gittings2015

GETTING HANDCUFFS ON AN OCTOPUS: MINIMUM
WAGES, EMPLOYMENT, AND TURNOVER

R. KAJ GITTINGS AND IAN M. SCHMUTTE*
Theoretical work on minimum wage policy emphasizes labor mar-

ket dynamics, but the resulting implications for worker mobility
remain largely untested. The authors show that in the teenage labor
market, higher minimum wage standards reduce worker flows and
increase job stability. Furthermore, they find that the employment
effects of a relatively higher minimum wage vary considerably across
markets with different levels of turnover and labor market tightness.
Results help to explain the small effects of minimum wage standards
on employment commonly found in the aggregate data and are
consistent with labor market models that involve search frictions.
Trying to understand the nature of unemployment is like trying to put handcuffs

on an octopus. You think you have it tied down and then another pair of tenta-
cles get you round the throat.
Baily 1982
M artin Baily’s colorful description of an economist-strangling octopus

captures the difficulty of modeling the effect of minimum wages on
employment. As he argued, a good model of the minimum wage, and its
effects on youth employment, should address the ease with which teenage
workers move in and out of jobs and do so in a manner that is consistent
with empirical evidence. In the thirty-plus years since his article was pub-
lished, theoretical models of the labor market have evolved to incorporate
the dynamics of labor market adjustment and the presence of search and
information frictions. In such models, the minimum wage can affect not just
employment but also turnover, the stability of employment, and the move-
ment of employment between shrinking and expanding firms. Yet much of
*R. KAJ GITTINGS is an Assistant Professor of Economics at Texas Tech University. IAN M. SCHMUTTE is an
Assistant Professor of Economics in the Terry College of Business, University of Georgia. The research
reported in this article used resources provided by Cornell University’s Social Science Gateway, which
was funded through NSF Grant #0922005. Molly Candon and Isaac Knowles provided excellent research
assistance. We gratefully acknowledge the helpful comments of Chad Cotti, Jed DeVaro, Barry Hirsch,
Julie Hotchkiss, David Mustard, David Neumark, Mike Strain, Ron Warren, and Paul Wolfson. Additional
results and copies of computer programs used to generate the results presented in the article are avail-
able from the authors at kaj.gittings@ttu.edu or schmutte@uga.edu.
KEYWORDs: minimum wage, job flows, turnover, unemployment
ILR Review, XX(X), December 201X, pp. 1–38

DOI: 10.1177/0019793915623519. Ó The Author(s) 2015
Journal website: ilr.sagepub.com
Reprints and permissions: sagepub.com/journalsPermissions.nav
Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

2 ILR REVIEW
the research in empirical microeconomics on minimum wage policy contin-

ues to focus on teasing out implications of the link between minimum wages
and the rate of employment. Baily’s argument was that without understand-
ing how labor market dynamics mediate their relationship, we cannot
understand how minimum wages affect employment. It is as if we are fight-
ing Baily’s octopus without any clear knowledge of what it looks like.
In this article, we examine the relationship between U.S. minimum wage
policy, labor market flows, and employment in the teenage labor market
using novel data from the Quarterly Workforce Indicators (QWI). The QWI
provide quarterly measures of earnings, employment, hiring, separations,
job creations and job destructions, employment stability, and information
on the speed at which workers move in and out of nonemployment.1 These
data can be disaggregated by age, by sector, and by county, allowing us to
use standard panel data methods to identify the effect of the minimum
wage on teenage employment across states, both in the aggregate and
across industries. We show that increases in the minimum wage are associ-
ated with decreased worker flows and increased employment stability but
not with employment or job reallocation across employers. Since the impact
is largely to reduce the flow of workers into and out of jobs as well as
increase the time workers spend in a match, minimum wages have a chilling
effect on labor market volatility.
We then consider what this chilling effect means for the health of the
labor market. There are two roles worker reallocation might play in mediat-
ing the effects of the minimum wage on employment. On one hand, the
speed with which workers move from job to job can be a barometer of labor
market competition and hence labor market health. Lazear and Spletzer
(2012) argued that reduced turnover can have substantial costs by prevent-
ing workers from moving to more productive matches. On the other hand,
turnover may also be costly for firms, given the expense of recruiting and
training workers. In this case, the chilling effect of a higher minimum wage
could help firms, who offset the increased unit labor cost with lower recruit-
ing costs, not unlike an efficiency wage. Hirsch, Kaufman, and Zelenska
(2015) found, for instance, that managers of fast-food restaurants report,
when surveyed, that they absorb the cost of minimum wage increases
through channels other than employment.
We document striking relationships between the effects of minimum
wage increases on employment and labor market conditions. We use two
measures of labor market conditions: the worker reallocation rate, which
measures turnover, and the duration of nonemployment experienced by
workers who separate, which measures labor market ‘‘tightness.’’ We find
that in low-turnover markets, minimum wage increases have a negative
1
Nonemployment is distinct from unemployment. Our QWI data track movements of workers into
and out of unemployment insurance (UI) taxable employment. However, the data cannot track whether
workers are out of the labor force, searching for work, or employed in a job not covered by UI.

