Most empirical findings in this hub did not survive replication. Card and Krueger 1994 did. Thirty years of hostile attempts to overturn the New Jersey-Pennsylvania fast-food study --- payroll re-analysis, contiguous-county designs, bunching estimators, 138 minimum-wage events --- have largely confirmed the original counter-textbook conclusion. Here is the rare empirical-economics result that earned a Nobel Prize the long way.

If you have been reading through this hub, you have watched canonical results in psychology, behavioral economics, social science, and macroeconomics get dismantled. Power posing collapsed. Ego depletion failed. Money priming evaporated. The Reinhart-Rogoff 90% debt threshold disintegrated under Excel-error scrutiny. The pattern, by now, is familiar: a high-profile empirical finding gets published, gets cited thousands of times, gets used to justify policy, and then either fails to replicate under preregistered conditions or collapses entirely once independent researchers get access to the data and methods.

A rational reader of this hub might conclude that empirical economics is more or less hopeless --- that the combination of motivated reasoning, publication bias, methodological flexibility, and policy incentive renders most quantitative claims about contested questions essentially worthless. That conclusion would be wrong, and this article is the second anti-example in the hub (alongside the default effect in behavioral economics) intended to make the point.

Because the most controversial empirical-economics finding of the past 35 years --- the claim that modest minimum-wage increases do not measurably reduce employment in the affected low-wage labor markets --- has survived 30 years of the most adversarial replication scrutiny any policy-relevant economics result has ever faced. It has been attacked with payroll-record re-analysis. It has been attacked with state-trend-corrected re-analysis. It has been attacked with the full machinery of methodological economics by researchers who very much wanted the result not to be true. And the most rigorous current evidence --- spanning 138 state-level minimum-wage changes between 1979 and 2016, multiple identification strategies, and several research groups --- still says approximately what David Card and Alan Krueger originally said in the American Economic Review in 1994.

This is the canonical anti-example of an empirical economics finding that has survived its own replication crisis. David Card was awarded half of the 2021 Nobel Memorial Prize in Economic Sciences “for his empirical contributions to labour economics,” and the minimum-wage work was a central part of the citation. The methodological revolution that work helped launch --- the natural-experiment / quasi-experimental design tradition in labor economics --- is itself one of the most important methodological reforms in the social sciences of the past four decades, and it grew partly out of this single contested study and the multi-decade scrap that followed it.

So here is the case for what the Card-Krueger result actually shows, what has happened to it under three decades of adversarial replication, and what it tells you about the difference between an empirical finding that holds up and an empirical finding that does not.

What Card And Krueger 1994 Actually Tested

The original paper is Card, D., & Krueger, A. B. (1994). “Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania.” American Economic Review, 84(4), 772—793. The full preprint is available at davidcard.berkeley.edu/papers/njmin-aer.pdf and the NBER working-paper version is at nber.org/papers/w4509.

The setup was a quasi-experiment. On April 1, 1992, New Jersey raised its state minimum wage from $4.25 per hour (the federal floor) to $5.05 per hour --- a roughly 19% increase that took the New Jersey floor approximately a dollar above the floor in neighboring Pennsylvania, where the minimum wage remained at $4.25. Eastern Pennsylvania, just across the Delaware River, offered an economically and demographically similar comparison region whose minimum wage was unchanged. The classic conditions for a difference-in-differences identification were in place: a treatment region experiencing the policy change, a control region not experiencing it, and a baseline period before the change against which post-change outcomes could be compared.

Card and Krueger surveyed 410 fast-food restaurants --- Burger King, KFC, Wendy’s, and Roy Rogers --- in New Jersey and eastern Pennsylvania. They surveyed each restaurant twice: once in February-March 1992 (before the New Jersey increase) and once in November-December 1992 (after it). The outcome variables of interest were full-time-equivalent employment per restaurant, hours worked, starting wages, prices of items on the menu, and store characteristics. The choice of fast food was deliberate. Fast-food restaurants employ many minimum-wage workers, so the New Jersey increase plausibly bound, and the industry has standardized job tasks across stores in different states, which removes confounds about industry mix.

