Datasets:
PaulPauls
/
openwebtext-sentences

Dataset card Data Studio Files Community
2
text
stringlengths
1
100k
Two approaches to studying the minimum wage The empirical literature on the impact of the minimum wage is large, but much of it (and all important recent studies) can be classified into one of two categories: one, studies that match and compare cases involving an increase in the minimum wage with a similar control group, and two, studies that do not match cases of a minimum-wage increase to a similar control group.
This distinction is only one of many possible ways of thinking about the empirical literature, but it is critical for answering the question of who is right about the employment effects of the minimum wage.
Matching studies Analyses of the minimum wage that use matching first received wide attention with David Card and Alan Krueger’s 1994 paper on an increase in New Jersey’s state minimum wage from $4.25 to $5.05.
Card and Krueger were concerned with distinguishing changes in employment at fast food restaurants that would have happened anyway from changes occurring in response to the minimum-wage increase.
Their solution was to use comparable restaurants in Pennsylvania immediately across the border from New Jersey as a control group of establishments operating in a similar environment, but not subject to the minimum-wage increase.
These Pennsylvania establishments provided a baseline for determining what would have happened in New Jersey if the minimum wage had remained constant.
Deviation from that baseline in the New Jersey restaurants could thus be safely attributed to the minimum wage.
A true experimental design would have randomly assigned increases in the minimum wage in order to control for alternative influences, but in the absence of random assignment the authors identified the next best alternative: a close match.
The Card and Krueger study concluded that there was no evidence that the minimum-wage increase in New Jersey reduced employment in that state relative to the comparison group of Pennsylvania restaurants.
Criticisms of the quality of the study’s phone survey data were raised at the time, which led the authors to analyze more reliable administrative payroll data from New Jersey and Pennsylvania.
Card and Krueger (2000) confirmed the original finding that the minimum-wage increase in New Jersey had no discernable employment effect.
The matching approach pioneered by Card and Krueger has been applied with increasing sophistication and stronger data sources than the initial phone survey data in the 20 years since the New Jersey analysis.
The most notable advance in matching has been in the work of Arindrajit Dube with several coauthors, which uses counties that neighbor each other across state borders as control cases.
Rather than a restricted analysis of one state’s minimum-wage increase, Dube, Lester, and Reich (2010) compare every pair of neighboring counties along every state border in the country (similar study designs are used in other papers by Dube and his colleagues).
By exploiting variation in the minimum wage across the country and over the course of 16 years, this research estimates minimum-wage effects from a larger sample than earlier matching studies, and produces estimates that are more representative of the typical response to a minimum-wage increase and not the special circumstances of a particular local labor market.
Dube and his colleagues consistently find no evidence for reduced employment as a result of regular increases in the minimum wage using the county pair match.
In fact, even before using county pairs, as Dube, Lester, and Reich (2010) add increasingly more precise geographic matching into their models, the negative impact of the minimum-wage increase identified in the nonmatching literature (discussed in more detail below) gradually evaporates.
Table 1 reports Dube, Lester, and Reich’s (2010) estimates of the percentage change in employment resulting from a percentage change in earnings as a result of an increase in the minimum wage.
The authors analyze two different samples of employment data: one that includes all counties (the first column), and one that includes pairs of neighboring counties (the second column), with county pair matching performed on the latter sample.
Table 1 Percentage change in employment for each percentage change in earnings due to a change in the minimum wage All county sample County pair sample No matching -0.784* -0.482** No matching, control for Census division differences -0.114 — No matching, control for state differences 0.183 — No matching, control for MSA differences 0.211 — County-level matching — 0.079 * Statistically significant at the 10 percent level.
** Statistically significant at the 5 percent level.
Source: Estimates drawn from Dube, Lester, and Reich (2010), Table 2 (this is not a reproduction of their Table 2) Share on Facebook Tweet this chart Embed Copy the code below to embed this chart on your website.
Download image The first row in Table 1, which presents results when no matching is done, is representative of most study designs before Dube, Lester, and Reich (2010), and many since.
When no matching is done, the minimum-wage increase is estimated to have a negative effect.
However, as the comparison is increasingly narrowed to more similar counties, first in the same Census division, then the same state, then the same metropolitan statistical area (MSA), the statistically significant negative effect of the minimum-wage increase is eliminated.
In the analysis that uses actual pair-matching of bordering counties to construct a comparison group (the last row), the higher minimum wage has an estimated positive effect on employment.
However, because this result is statistically insignificant it cannot be statistically distinguished from a finding that the minimum wage has no effect on employment.
In any case, the stronger designs that use matching strategies clearly contradict the theory that minimum-wage increases reduce employment.