MINIMUM WAGES, EMPLOYMENT, AND TURNOVER 3
effect on employment, while in high-turnover markets, there is a statistically

significant employment increase in response to the minimum wage.
Similarly, markets in which workers experience short durations of non-
employment (tight markets) also have negative employment effects of mini-
mum wage increases, but we observe positive employment effects in markets
in which the duration of nonemployment for separated workers is long
(slack markets). Important to the interpretation of this finding, we show
that the bindingness of the minimum wage is not correlated with turnover
or tightness in local labor markets.
Because our measures of labor market conditions vary by industry within
state, our results can be interpreted in two different ways. In part, they
reflect variation in the effects of the minimum wage between labor markets
with different levels of turnover and tightness. They also reflect variation in
the effects of the minimum wage across sectors within the same labor mar-
kets. To further disentangle the source of heterogeneity in minimum wage
effects, we classify markets by both turnover and tightness simultaneously.
The results suggest minimum wages have disemployment effects when mar-
ket conditions are most like those in a spot market. Conversely, minimum
wages generate positive employment effects only when turnover is high,
consistent with turnover as a cost channel for industries with high turnover
and low wages relative to other sectors in the same market.
Related Literature
Drawing in data on market outcomes beyond employment has been effec-
tive for evaluating minimum wage policy. Aaronson, French, and
MacDonald (2008) showed that the relationship between minimum wage
increases and restaurant prices is consistent with a competitive labor market
and not consistent with a model in which employers hold substantial
monopsony power. Neumark, Schweitzer, and Wascher (2004, 2005) found
that the effects of minimum wages on income and poverty are also consis-
tent with the predictions of neoclassical theory. Our findings similarly help
to distinguish competing models of market frictions.
We find that aggregate estimates showing small employment effects of
the minimum wage mask stronger effects across markets when disaggre-
gated by turnover and average nonemployment durations. These findings
are related to other work regarding the importance of treatment effect het-
erogeneity in minimum wage analysis. Neumark and Wascher (2002) and
Yuen (2003) documented variation in the extent to which the minimum
wage binds for teenage workers. Singell and Terborg (2007) and Addison,
Blackburn, and Cotti (2009) provided evidence of heterogeneity in the min-
imum wage effects across industries.
Other articles have directly examined how minimum wage increases
affect worker flows, but ours is the first to directly consider the role of turn-
over in mediating the employment effects of the minimum wage. The basic

4 ILR REVIEW
finding that minimum wages reduce worker flows is becoming a stylized

fact. Portugal and Cardoso (2006) used Portuguese employer-employee
matched data to show that a large increase in the minimum wage for young
workers decreased the rates at which they separated from, and were hired
to, firms. Similarly, Brochu and Green (2013) exploited variation in mini-
mum wages across Canadian provinces, finding that higher minimum wages
led to reductions in separations and hiring. They distinguished between
quits and layoffs and argued that the reduction in separations was largely
due to a reduction in layoffs.
In the U.S. context, Thompson (2009) used QWI data to study the effects
of the minimum wage on the employment of teenagers. As part of his analy-
sis, he showed that minimum wages also decrease the share of teenagers in
new hires. Our article is closely related to recent articles that use QWI data
to study the effects of minimum wages on rates of hiring and separation
(Dube, Lester, and Reich, forthcoming) and on job flows (Meer and West,
forthcoming). Like those authors, we show that minimum wages reduce the
flow of workers into and out of jobs, and we seek to interpret this finding in
the context of models of labor market frictions. In addition, we focus on
documenting heterogeneity in the response of employment to changes in
the minimum wage across markets with different levels of turnover and
labor market tightness. We are the first to directly demonstrate that the
employment effects of the minimum wage are strongly correlated with turn-
over and labor market tightness. Our results confirm key predictions of fric-
tional search models of the labor market and suggest that these measures
may be very useful summaries of the heterogeneity across markets.
A secondary contribution relative to this quickly developing literature is
that we estimate the effect of minimum wages on teenage employment,
worker flows, and job flows using a state panel data design. This is relevant
for two reasons. First, no article has studied all three outcomes using a uni-
fied empirical framework. The state panel data framework is common to
previous work studying the effects of minimum wages on teenage employ-
ment (Neumark and Wascher 1992; Burkhauser, Couch, and Wittenburg
2000), and our design incorporates the many forms of spatial heterogeneity
found to be important. Second, our research design complements Dube et
al. (forthcoming), who used county border pairs for identification, as in
Dube et al. (2010). Since our empirical findings regarding basic outcomes
such as employment and turnover are broadly consistent, the identification
strategies and results in the two articles using different sources of variation
should be viewed as complementary.
Finally, our work is also relevant to the literature on comparative labor
market institutions. Relatively high unemployment in some European coun-
tries is associated with muted levels of worker and job reallocation. To
explain this ‘‘Eurosclerosis,’’ several articles developed frictional matching
models in the spirit of Mortensen and Pissarides (1994) to analyze the
effects of labor market regulations, including minimum wage laws, on levels