The textbook microeconomic prediction was unambiguous. If the labor market for fast-food workers in New Jersey was approximately competitive and the pre-change wage was approximately the market-clearing wage, then a 19% minimum-wage increase --- well above market-clearing --- should reduce equilibrium employment. The magnitude of the predicted reduction depended on the labor-demand elasticity, but the sign was not in question. Standard intermediate-microeconomics textbooks of the era used the minimum wage as a canonical example of price-floor disemployment effects.

The result Card and Krueger reported was the opposite. Comparing the change in employment in treated New Jersey restaurants to the change in employment in untreated Pennsylvania restaurants, they found no reduction in New Jersey employment relative to Pennsylvania --- and in their preferred specifications, a modest relative increase in New Jersey employment, on the order of 13%. They reported essentially no effect on hours worked, a small positive effect on store openings, and an increase in starting wages at affected New Jersey stores roughly consistent with the policy change. The price of fast-food items rose modestly in New Jersey relative to Pennsylvania --- consistent with employers passing some of the higher wage cost to consumers --- but the employment-loss prediction simply did not materialize in the data.

This was, in 1994, an empirically explosive finding. It contradicted half a century of textbook microeconomics teaching. It contradicted decades of time-series estimates that had typically found minimum-wage elasticities for teen employment in the range of -0.1 to -0.3. And it directly challenged the policy framing under which minimum-wage increases were politically opposed: that you could not help low-wage workers by raising the wage floor because the unemployment effect would dominate the wage gain.

A year later, Card and Krueger published the book Card, D., & Krueger, A. B. (1995). Myth and Measurement: The New Economics of the Minimum Wage. Princeton University Press. (press.princeton.edu/books/paperback/9780691169125/myth-and-measurement). The book extended the New Jersey-Pennsylvania case to a series of additional minimum-wage episodes --- the 1988 California state increase, the 1990-91 federal-floor increases, and an analysis of teen labor markets across states --- and presented a sustained case that the standard textbook prediction was empirically wrong. The book was reviewed harshly in some quarters and favorably in others; the central empirical claim was the same as the 1994 paper.

The Pushback: Neumark And Wascher 2000

The original Card-Krueger paper used telephone-survey data on employment at each restaurant. Critics pointed out that survey self-reports of employment are noisier than administrative payroll records, and that small errors in measured employment could plausibly drive a difference-in-differences result of this magnitude.

The most influential rebuttal was Neumark, D., & Wascher, W. (2000). “Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania: Comment.” American Economic Review, 90(5), 1362—1396 (aeaweb.org/articles?id=10.1257/aer.90.5.1362). The corresponding NBER working paper is at nber.org/papers/w5224.

Neumark and Wascher obtained actual payroll records from a sample of New Jersey and Pennsylvania fast-food restaurants --- not survey reports of employment, but the underlying administrative records that the restaurants themselves used for payroll administration. They argued that this was a more reliable measurement of employment than the original Card-Krueger telephone-survey data, which had been collected by research-assistant phone calls and depended on whoever happened to answer the phone reporting employment accurately.

Their conclusion went the other way. Using their payroll-record sample, Neumark and Wascher estimated that the New Jersey minimum-wage increase reduced employment in New Jersey fast-food restaurants by approximately 4.6% relative to Pennsylvania, statistically significant at the 5% level. This implied an elasticity of fast-food employment with respect to the minimum wage of approximately -0.24 --- a number close to the prior time-series consensus and to the textbook prediction.

The Neumark-Wascher comment was a serious challenge. The implicit argument was that the Card-Krueger result was a measurement artifact of using survey data rather than administrative data, and that the textbook prediction had been correct all along. If Neumark and Wascher were right, the central Card-Krueger finding would have been a methodological cautionary tale --- the kind of story this hub is full of --- rather than a paradigm shift in labor economics.

It is worth noting that Neumark and Wascher’s payroll sample was not a clean random sample of the restaurants in the original Card-Krueger study. It was a convenience sample assembled with assistance from the Employment Policies Institute, an industry-funded research organization with strong views on minimum-wage policy. This does not by itself disqualify the analysis --- adversarial data collection can be informative regardless of who funds it --- but it did mean the Card-Krueger team had grounds to question how the Neumark-Wascher sample had been assembled and whether it was representative of the original population.

