The Policy Context

On 1 April 1992, New Jersey raised its minimum wage from 2Z5.05 per hour, a 19 percent increase. Pennsylvania's minimum wage remained at the federal level of $4.25. The fast-food industry was directly affected: it employs large numbers of minimum wage workers and its labour costs are a significant fraction of total costs.

Standard competitive labour market theory predicts that a binding minimum wage increase reduces employment: firms substitute away from labour as it becomes more expensive, move up the labour demand curve, and hire fewer workers. This prediction had been supported by a large literature of time-series and cross-sectional studies through the late 1980s, most influentially in a survey by Brown et al.(1982).

The Card–Krueger Design

Card and Krueger(1994) surveyed fast-food restaurants in New Jersey and eastern Pennsylvania by telephone in February–March 1992 (before the wage increase) and in November–December 1992 (after the increase). They collected information on employment (measured in full-time equivalents), wages, prices, and restaurant characteristics.

The DiD estimator compares the change in employment in New Jersey to the change in Pennsylvania: \[ \widehat{\Delta} = (\bar{Y}_{NJ,\text{post}} - \bar{Y}_{NJ,\text{pre}}) - (\bar{Y}_{PA,\text{post}} - \bar{Y}_{PA,\text{pre}}) \] Under the parallel trends assumption — that in the absence of the minimum wage increase, employment in New Jersey and Pennsylvania would have followed the same trend — this estimator identifies the average treatment effect on the treated.

The Parallel Trends Assumption

The parallel trends assumption requires that New Jersey and Pennsylvania were on the same employment trajectory before the treatment. Card and Krueger note that the two states are geographically adjacent, share a border economy, and had similar pre-period trends in employment. They also note that New Jersey restaurants that were already paying above the new minimum wage serve as a within-state control group: if the minimum wage is the cause of any employment change, restaurants unaffected by the binding constraint should show no differential change.

Main Results

Card and Krueger's headline result was that New Jersey fast-food employment increased relative to Pennsylvania after the minimum wage rise. Employment in New Jersey rose from an average of 20.4 full-time-equivalent workers before the increase to 21.0 after; in Pennsylvania it fell from 23.3 to 21.2. The DiD estimate is: \[ \hat{\Delta} = (21.0 - 20.4) - (21.2 - 23.3) = 0.6 - (-2.1) = 2.7 \] This positive and statistically significant estimate of roughly 2.7 full-time-equivalent workers per restaurant was directly contrary to the competitive model prediction.

Within-State Control

As a robustness check, Card and Krueger compared New Jersey restaurants that initially paid below the new minimum wage (and thus had to raise wages) to those that already paid above it (and thus were unaffected). This within-NJ comparison also found no negative employment effect, strengthening the evidence against the competitive prediction.

The Controversy

The paper provoked immediate and intense criticism, particularly from Neumark and Wascher(1992), who used payroll data rather than telephone survey data and found a negative employment effect. The debate crystallised around three methodological questions:

Measurement error. Neumark and Wascher(1992) argued that Card and Krueger's telephone survey data contained substantial measurement error, while payroll records were more reliable. Card and Krueger(1994) responded with a later dataset cross-validated against payroll records, which supported their original findings.

The comparison group. Is Pennsylvania a valid counterfactual for New Jersey? The two states differ in many dimensions beyond the minimum wage, and the labour markets on either side of the state border are not identical. Card and Krueger argue that the border region is economically integrated enough to support the parallel trends assumption; critics are less convinced.

Functional form. Employment is measured in full-time equivalents, which aggregates hours and bodies in a particular way. Subsequent studies have examined different margins — hours, employment rates, number of workers — and reached mixed conclusions.

Replications and the Subsequent Literature

The minimum wage debate has generated a large subsequent literature. A key development has been the use of border discontinuity designs, which compare counties on either side of state minimum wage boundaries, motivated by the Card–Krueger idea that adjacent labour markets serve as natural controls.

Dube et al.(2010) implement this design using a comprehensive panel of county-pair data across the United States, exploiting the variation created by many state minimum wage changes between 1990 and 2006. They find small to zero employment effects, consistent with Card and Krueger. Their key innovation is to control for local labour market trends by including county-pair fixed effects, which absorbs time-varying local shocks that might otherwise confound the comparison.

Neumark and Wascher(2014), by contrast, use a different methodological approach and find negative effects, particularly for young and low-skilled workers. The debate between these positions reflects genuine disagreement about the appropriate control group, not just about the data.

More recently, Cengiz et al.(2019) use a bunching estimator that aggregates many state minimum wage increases and examines the distribution of wages around the minimum. They find that minimum wage increases raise wages at the bottom of the distribution without detectable disemployment, consistent with Card and Krueger.

Methodological Legacy

Regardless of where one stands on the minimum wage question, Card and Krueger (1994) has an unambiguous methodological legacy.

Legitimising DiD in applied economics. Before 1994, DiD was known but was not a primary identification strategy in high-profile labour economics papers. The paper's prominence — it was published in the American Economic Review and garnered enormous attention — established DiD as a credible tool for policy evaluation.

The value of natural experiments. Card and Krueger made explicit the idea that policy changes can be exploited as natural experiments. The New Jersey minimum wage increase was not designed as an experiment, but the accident of geography — Pennsylvania as a neighbouring unaffected state — provided a control group.

Challenging received theory. The paper demonstrated that empirical evidence could directly challenge textbook predictions, and that clean identification was sufficient to overturn a theoretical consensus. This raised the evidentiary standard for labour economics claims.

The parallel trends assumption as a central object. The criticism that Pennsylvania is not a valid control for New Jersey forced the profession to think carefully about what conditions the parallel trends assumption requires and how to test it. This conversation ultimately led to the modern literature on pre-trend testing, sensitivity analysis (Rambachan and Roth(2023)), and the border discontinuity design.

The 2021 Nobel Prize

In 2021, David Card was awarded the Nobel Memorial Prize in Economic Sciences (together with Joshua Angrist and Guido Imbens) partly for his contributions to the analysis of labour markets, including his work on the minimum wage. The Nobel citation explicitly referenced Card and Krueger (1994) and the natural experiments tradition. The prize was, in many respects, a recognition of the credibility revolution in empirical economics that the paper helped to launch.

Conclusion

Card and Krueger (1994) is a paper worth revisiting not only for its substantive finding about the minimum wage, but for what it demonstrates about the power and the limitations of the difference-in-differences method. It showed that a simple, transparent research design could produce findings that challenged theoretical orthodoxy. It also provoked the criticism that would, over the next thirty years, refine and stress-test the DiD methodology in ways that have made the broader discipline more rigorous. The paper's legacy is visible in every application of the parallel trends assumption that is now submitted to referees who know to ask for pre-trend tests and sensitivity analyses.