of employment and labor market flows (Blanchard and Portugal 2001; Pries
and Rogerson 2005; Gorry 2013). In these models, firms create vacancies
and hire from a pool of unemployed workers. Labor market policies, includ-
ing minimum wages, affect the expected profit from a vacancy but also the
value of remaining in a match for the worker. Evaluating these models in a
cross-country setting is compromised by the difficulty of holding unobserva-
ble institutional features fixed. We avoid this criticism by exploiting varia-
tion across U.S. states in minimum wage policy. Our analysis is therefore
similar in spirit to the work of Autor, Donohue, and Schwab (2006) and
Kugler and Pica (2008) on employment protection legislation. While far
from conclusive, our findings suggest that further efforts to estimate fric-
tional labor market models using panel data on labor market flows, in the
spirit Kiyotaki and Lagos (2007), may be a productive way forward.
Data
We combine data from three sources to study the effect of minimum wages
on employment and labor market flows. From the Quarterly Workforce
Indicators we obtain quarterly measures of teenage worker and job realloca-
tion. We also construct our own panel of state-level minimum wage laws that
include information on the month in which each minimum wage became
effective. Finally, we use state-level aggregates from the Current Population
Survey Outgoing Rotation Groups (CPS-ORG) to generate control variables
at the state level and to establish that the minimum wage is binding for
teenage workers.
The Quarterly Workforce Indicators

For our purposes, the QWI data have two primary advantages. First, the QWI
measure many variables characterizing labor market dynamics, most impor-
tant, those on the movements of workers into and out of firms. Second, the
data can be disaggregated by detailed geography, industry, and demographic
characteristics, specifically age. The measurement of worker flows and the
ability to cut the data by both age and industry allow us to study the interac-
tion between minimum wages and employment across markets characterized
by different levels of labor market tightness. These features distinguish the
QWI from the Quarterly Census of Employment and Wages (QCEW), which
are also based on administrative unemployment insurance (UI) data.
Background on QWI Data
The QWI data are a public-use product of the Longitudinal Employer
Household Dynamics (LEHD) program at the U.S. Census Bureau. The
microdata underlying the QWI, the LEHD Infrastructure files, are
employer-employee matched data constructed from state UI records. LEHD
aggregates these records in partnership with state Departments of Labor
that administer UI systems. The basic data element is a ‘‘job,’’ defined by
the appearance of a UI record that reports the UI-taxable earnings paid to

6 ILR REVIEW
a specific individual by a particular employer. The LEHD program inte-

grates the UI-based job frame with survey and administrative data that con-
tain information on the individuals and employers involved. The microdata
created by the program are then compiled into public-use products, of
which the QWI are one. Because they are built from UI records, the LEHD
data cover roughly 98% of all private-sector, nonagricultural employment,
facilitating the release of statistics at high levels of demographic, geo-
graphic, and industry detail.2
The QWI data used in this article cover 49 states for the period 1990: Q1
through 2010: Q4.3 Our initial analysis uses data reported at the state level
for two different age groups: teenagers, which QWI reports as 14 to 18 year
olds, and adult workers, aged 25 to 54. We go on to further disaggregate
the data by the North American Industry Classification System (NAICS)
3-digit industry. We use the following QWI variables:
B: employment at the start of the quarter

E: employment at the end of the quarter
A: workers newly hired this quarter (accessions)
S: workers newly separated this quarter (separations)
C: total jobs created across all firms this quarter
D: total jobs destroyed across all firms this quarter
Y: total earnings of workers employed at the end of the quarter
SH: workers hired into ‘‘stable’’ jobs this quarter
SS: workers separated from stable jobs this quarter4
NH: average number of periods of nonemployment for new hires
NS: average number of periods of nonemployment for separated workers5
In the process of aggregating the raw job-level records to geography-

industry-demographic group cells, the LEHD program imposes two data
quality controls that may affect analysis. First, the following accounting iden-
tities are imposed on the release statistics:
JF: net job flows, with JF = E2B = A2S = C2D

WR: worker reallocation, WR = A + S
JR: job reallocation, JR = C + D
Bt = Et21
2
For a comprehensive overview of the LEHD data and production of the QWI, see Abowd et al.
(2009).
3
We used the R2012Q1 version of the public use Quarterly Workforce Indicators. The raw data are
available for download from http://www.vrdc.cornell.edu/news/data/qwi-public-use-data/.
4
Stable jobs in the QWI are those that last at least one full quarter. Formally, a ‘‘full quarter’’ job is
one in which the worker is employed in that quarter, the quarter before, and the quarter after. The
inference is that the job in question was in progress throughout the entire reference quarter.
5
The variable NH measures periods of nonemployment for new hires as the average number of quar-
ters of nonemployment experienced prior to being hired: average (max [nonemployed quarters of new
hire, 4]). Periods of nonemployment for separations represent the average number of quarters of
nonemployment post separation: average (max [nonemployed quarters of separated worker, 4]).

These accounting identities do not directly affect our analysis, but they
restrict the number of independent sources of variation. Second, the QWI
data are prepared using a novel confidentiality protection procedure of
noise infusion so that small cells can be published rather than suppressed
in the released statistics.
Historical Availability
As illustrated in Figure 1, the historical availability of QWI data varies by
state. Entry dates are determined primarily by the adoption of computer-
ized record-keeping. The earliest states appear in 1990 (MD, IL, WA, and
WI). By 2000 the panel is almost balanced. Since our identification strategy
relies on cross-sectional variation in state minimum wage laws, the entry pat-
tern has some potential to affect our analysis. Our benchmark model uses
all of the available data, but for robustness, we also report estimates
restricted to the post-2000 sample.
Choice of Dependent Variables

When working with linear models, the accounting identities imposed in cre-
ating the QWI mean have really only three independent sources for
Figure 1. Annual State Minimum Wage Laws overlaid with QWI availability
Notes: Cells contain the maximum state minimum wage that exceeds the federal minimum wage in a
given year. The cell contains no entry when the state minimum is never binding. The cells are high-
lighted black if QWI data are available in that year. While the gure is annual, the minimum wage and
QWI data used in our analysis are quarterly. Full data are available from the authors upon request.

8 ILR REVIEW
measuring employment flows. Given the level of job flows JF, and worker
reallocation, WR, we have
WR + JF
ð1Þ A=
2
WR JF
ð2Þ S=
2
In addition, given job reallocation, JR,
JR + JF
ð3Þ C=
2
JR JF
ð4Þ D=
2
Therefore, we face a choice of whether to directly model worker realloca-
tion (turnover), job reallocation, and job flows, treating accessions, separa-
tions, creations, and destructions as derived, or vice versa. The related
literature models accessions and separations directly, along with creations and
destructions. We focus on turnover and job reallocation, since most of the
movements in hires and separations are highly correlated. For comparability
with existing research, we also report estimates using accessions, separations,
job creation, and job destruction as dependent variables (see Table 5).
Note that A and S are very highly correlated and have nearly identical
level- and time-series variation within states. Figures 2a and 2c plot the least-
squares residuals in worker reallocation and employment from two repre-
sentative states, California and Illinois, after removing state and period
effects. Figures 2b and 2d plot the least-squares residuals in hires and
separations.6 The series for hires and separations track each other excep-
tionally closely, which reflects that most hiring activity is driven by the need
to replace workers who turn over. These observations support our primary
focus on turnover and job reallocation.
Noise Infusion and Cell Suppression

Like the Business Dynamics Statistics (BDS) and the QCEW, the QWI data
are prepared using confidentiality protection procedures to protect small
cells. The QWI use a novel noise infusion method that maintains the analy-
tic validity of the released statistics but reduces the incidence of individual
cell suppression and eliminates the need for complementary suppression
(Abowd et al. 2012). Briefly, the procedure works by adding a small, but
non-zero, amount of noise to the underlying microdata prior to aggrega-
tion. The noise makes it impossible to ‘‘reverse engineer’’ the actual cell
counts from the release data.
6
We show plots from just two states for brevity. Plots of the data from all states are available from the
authors upon request.

Figure 2. Residual Plots: Employment and Worker Reallocation Rate and Separation and
Hiring Rate for California and Illinois
All cells and all QWI statistics are distorted through this process, but the
upshot is that the QWI data are therefore subject to very little cell suppres-
sion relative to other protection schemes and produce a more representative
sample in the published statistics. The QWI include a flag when cells are ‘‘sig-
nificantly distorted’’—that is, when the published value deviates from the con-
fidential value by a specific (confidential) threshold. To ease the reader’s
mind about potential measurement issues, we drop the flagged significantly
distorted cells as a robustness check.7 Because of the high level of aggrega-
tion, our state-level data have no cells flagged as significantly distorted.
Other Concerns
The QWI data also do not currently include reliable or consistent informa-
tion on public-sector employment or self-employment.8 We restrict our
7
It should be noted, however, that it is not clear whether dropping these cells produces better or worse
estimates. In fact, since it is clear that these cells are small to begin with, dropping them generates a sam-
ple selection issue. Regardless, even when doing so, our results are unchanged.
8
As of this writing, the LEHD program is in the process of incorporating the self-employed and public-
sector employment into the QWI, so the limitation to private-sector employment is not permanent.

10 ILR REVIEW
analysis to jobs for employers that report being privately owned. We also
have no measure of labor utilization at the intensive margin. Like much of
the minimum wage literature, our analysis is restricted, by virtue of the avail-
able data, to the extensive margin of employment adjustment.
Minimum Wage Data

We exploit state-level variation in the level and timing of minimum wage
laws for identification. To better measure timing, we have collected new
data that identify the month in which each minimum wage law was enacted
by states between 1979 and 2010. We constructed our minimum wage data
from the Monthly Labor Review in combination with primary data from
state Departments of Labor. The levels in our series are nearly identical to
those reported in other published work, in particular Sabia (2009). Our
series is available upon request.
We record 246 increases in the minimum wage between 1990 and 2010.
That count includes seven increases in the federal minimum wage, which
we represent as separate increases in each state that previously had a mini-
mum wage lower than the new federal minimum. There are, on average,
five changes in the effective minimum wage for each of the 49 states in the
QWI. Among those 246 changes, 87 (35.3%) occur in the first quarter of
the year, 11 (4.5%) in the second quarter, 94 (38.2%) in the third quarter,
and 54 (22%) in the fourth quarter.
Figure 1 shows the variation within and across states of the minimum
wage. While our analysis is based on quarterly changes in the minimum
wage, the figure is reported annually for ease of presentation. Thirty-four
states had minimum wages above the federal level at some point between
1990 and 2010. The black highlighting identifies the pattern of QWI avail-
ability. Numeric entries indicate the level of a particular state’s minimum
wage if it superseded the federal minimum wage at some point during the
year. Blank entries indicate that the federal minimum wage superseded the
state minimum wage for the entire year.9
Current Population Survey

We use the National Bureau of Economic Research (NBER) extracts from
the Current Population Survey Merged Outgoing Rotation Groups to con-
struct contextual state-level variables used in our main analysis, to analyze
the minimum wage effects on the teen wage distribution, and to compare
our results across employment measures in the QWI and CPS.10 We
9
Figure 1 conveys that the timing of entry into QWI through the 1990s misses some of the state-level
variation in minimum wages. This observation motivates us to perform robustness checks of our main
results by restricting the sample to the more balanced 2000:Q1–2010:Q4 period. The results are largely
consistent across the sample periods.
10
The Merged Outgoing Rotation Group files are available from NBER at http://www.nber.org/
morg/annual/ or by request from the authors.

aggregate the household-level data using the CPS sampling weights, into
the following state-quarter variables: the employment-to-population ratio of
teenagers, the teenage share of the state population aged 16 to 65, the aver-
age wage rate for teenagers (aged 16 to 18), the average wage of adults aged
25 to 54, and the unemployment rate for men aged 25 to 54.11
Dependent Variables
We focus on the following variables:
EPOP: end-of-quarter teenage QWI employment to population ratio (E/POP)

WRR: worker reallocation rate ((A + S)/EMP )
JRR: job reallocation rate ((C + D)/EMP )
ln Y: log of average monthly earnings of teen workers employed at the end of
the quarter
HSF: fraction of hires that are into stable jobs = SH/A
SSF: fraction of separations that are from stable jobs = SS/S
NS: average number of periods of nonemployment for separated workers
NH: average number of periods of nonemployment for new hires
where POP is teenage population, and EMP = B + 2

E
is average QWI teen
employment over the quarter. Our decision to normalize WRR and JRR
(and in later analyses hiring, separation, creation, and destruction) by
employment is for consistency with the literature on labor market flows, but
the results are not sensitive to normalization by the teenage population.
The worker reallocation rate measures the flow of workers across employers.
The job reallocation rate measures the flow of employment (jobs) from con-
tracting to expanding employers. Hence, worker reallocation can be low or
high in markets with stable employment, simply through turnover.
Descriptive Statistics
Table 1 reports summaries of the state-level data. The entries are means of
state-quarter observations, weighted by the teenage population. The average
worker reallocation rate for teenage workers in the sample is 1.19. That is,
teenage workers turn over a little more than once per quarter. The average
job reallocation rate is 0.38, which means that a little more than a third of
teenage jobs are reallocated from employers reducing their teen workforce
to those expanding it during a quarter. In the QWI, the average ratio of
worker to job reallocations is approximately three. For every job that is real-
located across firms during the quarter, three workers are reallocated.
11
See the online Appendix (http://ilr.sagepub.com/supplemental) for details.

12 ILR REVIEW
Table 1. Summary Statistics
Variable State level
QWI: (EPOP) Teen end-of-quarter employment-population ratio 0.28

(0.08)
QWI: (lnY) Log avg. end-of-quarter teen earnings 6.22
(0.243)
QWI: (WRR) Teen worker reallocation rate 1.19
(0.312)
QWI: (JRR) Teen job reallocation rate 0.38
(0.060)
QWI: (HSF) Teen fraction of stable hires 0.34
(0.114)
QWI: (SSF) Teen fraction of stable separations 0.27
(0.060)
QWI: (NS)Teen avg. periods of nonemployment for all separations 1.81
(0.219)
QWI: (NH)Teen avg. periods of nonemployment for new hires 2.63
(0.222)
QWI: Log avg. end-of-quarter adult earnings 8.10
(0.220)
CPS: Share of teens in working-age population 0.09
(0.010)
CPS: Prime-male unemployment rate 0.05
(0.026)
CPS: Teen quarterly employment-population ratio 0.38
(0.102)
CPS: Log teenage wage 1.96
(0.170)
CPS: Log adult wage 2.85
(0.179)
MW: Log state minimum wage(quarterly) 1.73
(0.184)
Number of observations 2,786
Notes: Summary statistics for the state-level combined QWI and CPS data. The universe is all state-
quarter observations that appear in the QWI. The source for each variable is indicated in its title. All
reported summary statistics are weighted by the teenage population. Standard deviations in
parentheses.
The variable ‘‘teen fraction of stable hires’’ measures the fraction of total
hires during the quarter that become stable jobs. The reported value of
0.34 means that, on average, a third of jobs that teenagers are hired into
are stable, as opposed to temporary, jobs. We also report the corresponding
statistic for separations. Slightly fewer than a third of the jobs that teenage
workers separate from were stable jobs.
The QWI also report the number of quarters of nonemployment
experienced in the previous year by the worker before being hired. For all
separations, they report the number of quarters of nonemployment experi-
enced by the worker subsequent to leaving employment. Consistent with
the strong seasonality of their employment, teenage workers experience on
average 2.63 quarters of nonemployment after separating from jobs and
have experienced 1.81 quarters of nonemployment before being hired.

Empirical Strategy
In this section, we introduce our main empirical strategy, which is an exten-
sion of the state panel data approach used in much of the related literature.
We discuss the assumptions underlying identification and our approach to
assessing robustness to violations of those assumptions. Also in this section,
we demonstrate that 1) increases in the statutory minimum wage materially
affect the wages earned by teenage workers and 2) minimum wages are
binding within and across states and industries for teenagers.
Empirical Models and Identification

For the state-level analysis, our benchmark specification is:
yst = a + lt + ms + ths + vr ðs Þ, t + xs, q(t)

ð5Þ
+ ps RECESSIONt + b In MW st + Xst G + est
The variable s indexes the state and t indexes the period (quarter).
Variable yst can be one of many labor market outcomes, including employ-
ment, worker reallocation rate, job reallocation rate, or others. State- and
period-specific effects are denoted ms and lt , and ths is a state-specific lin-
ear time trend. We also include vr ðs Þ, t , which allows for the period-specific
shocks to vary by Census region. The function r(s) maps states to its Census
region. The variable RECESSIONt is equal to 1 in periods of recession, as
dated by the NBER, and zero otherwise. The term ps measures any state-
specific effect on the outcome from being in a recessionary period. State-
specific seasonal fluctuations in the outcome are included as xs, q(t) . The
function q(t) maps period to the quarter of the year (I, II, III, IV). We denote
the natural logarithm of the effective minimum wage in states as of the
beginning of period t (first month of the quarter) as lnMWst, and Xst is a vec-
tor including the log adult wage, the share of teenagers in the adult popula-
tion, and the unemployment rate of prime-age males. The residual, nst,
accounts for unobservables affecting state-level outcomes in period t.
Heterogeneity by Labor Market Conditions

The QWI data allow us to study variation in the effect of minimum wage
increases on teenage employment across state and industry (3-digit NAICS).
A well-developed theoretical literature postulates that the employment
effects of the minimum wage may vary greatly depending on how competi-
tive, or ‘‘tight,’’ the labor market is. In markets with substantial market fric-
tions, such as in the static monopsony model, it is even possible that
increases in the minimum wage could correspond to increased employment.
Motivated by these theoretical observations, we aim to study variation in
the employment effects of the minimum wage across different levels of turn-
over and labor market tightness. We measure turnover using the average
worker reallocation rate. We proxy for labor market tightness using the

14 ILR REVIEW
average number of periods of nonemployment for separating workers. We

adopt a semi-parametric approach whereby we assign each state-industry
pair to a decile in the state-industry distribution of both measures
separately.
The base specification is the analogue of Equation (5), but now the unit
of observation is a state-industry-quarter:
ykst = a + lt + mks + thks + vr ðsk Þ, t + xks, q ðt Þ + pks RECESSIONt

ð6Þ X
10
+ Idks (k d + bd lnðMWst Þ) + Xst g + ekst
d =1
where k indicates the NAICS 3-digit industry, and Idks is an indicator equal to
1 when state-industry pair ks is in decile d and zero otherwise. The para-
meter bd is the effect of interest and measures heterogeneity in the respon-
siveness of markets to the minimum wage with respect to the level of
turnover or tightness. The remaining effects are the same as they were in
Equation (5) but defined at the state-industry level here in this model.12
We compute the within-sample average of turnover and tightness sepa-
rately for each state-NAICS 3-digit industry and pool these measures into
deciles for estimation. We adopt several alternative methods that vary the
rule for decile assignment, as described below in the subsection on estima-
tion details. Each measure has advantages and potential drawbacks. As we
show, both lead us to similar though not identical conclusions, opening
fresh questions about the link between minimum wages, employment, and
the competitive structure of teenage labor markets.
Identification
Identification of Equation (5) (and similarly Equation (6)) relies on the
assumption that changes in the minimum wage are not correlated with resi-
dual movements in the outcome. The appropriate model and research
design continues to be the subject of much controversy (Allegretto, Dube,
Reich, and Zipperer 2013; Neumark, Salas, and Wascher 2014). Our bench-
mark specification is highly saturated and controls for many different con-
founding unobservables considered in the recent literature, including state-
specific linear time trends (Addison et al. 2009), region-period shocks
(Meer and West, forthcoming), and state-specific responses to recession
(Allegretto, Dube, and Reich 2011; Neumark et al. 2014). In the empirical
work, we establish that our results are robust to further saturating the model
12
There are 4,269 unique state-NAICS 3-digit industry pairs, and estimating all the effects in Equation
(6) at the NAICS 3-digit level results in a very large number of parameters. Therefore, some of the effects
will be estimated at the state-sector level to ease the computational burden. Those instances are made
explicit in both the text and the tables. Also note that in Equation (6), kd and lt are not identified since
we control for state-industry and region-period effects. The notation is preserved for consistency of pre-
sentation with Equation (5).

as well as to more relaxed specifications. We also compare our results

against others in the quickly developing literature that studies the link
between minimum wages and labor market flows, including Dube et al.
(forthcoming), who studied worker flows using a county border-pair
research design, and Meer and West (forthcoming), who studied job growth
and flows using state-level panel data. Altogether, our results are very robust,
both to our own specification changes and by comparison to the literature.
For users unfamiliar with the QWI, note that these data are not seasonally
adjusted and exhibit clear seasonal patterns in the raw data. For this reason,
our benchmark specifications also control for state-specific seasonal cycles
(or state-sector cycles in the heterogeneity analysis). This feature of the data
has not been addressed in other minimum wage research using the QWI
but is important, as we find our results are sensitive to this control. We sus-
pect this may be associated with a coincidental link between the timing of
minimum wage changes and seasonal employment of teenage workers.
Does the Minimum Wage Bind on Teenage Workers?

To affect employment, changes in the statutory minimum wage should have
a first-order effect on the wages of employed teenage workers. We show this
is the case using individual-level data on the wages of teenagers from the
CPS. In Table 2, we report estimated minimum wage elasticities for the
mean and each decile of the teenage wage distribution. The elasticities are
estimated using the specification in Equation (5).13 Each row in the table
presents a different cut of the data. The two rows labeled (QWI) are
restricted to state quarters that appear in the QWI, allowing us to see any
effects of the selection imposed by the timing of state entry to the QWI.
The last two rows restrict the sample to the 2000 to 2010 period, in which
the QWI is nearly balanced. Raising the minimum wage increases the mean
and increases all deciles of the teen wage distribution up through the med-
ian but has little effect past the sixth decile. We interpret these results as
confirmation that the minimum wage is binding on teenage wages during
the period of our study. The restriction to observations with QWI data has a
mild attenuating effect in the full sample and no statistically meaningful
effect in the 2000 to 2010 sample.
Variation in Bindingness of the Minimum Wage by State and Industry

A key advantage of our research design is that we observe a low-wage
group—teenage workers—in every industry across states. This allows us to
examine heterogeneity in the response to the minimum wage across state-
industry markets. However, there could be substantial variation in the bind-
ingness of the minimum wage for teen workers across states and industries.
13
Note that the number of teenage workers in each state-month cell is relatively small. However, any
measurement error this generates should attenuate our estimates.