The Reply: Card And Krueger 2000

Card and Krueger replied in the same issue of the AER with Card, D., & Krueger, A. B. (2000). “Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania: Reply.” American Economic Review, 90(5), 1397—1420 (aeaweb.org/articles?id=10.1257/aer.90.5.1397).

Their reply did the methodologically clean thing: rather than re-defend the original survey-based estimates against a payroll-based critique, they went and got their own administrative-payroll dataset. Specifically, they used the Bureau of Labor Statistics ES-202 data --- the universal unemployment-insurance payroll records that every employer in the United States is required to submit and that cover essentially the full population of legal-sector employment in any covered industry. The ES-202 data, by construction, is not subject to survey-response error. It is what employers actually paid people and reported to the state unemployment-insurance system.

When they re-ran the New Jersey-Pennsylvania difference-in-differences analysis on the ES-202 fast-food employment data for the same period as the original 1992 minimum-wage increase, the answer reproduced the original Card-Krueger finding. There was no statistically significant reduction in New Jersey fast-food employment relative to Pennsylvania around the April 1992 wage increase. Card and Krueger’s summary judgment in the reply was 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” --- essentially the original 1994 conclusion, re-derived from a different and more reliable data source.

The reply also offered a forensic analysis of why the Neumark-Wascher payroll sample might have produced a different answer. The Neumark-Wascher sample was disproportionately drawn from one franchise chain (Burger King) in particular regions, and the time period covered by their data did not align cleanly with the immediate pre/post window around the April 1992 wage change. The Card-Krueger reply argued that these compositional issues plausibly explained the divergent estimate.

The 2000 exchange did not produce a clean intellectual victory for either side. Neumark and Wascher continued to argue that their payroll data was the more reliable source. Card and Krueger countered that the ES-202 universal-payroll data they had used in the reply was at least as reliable and pointed the other way. What the exchange did establish was that the dispute was not going to be resolved by one more analysis of the original 1992 New Jersey episode. It was going to require many more minimum-wage episodes, across many more states and time periods, with several different identification strategies, before anyone could claim a confident view of the underlying labor-market effect.

That is exactly what then happened over the following two decades.

Continued Replication: Dube-Lester-Reich 2010 And The Contiguous-County Design

The next major methodological step was Dube, A., Lester, T. W., & Reich, M. (2010). “Minimum Wage Effects Across State Borders: Estimates Using Contiguous Counties.” The Review of Economics and Statistics, 92(4), 945—964. DOI: 10.1162/REST_a_00039 (direct.mit.edu/rest/article/92/4/945).

The Dube-Lester-Reich contribution was to recognize that the New Jersey-Pennsylvania design generalized. There are 1,182 county pairs in the United States that straddle a state border, and many of those border pairs at various times had different state-level minimum wages on the two sides. Each such pair offered a Card-Krueger-style natural experiment: nearly identical local labor markets, often the same metropolitan area, divided arbitrarily by a state line that determined the legal wage floor. Pooling all those pairs over 1990 to 2006, Dube, Lester, and Reich constructed a difference-in-differences estimator that controlled for local labor-market conditions much more aggressively than the state-by-state time-series approaches that had been the prior literature standard.

Their result was consistent with Card-Krueger: no statistically significant adverse employment effect of state minimum-wage increases on restaurant or other low-wage employment in the contiguous-county comparison. They argued that prior time-series studies had produced spurious negative employment effects because they failed to control for the substantial spatial heterogeneity in employment trends across U.S. regions. Once you forced the identification to come from within-pair, near-neighbor comparisons, the disemployment estimate fell to approximately zero.

A back-and-forth ensued. Neumark and his collaborators argued that the contiguous-county design absorbed too much identifying variation and that the resulting null result reflected statistical power loss rather than a true zero effect. Dube and his collaborators argued that the prior cross-state designs were biased by uncontrolled regional trends and that the contiguous-county design was the better identification strategy.

