compared to Pennsylvania. They attribute the discrepancies between their findings and ours to problems in our fast-food restaurant data set. Specifically, they argue that our use of employment data derived from telephone surveys, rather than from payroll records, led us to draw faulty inferences about the effect of the New Jersey minimum wage.

In this paper we attempt to reconcile the contrasting findings by analyzing administrative employment data from a new representative sample of fast-food employers in New Jersey and Pennsylvania, and by reanalyzing NW’s data. Most importantly, we use the Bureau of Labor Statistics’s (BLS’s) employer-reported ES-202 data file to examine employment growth of fast-food restaurants in a set of major chains in New Jersey and nearby counties of Pennsylvania.2 We draw two samples from the ES-202 files: a longitudinal file that tracks a fixed sample of establishments between 1992 and 1993, and a series of repeated cross sections from the end of 1991 through 1997. Because the BLS data are derived from unemploymentinsurance (UI) payroll-tax records, the employment measures are free of the kinds of survey errors that NW allege affected our earlier results. In addition, because the ES-202 data include information for all covered employers in a fixed group of restaurant chains, there is no reason to doubt the representativeness of the BLS sample.

A comparison of fast-food employment growth in New Jersey and Pennsylvania over the period of our original study confirms the key findings in our 1994 paper, and calls into question the representativeness of the sample assembled by Berman, Neumark, and Wascher. Consistent with our original sample, the BLS fast-food data set indicates slightly faster employment growth in New Jersey than in the Pennsylvania border counties over the time period that we initially examined, although in most specifications the differential is small and statistically insignificant. We also use the BLS data to examine longer-run effects of the New Jersey minimum-wage increase, and to study the effect of the 1996 increase in the federal minimum wage, which was binding in Pennsylvania but not in New Jersey, where the state minimum wage already exceeded the new federal standard. Our analysis of this new policy intervention provides further evidence that modest changes in the minimum wage have little systematic effect on employment.

In light of these results we go on to reexamine the Berman-Neumark-Wascher (BNW) sample and evaluate NW’s contention that the rise in the New Jersey minimum wage caused employment to fall in the state’s fast-food industry. Our reanalysis leads to four main conclusions. First, the pattern of employment growth in the BNW sample of fast-food restaurants across chains and geographic areas within New Jersey is remarkably consistent with our original survey data. In both data sets employment grew faster in areas of New Jersey where wages were forced up more by the 1992 minimum- wage increase. The differences between the BNW sample and ours are attributable to differences in the BNW sample of Pennsylvania restaurants, which unlike the more comprehensive BLS sample, and our original sample, shows a rise in fast-food employment in the state. Second, the differential employment trend in the BNW Pennsylvania sample is driven by data for restaurants from a single Burger King franchisee who provided all the Pennsylvania data in the original Berman sample.

Third, the employment trends measured in the BNW sample are significantly different for restaurants that reported their payroll data on a weekly, biweekly, or monthly basis. Establishments that reported on a biweekly basis had faster growth than those that reported on a monthly or weekly basis. We suspect that the different reporting bases matter because the BNW employment measure is based on payroll hours (rather than actual numbers of employees) and because weekly, biweekly, and monthly averages of payroll hours were differentially affected by seasonal factors, including the Thanksgiving holiday and a major winter storm in December 1992. Regardless of the explanation, a higher fraction of Pennsylvania restaurants reported their data in biweekly intervals, leading to a faster measured employment growth in that state. Once the employment changes are adjusted for the reporting bases, the BNW sample shows virtually identical growth in New Jersey and eastern Pennsylvania. Finally, a reanalysis of publicly available BLS data on employment trends in the two states shows no effect of the minimum wage on employment in the eating and drinking industry.

Based on all the evidence now available, including the BLS ES-202 sample, our earlier sample, publicly available BLS data, and the BNW sample, we conclude that the increase in the New Jersey minimum wage in April 1992 had little or no systematic effect on total fast food employment in the state, although there may have been individual restaurants where employment rose or fell in response to the higher minimum wage.

Vir: David Card & Alan KruegerAmerican Economic Review, 2000

%d bloggers like this: