The Card-Krueger Study
- Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania," David Card &Alan B. Krueger, The American Economic Review, Vol. 84, No. 4. (Sep., 1994), pp. 772-793.
They find that increasing the minimum wage increases employment at fast-food restaurants. This result is destroyed by two other papers. Neumark & Wascher replicate the study using clearly better data and find that employment falls, not rises. Welch notes that the Card-Krueger data was collected with a confusing survey method and the data clearly has big mistakes, showing wild swings in employment and wages, including wages falling in many restaurants after the minimum wage increase, and to illegally low levels.
The Neumark-Wascher Critique: Using Better Data Reverses the Conclusions
- "The Effect of New Jersey's Minimum Wage Increase on Fast-Food Employment: A Re-Evaluation Using Payroll Records," David Neumark & William Wascher. Card and Kreuger's employment data was bad. They got it by phoning up restaurants---or having research assistants do it. N-W got administrative payroll data from the restaurants. This took 5 years to appear in print, as: Neumark, David, and William Wascher. 2000. “Minimum Wages and Employment: A. Case Study of the Fast-Food Industry in New Jersey and Pennsylvania: Comment.” American Economic Review 90, no. 5 (December): 1362-1396.
- David Card & Alan B. Krueger, "Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania: Reply." American Economic Review, December 2000, 90(5), pp. 1397-420.
The Card-Krueger data were elicited from a survey that asked managers or assistant managers “How many full-time and part-time workers are employed in your restaurant, excluding managers and assistant managers?” This question is highly ambiguous, as it could refer to the current shift, the day, or perhaps the payroll period, and the respondents’ interpretation of it could differ in the observations covering the periods before and after the minimum wage increase. In contrast, the payroll data referred unambiguously to the payroll period used by the restaurant. Reflecting this difference, the data collected by Card and Krueger had much greater variability across the two observations than did the payroll data, with changes that were sometimes implausible.
In contrast both to their original study and to our replication, their reanalysis generally finds small and statistically insignificant effects of the increase in New Jersey’s minimum wage on employment, and they conclude that “the increase in New Jersey’s minimum wage probably had no effect on total employment in New Jersey’s fast-food industry, and possibly had a small positive effect”.
- "MINIMUM WAGES AND EMPLOYMENT: A REVIEW OF EVIDENCE FROM THE NEW MINIMUM WAGE RESEARCH," David Neumark and William Wascher October 2006:
In a review piece on Card and Krueger’s book Myth and Measurement, Richard Freeman (1995) wrote that “their analysis is a model of how to do empirical economics” (p. 831). Similarly, Paul Osterman (1995) asserted that Card and Krueger “make a powerful case that what they term ‘natural experiments’ are a more appropriate way to conduct policy analysis than cruder research based on time-series or broad cross-sections” (p. 839).
However, other labor economists were more critical of these studies. In particular, Finis Welch (1995)—again referring to Myth and Measurement—states that: “I am convinced that the book’s long-run impact will instead be to spur, by negative example, a much-needed consideration of standards we should institute for the collection, analysis, and release of primary data” (p. 842). Likewise, Hamermesh (1995) concludes that “even on its own grounds, CK’s strongest evidence is fatally flawed” (p. 838).
In general, the criticisms of the case study approach focused on four issues. The first is the question of adequacy of the control groups used in the studies, a concern emphasized by both Hamermesh and Welch. On its face, for example, it seems reasonable to question the use of Georgia, Florida, and Dallas/Ft. Worth as adequate control groups in Card’s study of the California minimum wage increase, given that these places are far from California and likely influenced by very different demand conditions. But even for states with geographic proximity, using one state as a control may be problematic. For example, Deere et al. (1995) pointed out that teenage employment rates (as measured by the CPS) in New Jersey diverged significantly from those in Pennsylvania beginning in 1988, casting doubt on Card and Krueger’s claim that Pennsylvania represents a sensible control group with which to compare New Jersey.
More broadly, Hamermesh questioned the practicality of this entire approach for studying the effects of minimum wages, noting that the variance in employment seems to be dominated by demand shocks, which suggests that “any changes in the relative demand shocks” affecting two geographic areas will easily “swamp the effect of a higher minimum wage” (p. 837).
The Welch Critique of the Data
Welch, Finis. "Myth and Measurement: The New Economics of the Minimum Wage: Review Symposium: Comment." Industrial and Labor Relations Review, July 1995, 48(4), pp. 842-49. Welch absolutely destroys Card & Krueger. He shows that their data just can't be correct because individual restaurants have wild swings in employment and wages, and how that was probably the result of bad survey design that confused the restaurants into giving inconsistent answers:
Although the notoriety surrounding Myth suggests important conclusions that challenge economists' fun- damental assumptions, I am convinced that the book's long-run impact will instead be to spur, by negative example, a much- needed consideration of standards we should institute for the collection, analysis, and release of primary data. ...