The most thorough adjudication was Allegretto, S., Dube, A., Reich, M., & Zipperer, B. (2017). “Credible Research Designs for Minimum Wage Studies: A Response to Neumark, Salas, and Wascher.” ILR Review, 70(3), 559—592 (journals.sagepub.com/doi/abs/10.1177/0019793917692788). Allegretto, Dube, Reich, and Zipperer used data from 1979 to 2014 and showed that the apparent disemployment effect in conventional cross-state specifications was largely an artifact of differential pre-existing trends across U.S. states. When they applied a data-driven LASSO procedure designed to optimally adjust for state-specific trends, the estimated minimum-wage employment elasticity for teen employment fell to approximately -0.01 --- effectively zero. A synthetic-control analysis produced an elasticity of approximately -0.04, also near zero.

By the mid-2010s, the methodological dispute had a structure. There was a research camp --- Card, Krueger, Dube, Reich, Zipperer, Cengiz, and collaborators --- arguing for natural-experiment designs that exploited within-region variation and that consistently found small-to-zero employment effects of minimum-wage increases in the range studied. There was a counter-camp --- Neumark, Wascher, and collaborators --- arguing that the natural-experiment designs absorbed too much variation and that broader specifications still showed meaningful disemployment effects. The dispute was not over the existence of the Card-Krueger result; it was over which identification strategy was the right one for estimating minimum-wage effects.

The Cengiz 2019 Synthesis: 138 Events, Bunching, And A New Methodological Approach

The methodological capstone of the post-Card-Krueger replication literature is Cengiz, D., Dube, A., Lindner, A., & Zipperer, B. (2019). “The Effect of Minimum Wages on Low-Wage Jobs.” The Quarterly Journal of Economics, 134(3), 1405—1454. DOI: 10.1093/qje/qjz014 (academic.oup.com/qje/article-abstract/134/3/1405/5484905).

The Cengiz design is methodologically distinct from anything that had come before, and it gets around the central dispute between the cross-state and contiguous-county camps. Rather than estimate the overall employment effect of minimum-wage increases using difference-in-differences on aggregate employment, the Cengiz approach exploits a feature of the wage distribution: a binding minimum-wage increase mechanically creates “bunching” of workers at the new minimum, and a “missing mass” of workers at wages below the new minimum who used to exist but no longer do. By comparing the size of the bunching at the new minimum to the size of the missing mass below it, Cengiz and colleagues can directly estimate how many of the previously-below-minimum jobs got bumped up to the new minimum (the wage-gain channel) versus how many were eliminated entirely (the disemployment channel).

The dataset covers 138 prominent state-level minimum-wage changes in the United States between 1979 and 2016. The bunching design effectively turns each minimum-wage change into its own internally controlled experiment, with the rest of the wage distribution serving as the control.

The headline finding: “the overall number of low-wage jobs remained essentially unchanged over the five years following the increase.” The bunching at the new minimum was almost entirely offset by the missing mass just below it, with very little net employment loss. The implied elasticity of low-wage employment with respect to the minimum wage was small and statistically indistinguishable from zero in the range of minimum-wage changes studied. The wage gain to affected workers was real and substantial; the disemployment offset was small.

The Cengiz design is methodologically important because it does not rely on between-state comparisons at all. The identification comes from the structure of the wage distribution within each state around each minimum-wage event. The cross-state-control concerns that animated the earlier Neumark-Wascher critique do not apply, because no cross-state comparison is being made. And the result still says approximately what Card and Krueger said in 1994: in the range of minimum-wage changes the United States has actually implemented over the past 35 years, employment effects on the affected low-wage labor markets are small to zero.

This is the closest the literature has come to a definitive methodological adjudication. The result has been broadly accepted by the labor-economics mainstream, and the methodology has been extended to additional country contexts (the United Kingdom’s National Living Wage in particular) with similar results.

What Distinguishes This From Reinhart-Rogoff

A useful way to understand why Card-Krueger survived and many other empirical-economics findings did not is to contrast it with the Reinhart-Rogoff 2010 Excel-error episode, which is the canonical empirical-economics failure of the past two decades.

Reinhart and Rogoff claimed that government debt above 90% of GDP was associated with a discrete decline in average growth rates. The claim was used to justify austerity policy across multiple economies. When Herndon, Ash, and Pollin re-analyzed the underlying data in 2013, they found a spreadsheet coding error, selective country exclusions, and a non-standard data weighting that, when corrected, eliminated the discrete 90% threshold. The Reinhart-Rogoff claim collapsed because the original analysis was fragile to ordinary methodological scrutiny --- the underlying pattern was not robust to obvious choices about how to aggregate the data, and the headline number was an artifact of a small number of analyst decisions.

The Card-Krueger finding has been subjected to substantially more adversarial scrutiny than Reinhart-Rogoff ever was. The original 1994 survey-based result was attacked. The 2000 payroll-based reply was attacked. The 2010 contiguous-county design was attacked. The 2017 state-trend-correction was attacked. The 2019 bunching analysis was attacked. The attackers included substantial industry-funded research dollars, multiple academic research groups, and decades of motivated empirical effort by economists with strong priors that the textbook prediction had to be right. The result has been re-derived from at least four substantially different identification strategies, using survey data, ES-202 administrative payroll data, contiguous-county difference-in-differences, state-trend-corrected panel data, and bunching analysis on the wage distribution.

Each successive replication came at the question from a different methodological angle, and each one returned an answer broadly consistent with the original. This is what survival under hostile replication looks like. It is not that the result is unanimous or beyond critique --- there are economists who still dispute it, and there are minimum-wage levels far above what has been studied where the effect could plausibly be different. But the core finding --- that modest minimum-wage increases in the range studied have small-to-zero employment effects on low-wage jobs --- has emerged from 30 years of methodological combat largely intact.

Reinhart-Rogoff did not. The 90% threshold did not survive its first serious replication attempt by an independent group. The difference is not a difference in topic prestige or author reputation; both papers were published by elite economists in top venues and were used to justify major policy positions. The difference is the underlying empirical pattern. One was real and robust to alternative specifications. The other was an artifact.

The 2021 Nobel Prize And The Credibility Revolution

In October 2021, the Royal Swedish Academy of Sciences awarded the Nobel Memorial Prize in Economic Sciences jointly to David Card, Joshua Angrist, and Guido Imbens. Card received half the prize “for his empirical contributions to labour economics”; Angrist and Imbens shared the other half “for their methodological contributions to the analysis of causal relationships.” The official announcement and background materials are at nobelprize.org/prizes/economic-sciences/2021/summary/.

The Nobel citation explicitly named the minimum-wage work as central to the prize. The committee’s framing was that Card’s empirical studies of the labor market --- minimum wages, immigration, and education --- demonstrated what natural-experiment designs could deliver in policy domains where direct experimental manipulation was impossible. The methodological contributions of Angrist and Imbens --- the formalization of instrumental-variables estimation under heterogeneous treatment effects, the local-average-treatment-effect framework --- provided the statistical foundation for the kinds of natural-experiment designs Card had pioneered.

The combined effect of this body of work has been what economists call the “credibility revolution” in empirical labor economics. The standard for what counts as a credible empirical claim shifted, over the 1990s and 2000s, from time-series correlations with weak identification to natural-experiment designs that exploit policy changes, geographic discontinuities, or other exogenous variation to identify causal effects. The Card-Krueger New Jersey-Pennsylvania design was one of the most influential examples of this methodological shift, and the multi-decade scrap over the minimum-wage finding was one of the most rigorous tests of whether the credibility-revolution standard actually produced more reliable knowledge than the prior approach.

The Nobel committee’s answer was yes. The methodological standard that Card, Angrist, Imbens, Krueger (who died in 2019 and was therefore not eligible for the prize, which is not awarded posthumously), and their many collaborators helped establish has substantially improved the average reliability of empirical-economics findings on policy questions. The Card-Krueger minimum-wage result is one of the showcases of what that methodology can deliver when the empirical pattern is robust.

What Is Honest To Say About Minimum-Wage Effects Now

A scrupulous summary of the current state of evidence, after 30 years of replication, would say something like this.

For modest minimum-wage increases in the range studied --- roughly speaking, increases that take the minimum wage to no more than approximately 55-60% of the local median wage --- the best-identified evidence consistently shows small-to-zero effects on overall low-wage employment, statistically significant wage gains for affected workers, and modest pass-through to prices in affected industries. This is what the Cengiz 2019 synthesis says, what the Dube 2010 and Allegretto 2017 papers say, and what the Card-Krueger 2000 administrative-data analysis said. The original 1994 finding was approximately correct, and the methodological elaborations since have largely confirmed it.