Other examples of this approach include Addison, Blackburn, and Cotti (2009; 2012), which have conclusions that are similar to Dube, Lester, and Reich (2010) and other matching studies.
One possible critique is that by over-parameterizing (i.e., adding too many controls to) their models, Dube, Lester, and Reich (2010) are mistakenly attributing true employment-discouraging effects of minimum-wage increases to other variables in their model, or that statistical significance is lost due to the difficulty of estimating such a complex model.
However, the authors point out that these fears can be easily dismissed by comparing estimates of the impact of the minimum wage on employment with estimates of the impact on earnings.
Only the estimate of the impact on employment becomes positive—and loses statistical significance—as more rigorous matching strategies are introduced.
The effect of the minimum wage on earnings stays consistent across these models.
Since the same statistical model with the same risks of over-parameterization is being used regardless of the dependent variable (earnings in one case, employment in the other), the case that specification problems are driving the result is harder to justify.
There are many different explanations for the lack of substantial disemployment effects in matching studies.
One suggestion is that employers exercise “monopsony power,” or bargaining power associated with being one of a small population of buyers in a market (an analog to the monopoly power exercised by sellers).
Just as a monopoly will not reduce its output in response to an imposed price reduction, a monopsonist can absorb a price increase (such as a minimum-wage increase) without reducing demand for workers.
Although such theoretical explanations are possible, a more straightforward argument is that an increase in the minimum wage does not have a disemployment effect because the increased labor costs are easily distributed over small price or productivity increases, or because fringe benefits are cut instead of employment levels.
Less work has been done on the impact of the minimum wage on these outcomes than on the employment impact.
Alternatively, disemployment effects might be avoided due to reduced fixed hiring costs as a result of lower turnover.
The most comprehensive and best known matching studies find that a higher minimum wage does not have a negative impact on employment, but this finding is not unanimous.
Some matching studies do find disemployment effects.
For example, Sabia, Burkhauser, and Hansen (2012) find negative effects on employment when they compare New York state with several comparison states, and Hoffman and Trace (2009) find that a minimum-wage increase in Pennsylvania reduced the employment prospects of “at-risk” workers relative to comparable workers in New Jersey.
Perhaps the best quality study using matching methods that identifies a disemployment effect is that of Singell and Terborg (2007), who find negative effects associated with much larger increases in the minimum wage in Oregon and Washington.
Finally, Neumark, Salas, and Wascher (2013) use a “synthetic control method” and find negative minimum-wage effects.
This important contribution to the matching literature is discussed in more detail below.
Each of these studies is open to criticism.
Hoffman (2014) shows that rectifying questionable data choices eliminates Sabia, Burkhauser, and Hansen’s (2012) negative result.
Finally, all of these analyses use state-wide data, which arguably provide a weaker match than Card and Krueger (1994), Dube, Lester, and Reich (2010), and other studies that match neighboring counties rather than states.
Even if these negative results are taken at face value, the strongest studies investigating the widest range of minimum-wage increases by Dube and his colleagues find that on average, minimum-wage increases have little or no effect on employment.
Studies without matching The alternative to a matching approach is to run a model using state-level or individual-level panel data (i.e., data collected over time) on employment levels to estimate how employment changes after states enact a higher minimum wage.
These models have a number of valuable features, most notably their ability to control for idiosyncratic differences between states or individuals that do not change over time.
These stable differences are called “fixed effects,” and the models are therefore referred to as fixed-effects models.
Regardless of whether fixed-effect models use state or individual-level data, they rely on variations in the minimum wage among states to determine the effect of the policy.
Notably absent from the fixed-effects models is any matching of comparison cases to treatment cases.
While Dube, Lester, and Reich (2010) used counties immediately across a state border as comparison cases, the fixed-effects models implicitly treat every state not experiencing a minimum-wage increase as a coequal comparison case to every state that does have a minimum-wage increase.
This potentially introduces “selection bias” into the results.
Minimum-wage laws are not imposed under experimental conditions.
This means that states that “select into” higher minimum wages by enacting increases may be systematically different from states that do not.
Fixed-effects models can handle this problem if the researcher has data on the factors that are associated with the differential adoption of minimum-wage laws or if these factors do not change over time (in that case, the inclusion of fixed effects controls for the nonrandomness that is introduced due to the lack of a true experiment).
However, if factors correlated with the adoption of minimum-wage laws vary over time and across states, fixed-effects models will produce biased estimates of the effect of the minimum wage.
This sort of bias is very plausible in practice.
Many states in the South and Central United States are experiencing rapid population and economic growth.
In contrast, communities in the Midwest and Northeast are already densely populated and in many cases undergoing a structural transition associated with the decline of manufacturing.