The numbers are incredible. It is not clear that the interview process was formalized. There has been no re- sponse to my requests for anything regard- ing interviewer instructions and training, coder instructions, or pre-test results. The two key questions, concerning wages and employment, invite inaccurate responses. Immediately after the introduction the employment quiz begins:
Q1. How many full-time and part- time workers are employed in your restaurant, excluding managers and assistant managers?
Q2. And how many managers and assistant managers?
The interview form has two spaces for the answer to the first question (full-time and part-time) and one space for the sec- ond. Question 1 makes the assumption that assistant managers cannot be part-time; it fails to define part- and full-time; and its compound nature probably confused re- spondents. The wage questions are no better:
Q4. What is the average starting wage for a nonmanagement employee at your restaurant today?
Q5. Is it the same starting rate for full-time and part-time workers?
Question 4 has two blanks in which an interviewer would write a response. The first blank is for a numeric response and the second is for "minimum wage" so that the coder can fill in the applicable value. Question 5 has two spaces, one for part-time and the other for full-time. Although the second wage question appears to have been intended to draw two responses, many might believe a simple "yes" or "no" would suffice. In any case, reviewers cannot exam- ine the coded responses to Questions 4 and 5 for evidence of the confusion one expects them to elicit, because the "analysis" file that is distributed contains only one start- ing wage for each survey wave with nothing to indicate whether it vey wave with nothing to indicate whether it is taken from Ques- tion 4 or Question 5. ...
Among the 79 survey records shown for Pennsylvania, 68 have valid wage records in each of the two waves. At the baseline interview, 22 of these 68 are coded as hav- ing a starting wage of $4.25 (the applicable minimum), while the remaining 46 had starting wages ranging from $4.35 to $5.50 per hour. By the time of the second wave, 13 of the 22 (59%) that initially paid $4.25 had increased their starting wage. It may not be surprising that starting wages in- creased, but would it be surprising if they fell? Among the 46 who initially paid more than the minimum, who therefore could reduce wages without violating federal law, 27 (also 59%) are coded as having done so by the time of the second interview!
Among the 331 records for New Jersey, 302 have valid starting wage observations for each survey wave and 23 of them had first wave starting wages above $5.05 per hour. They are the only ones that could lower the start- ing wage without violating New Jersey law. The analysis files show that 19 of the 23 (83%) lowered their starting wage and 18 of the 19 lowered the starting wage to the second wave minimum of $5.05 per hour! ...
Strikingly, 46 of the 311 (15%) that initially had full-time employees reportedly had none by the second interview, 7-8 months later. Conversely, 47 of the 67 (70%) that initially had no full-time em- ployees are coded as having added them by the time of the second interview. The magnitude of these swings is not trivial. Employment averaged a little over 21 work- ers in both interviews.7 Among stores that lost all full-time workers, the average loss was 10.7. Among stores that initially had none and are coded as having added them by the second interview, the average gain was 10.4. Is the technology so flexible and in such rapid flux that these numbers are unremarkable, or are the coded responses dominated by error?
Among the 378 stores with valid non- zero employment in both survey waves, average first wave employment is 21.14 and average second wave employment is 21.27. The trivial difference in averages between survey waves masks astonishing changes within stores. The largest gain is 34 employees, the largest loss is 41.5, and the standard deviation of the change is 9.0 employees! In examining changes in employment between the two survey waves it is important to recall that noise-dominated data regress toward means. ...
The 21 restaurants with the smallest employment averaged only 7.5 employees at the time of the baseline survey. In the interval between the first and second interviews, employment increased for every one of these restaurants, with an average gain of 5.4 employees---a 71 % increase in 7-8 months. At the opposite extreme, the 22 largest restaurants averaged 47.4 employees at the baseline and employment fell for every one of them in the following 7-8 months, with an average loss of 19.8 employees---a 42% decrease.
- "Reconciling the evidence of Card and Krueger (1994) and Neumark and Wascher (2000)," Olli Ropponen, J. Applied Econometrics, 2011. I don't know if this is any good.
Aaron Mamula, 6/17/2021. A nice replication with links to the original data. No surprises.
Brown, Charles. 1995. “Myth and Measurement: The New Economics of the Minimum Wage: Comment.” Industrial and Labor Relations Review 48, no. 4 (July): 828-830.
Deere, Donald, Kevin M. Murphy, and Finis Welch. 1995. “Employment and the 19901991 Minimum-Wage Hike.” American Economic Review Papers and Proceedings 85, no. 2 (May): 232-237.
Freeman, Richard B. 1995. “Myth and Measurement: The New Economics of the Minimum Wage: Comment.” Industrial and Labor Relations Review 48, no. 4 (July): 830-834.
Hamermesh, Daniel S. 1995. “Myth and Measurement: The New Economics of the Minimum Wage: Comment.” Industrial and Labor Relations Review 48, no. 4 (July): 835-838.
Kim, Taeil, and Lowell J. Taylor. 1995. “The Employment Effect in Retail Trade of California’s 1988 Minimum Wage Increase.” Journal of Business and Economic Statistics 13, no. 2 (April): 175-182.