For very large minimum-wage increases --- those approaching or exceeding the local median wage --- the empirical evidence is genuinely thinner, because the United States has not implemented many such increases at scale and because the limited evidence from very high minimum-wage cases (Seattle’s $15 increase in 2014-2017 in particular) has been methodologically contested in ways that are not fully resolved. It is honest to say that the Card-Krueger consensus applies to the range of minimum-wage changes the U.S. has actually implemented, and that extrapolation to substantially higher minimum-wage floors is genuinely uncertain. The textbook prediction --- that at some sufficiently high level, a binding minimum wage will produce meaningful disemployment effects --- may very well be true; what the empirical evidence shows is that the threshold at which this happens is higher than the textbook simple-supply-and-demand model implies, not that no such threshold exists.

There are still labor economists who dispute the central finding. Neumark and various collaborators continue to publish analyses arguing for negative employment effects in particular specifications. The dispute has not been settled in the sense that all serious economists agree on a single number. What has happened is that the median labor-economics view, weighted by methodological rigor and replication record, has shifted decisively toward the Card-Krueger position over the past 30 years, and the Nobel committee’s 2021 decision is one indicator of how that consensus has crystallized.

It is worth being explicit that the original textbook prediction was not crazy. The simple competitive-labor-market model in which the minimum wage prices low-productivity workers out of employment is a coherent and defensible theoretical framework, and it had empirical support from time-series studies that produced consistent (if small) negative elasticity estimates for decades before Card-Krueger. The reason the empirical view shifted is that better-identified studies produced different answers. The labor-economics profession updated --- imperfectly, slowly, with substantial internal dispute, but decisively --- toward a more empirically grounded view of what the labor market actually does in the range of minimum-wage policies actually implemented. This is what a healthy empirical science looks like.

What This Anti-Example Tells Us About Robust Empirical Economics

Stepping back from the specific minimum-wage debate, the Card-Krueger story illustrates several features of an empirical-economics finding that survives hostile replication --- features that, if absent, should make a reader more skeptical of any empirical claim.

Multiple independent identification strategies converging on the same answer. The Card-Krueger result has been re-derived from survey data, administrative payroll data, contiguous-county difference-in-differences, state-trend-corrected panels, and bunching analysis of the wage distribution. These are methodologically distinct identification approaches, drawing on different parts of the underlying data, with different threats to identification. They have nonetheless produced broadly consistent answers. This is much stronger evidence than any single study could be, because any single study could be undermined by the specific identification assumptions it relies on. A finding that survives across identification approaches is not a finding produced by any one set of assumptions.

Adversarial replication by researchers who want the result not to be true. Some of the most important replication work on minimum wages was done by Neumark, Wascher, and collaborators, who had strong priors against the Card-Krueger finding and were funded in part by sources sympathetic to the same priors. The fact that adversarial replication with substantial methodological skill failed to demolish the result is much stronger evidence than friendly replication would be. Adversarial replication is the gold standard for credibility in any empirical literature, and the minimum-wage literature has had decades of it.

Data and code sharing. The Card-Krueger team made the original 1994 dataset publicly available, which is what allowed the subsequent decades of replication and re-analysis. The Neumark-Wascher payroll data was at least partially shareable. The Dube and Cengiz datasets are publicly accessible. This kind of open-data infrastructure is what makes hostile replication possible, and it is one of the major differences between findings that survive (Card-Krueger) and findings that collapse (Reinhart-Rogoff, which depended on a spreadsheet error that was only catchable because the spreadsheet was eventually shared).

Methodological innovation as a byproduct. The replication debate around Card-Krueger did not just confirm or fail to confirm a particular empirical answer. It produced the contiguous-county research design, the bunching estimator, the synthetic-control approach to state policy analysis, and a generation of labor economists trained in natural-experiment methods. The empirical answer is one output of the methodological debate; the improved methodological toolkit is another. A finding that catalyzes methodological progress is doing more useful work than a finding that just sits as an empirical claim.