None of these changes are the result of the minimum-wage policy, but all are correlated with the minimum wage, which tends to be lower in the South and Central United States and higher in the Midwest and Northeast.
Other trends specific to states or counties rather than regions are also conceivable.
Some of these trends may be controlled for in certain studies, but fixed-effects models are not structured to capture the more comprehensive set of state-specific trends that matching studies can account for.
State-specific time trends that are not accounted for will move a fixed-effects model further away from results that would have been estimated by a randomized experiment.
The economists most closely associated with the fixed-effects model approach to studying the minimum wage are David Neumark and William Wascher.
In 2007, Neumark and Wascher conducted a thorough review of 102 minimum-wage studies, covering policies implemented both inside and outside the United States, and at the federal and state level.
They identified a subset of studies that they deemed “credible,” most of which fall into the category of state and individual-level fixed-effects models.
This subset of studies, selected for special mention by the most prolific authors who use the fixed-effects method, is therefore an excellent vantage point for understanding the consensus of this literature.
Most of the studies mentioned below come from this list.
Neumark and Wascher’s most recent minimum-wage study with J.M.
Salas is not a standard fixed-effects model.
This is discussed in more detail in the next section.
A typical state-level fixed-effects approach is offered by Neumark and Wascher (1992), published two years before the great disruption of the Card and Krueger (1994) study.
This research estimated that a 10 percent increase in the minimum wage reduced teenage employment by 1 to 2 percent and young adult employment by 1.5 to 2 percent.
These findings were notable because they were comparable to earlier estimates from the time series literature, which relied on variation over time rather than across states to estimate employment effects.
Neumark and Wascher (1996), Neumark (2001), and others soon extended the fixed-effects modeling framework to individual-level data to understand the impact of the minimum wage on specific vulnerable groups.
The authors find in both cases that increases in the minimum wage reduce employment for the population of interest (typically teenagers or low-skill workers).
These studies use the same design as the state-level studies, relying on variation among states and over time to estimate how changes in the minimum wage affect employment.
As such, they are vulnerable to the same criticisms outlined above.
Individuals in a high-minimum-wage state may experience lower employment rates, but it is difficult to determine whether that is the result of fundamentally different local labor market conditions that are unrelated to the minimum wage.
The most comprehensive exploration of the sensitivity of the fixed-effects model results to their ability to control for differences among states is by Allegretto, Dube, and Reich (2011).
This study uses Neumark and Wascher’s preferred fixed-effects modeling framework, but includes controls for Census division and state-specific labor market trends that Dube, Lester, and Reich (2010) suggest might be driving the strong negative employment effects in most fixed-effects analyses.
After controlling for these trends, the standard disemployment effects become statistically indistinguishable from zero effects.
What is notable about Allegretto, Dube, and Reich’s (2011) contribution is that the result of little or no disemployment effects of the minimum wage is not generated from models related to the matching studies described in the previous section.
Instead, the study uses the methods that are usually employed by Neumark and Wascher.
The method has also been extended beyond standard employment outcomes for the United States.
Couch and Wittenburg (2001) use a fixed-effects model to assess the impact of the minimum wage on hours worked, while Neumark and Wascher (2004) use these techniques to understand how labor market institutions are relevant for international differences in the effect of the minimum wage.
Both studies find the traditional negative impact.
Meer and West (2013) use state fixed-effects models and numerical examples to argue that matching studies that include location-specific time trends (discussed in more detail in the next section) may provide inappropriate employment estimates if the principal impact of changes in the minimum wage is on employment growth rates.
Which approach makes more sense?
Matching cases of minimum-wage increases to a control group is essential because it is often the closest social scientists can get to the gold standard of an experiment using random assignment.
Although the minimum-wage literature as a whole is divided on the question of the impact of minimum-wage increases, the strongest studies that use matching strategies find little or no evidence that such increases have a negative impact on employment.
It is difficult to overstate how uncontroversial it is in the field of labor market policy evaluation to assert the superiority of matching methods to the nonmatching approaches described above.
The seminal evaluations of the effects of job training programs, work-sharing arrangements, employment tax credits, educational interventions, and housing vouchers all use at least some sort of matching method, if not an actual randomized experiment.
In their widely cited survey article on non-experimental evaluation, Blundell and Costa Dias (2000) do not even mention state-level fixed-effects models when they list the five major categories of evaluation methods.
In a similar article, Imbens and Wooldridge (2009) do mention fixed-effects models as a tool for policy evaluation, but clarify that these were used before more advanced methods were developed, noting that the modern use of fixed-effects models is typically in combination with other more sophisticated techniques.