Coherent mechanism story, even when the textbook story is wrong. The Card-Krueger result requires some account of why modest minimum-wage increases do not produce the disemployment the textbook predicts. There are several candidate stories: employer monopsony power in low-wage labor markets, demand-side macroeconomic effects of higher wages, productivity effects of better-paid workers, reduced turnover offsetting higher unit labor costs. None of these is conclusively established as the dominant mechanism. But the existence of multiple plausible mechanisms, each consistent with the empirical pattern, makes the finding less dependent on any one theoretical commitment. Compare to Reinhart-Rogoff, where the 90% debt threshold required some specific story about how debt above that exact level became uniquely growth-damaging, and no coherent mechanism story for the discrete threshold ever developed.

What This Means For CEOs And Strategists Evaluating Economics Claims

Most of the empirical-economics claims that show up in business strategy, policy advocacy, and management consulting do not have the Card-Krueger profile. They have something closer to the Reinhart-Rogoff profile: a single high-prestige paper, a striking headline number, and a policy implication that is politically convenient for some constituency. When you encounter an empirical-economics claim in the wild, run it through the calibration checklist below.

How many independent identification strategies have produced the same answer? A finding that has been re-derived from multiple methodologically distinct designs is meaningfully different from a finding that exists in one paper. The Card-Krueger result has been re-derived four or five different ways. Most claims you will encounter have not.

Has the result survived adversarial replication? A finding that has been attacked by researchers with strong priors against it, and survived, is much more credible than a finding that has only been studied by sympathetic researchers. The minimum-wage literature has had three decades of well-funded adversarial replication. Most of the empirical claims in your consulting deck have had essentially none.

Are the data and methods publicly available? A finding produced from a dataset and codebase that anyone can re-run is structurally different from a finding produced from proprietary data the authors will not share. The first kind can be checked; the second kind cannot. Default to skepticism for findings of the second kind, especially when they are politically or commercially convenient.

Does the result depend on a specific identification assumption that nobody else has been able to validate? A finding that requires one particular instrumental variable, one particular structural model, or one particular sample restriction to come through is meaningfully different from a finding that holds across reasonable variations in those choices. Single-assumption findings are fragile; multi-assumption-robust findings are not.

Is the result politically or commercially convenient for whoever is presenting it? This is not by itself disqualifying --- some politically convenient findings are also empirically robust --- but it should increase your demand for evidence that the result has survived adversarial replication. Findings convenient to a constituency that has the resources to fund their own confirmatory research should be subjected to additional scrutiny, not less.

Has the finding generated methodological progress, or has it just sat as a contested empirical claim? Findings that catalyze methodological innovation tend to be more empirically substantive than findings that do not. The minimum-wage debate produced the contiguous-county design and the bunching estimator; the Reinhart-Rogoff debate produced a few replication papers and a cautionary tale about spreadsheet workflow.

The general rule for a CEO or strategist evaluating empirical-economics claims is to weight findings by their replication profile, not by their headline prestige. A finding from a Nobel laureate published in a top-five economics journal that has not been independently replicated is structurally less reliable than a finding that has been re-derived from multiple methodologies by adversarial research groups, even if the latter finding is less famous. Most of what you will see in business contexts is the first kind. The Card-Krueger result is one of the relatively few examples of the second kind in policy-relevant empirical economics, and it earned that status the long way.

Sources

Browse the full Replication Crisis Hub for other empirical-research findings, including:

FAQ

Does any minimum wage increase have no employment effects?

No. The Card-Krueger and subsequent replication literature shows that the modest minimum-wage increases the United States has actually implemented over the past 35 years --- typically taking the minimum wage to no more than approximately 55-60% of the local median wage --- have produced small-to-zero employment effects in affected low-wage labor markets. Larger minimum-wage increases, or increases approaching or exceeding the local median wage, are not well-covered by this evidence base. The honest claim is that modest increases in the range studied have small effects, not that any increase of any size would.

What about the $15 minimum wage debates?

The $15 minimum wage at the federal level would represent a substantially larger increase than any of the state-level changes covered by the Cengiz 2019 evidence base, particularly in lower-wage states where $15 would push the minimum wage close to or above the local median wage. The Seattle $15 implementation has been studied (notably by Jardim and colleagues at the University of Washington, with contested results) and the conclusions there have not converged in the way the broader minimum-wage literature has. It is honest to say that the empirical evidence for modest minimum-wage increases does not necessarily extrapolate to $15 in lower-wage states. The evidence is somewhat stronger for $15 in high-wage cities where the resulting minimum-to-median ratio remains in the range covered by prior research.

What does textbook microeconomics actually say?

The simple competitive-labor-market model predicts that a binding minimum wage above the market-clearing wage will reduce equilibrium employment, with the magnitude depending on the demand and supply elasticities. This is the textbook prediction the original Card-Krueger result contradicted. More sophisticated models --- monopsony, search frictions, efficiency wages --- can generate small or zero employment effects of modest minimum-wage increases without contradicting basic economic theory. The empirical finding has shifted the field’s view toward the more sophisticated models for low-wage labor markets, while the simple competitive model remains a useful baseline.

Is there any consensus on minimum wage among economists?

There is now broad (though not unanimous) consensus among labor economists that modest minimum-wage increases in the range studied have small employment effects --- substantially smaller than older time-series studies had estimated, and not large enough to dominate the wage gain for affected workers. There is much less consensus on what happens at substantially higher minimum-wage levels, on the long-run effects on labor-force composition and firm entry, or on the optimal level of the minimum wage from a welfare standpoint. The 2021 Nobel committee’s framing reflects this: the credibility-revolution evidence on minimum wages is real and durable, but it bounds the range over which we can confidently make claims.

Why did this result survive when so many others did not?

Multiple independent identification strategies (survey, administrative payroll, contiguous-county, state-trend-corrected, bunching) converged on the same answer. Adversarial replication by researchers with strong priors against the finding (Neumark, Wascher, industry-funded research) failed to demolish it. The data and methods have been publicly available throughout, allowing 30 years of independent re-analysis. The dispute generated methodological innovation (contiguous-county design, bunching estimator) that has paid off in adjacent literatures. And the result has multiple plausible mechanism stories (monopsony, search frictions, productivity effects) consistent with it, rather than depending on one specific theoretical commitment.

How big is the wage gain for affected workers?

Approximately commensurate with the policy change. A 10% increase in the legal minimum wage produces approximately a 10% wage gain for workers previously paid at or just above the old minimum, with modest spillovers up the wage distribution. The Cengiz 2019 estimates suggest that average earnings for affected workers rise by close to the full policy increase, with very little of the gain being offset by reduced hours or eliminated jobs. This is the practical implication of the small-employment-effect finding: minimum-wage increases in the studied range function approximately as redistributive transfers from consumers and employers to low-wage workers, not as job-destroying interventions.

Should I trust empirical economics claims in general?

You should trust empirical economics claims to the extent they have the Card-Krueger profile: multiple independent identification strategies, adversarial replication, data and method sharing, methodological coherence. Most empirical economics claims you will encounter in business and policy contexts do not have this profile --- they exist in one paper, have not been independently replicated, depend on data the authors will not share, and are politically or commercially convenient for whoever is citing them. Default to skepticism for the second kind, and apply the calibration checklist in the strategist section above to anything you are tempted to act on.

What’s the single best version of the current evidence I should read?

For a quick overview, the Cengiz, Dube, Lindner, and Zipperer 2019 paper in the Quarterly Journal of Economics is the methodological capstone of the post-Card-Krueger literature and the best single reference for current evidence on minimum-wage employment effects. For the original empirical case, the Card and Krueger 1994 AER paper and the 2000 reply with administrative-payroll data are the foundational documents. For the broader methodological context, the 2021 Nobel committee’s background materials at nobelprize.org provide an authoritative summary of why the credibility-revolution approach has earned the field’s recognition.

replication-crisis card-krueger minimum-wage economics-research evidence-evaluation
Share this article
LinkedIn (opens in new tab) X / Twitter (opens in new tab)
Atticus Li

Experimentation and growth leader. CXL-certified CRO practitioner, Mindworx-certified behavioral economist (1 of ~1,000 worldwide). 200+ A/B tests across energy, SaaS, fintech, e-commerce, and marketplace verticals.

About LinkedIn Newsletter