Table 2. Minimum Wage Elasticities along the Teen Wage Distribution
Decile
Sample 1 2 3 4 5 6 7 8 9 Mean
1990–2010 0.32*** 0.43*** 0.33*** 0.22*** 0.17*** 0.10*** 0.04 0.07* 20.00 0.15***
(N = 4,284) (0.055) (0.027) (0.026) (0.024) (0.025) (0.030) (0.033) (0.043) (0.044) (0.039)
1990–2010 (QWI) 0.30*** 0.38*** 0.32*** 0.20*** 0.11*** 0.05 0.01 0.04 20.04 0.11***
(N = 2,754) (0.070) (0.031) (0.037) (0.034) (0.032) (0.040) (0.044) (0.050) (0.051) (0.048)
2000–2010 0.29*** 0.36*** 0.30*** 0.18*** 0.08*** 0.05 20.00 0.06 20.01 0.11**
(N = 2,244) (0.095) (0.031) (0.039) (0.030) (0.033) (0.047) (0.049) (0.040) (0.064) (0.051)
2000–2010 (QWI) 0.27*** 0.35*** 0.29*** 0.17*** 0.08** 0.04 0.01 0.07 20.01 0.10*
(N = 2,058) (0.099) (0.032) (0.040) (0.032) (0.035) (0.050) (0.051) (0.043) (0.067) (0.053)
Notes: Minimum wage elasticities estimated under the state-level panel data model described in the section on employment, turnover, and labor market tightness. The
column lists the dependent variable and the row heading lists the sample restriction. The models are estimated under the benchmark specification that controls for
average adult wage, prime-male unemployment, the share of teenagers in the working-age population, state and period effects, state-specific trends, state-specific season
cycles, state-specific recession effects, and region-period shocks. The models are weighted by the teenage population. The parenthesized values are robust standard errors
clustered at the state level.
*, **, and *** indicate the estimate is statistically different from zero at the 10, 5, and 2.5% level, respectively.

Table 3. Bindingness of the Minimum Wage by State and Industry
Industry State-Industry State
Decile Teen Adult Teen Adult Teen Adult
Decile 1 .00 .23 –.02 .21 –.02 .42

Decile 2 .01 .31 .00 .31 .00 .44
Decile 3 .04 .43 .00 .42 .00 .44
Decile 4 .07 .49 .03 .49 .00 .49
Decile 5 .09 .54 .06 .56 .02 .50
Decile 6 .12 .61 .08 .61 .02 .52
Decile 7 .15 .66 .12 .66 .03 .54
Decile 8 .17 .71 .15 .73 .04 .56
Decile 9 .22 .83 .23 .85 .07 .60
Source: CPS-MORG 2000–2010, authors’ calculations.

Notes: The table entries measure bindingness of the minimum wage and are calculated in two stages.
The raw CPS microdata are merged to the minimum wage based on the observed state and month. We
then compute for each worker the log gap between their reported wage and the minimum wage. We
then compute the population-weighted first quintile (20th percentile) of the distribution of this gap by
NAICS 3-digit industry, by state, and by state-NAICS3 pair. The table entries report the distribution of
these quintile cutoffs across sectors and states for teenagers and for adult workers. For state-NAICS3
pairs, we omit cells with fewer than 10 (unweighted) observations.
If so, minimum wage effects that vary by state-industry might be an artifact of

the extent to which the minimum wage binds across state-industry pairs
rather than variation in market tightness. Here we show that the minimum
wage is strongly binding for teenage workers in all state-industry markets.
Table 3 uses the CPS microdata to provide evidence that the minimum wage
is binding for teenagers across most state-industry pairs. For each worker, i,
we measure the gap between the log of their reported wage and the log of
the minimum wage in their state of residence (s(i)) in the month of collec-
tion (t(i)):

ð7Þ GAPi = lnðwagei Þ ln MWs ði Þ, t ði Þ :
For each industry k, GAPk is the number such that 20% of workers in k
have GAPi GAPk . The first two columns of Table 3 show the distribution
of GAPk across all NAICS 3-digit industries for teenagers and for adult work-
ers. The third and fourth columns show the distribution of GAPsk computed
within state-industry pairs. The fifth and sixth columns, which complement
the state-level bindingness results in Table 2, show the distribution of GAPs
computed by state.14
The entry .04 in the first column indicates that at the third decile of the
industry distribution, 20% of teen workers earn within 4% of the minimum
wage. In the median industry, 20% of teen workers earn within 9% of the
current minimum wage. By contrast, in the median industry, 20% of adult
workers have wages 54% higher than the minimum. At the ninth decile,
14
GAPi is measured in log points and so the entries in Table 3 approximate percentages.

18 ILR REVIEW
20% of teen workers earn within 22% of the minimum wage. The corre-
sponding figure for adults is 83%. These patterns are nearly identical in the
third and fourth columns, which computed the gap measure for state-
industry pairs. At the state level (which masks the variation in the binding-
ness across industries) at the 90th percentile of the gap distribution across
states, 20% of teens earn within 7% of the minimum wage.
Taken together, the results suggest that while there is indeed more varia-
tion in the bindingness across industries than across states, the minimum
wage has some bite virtually everywhere. The average increase in the mini-
mum wage in our data is 10.2%, with an interquartile range of 4.8–12.0%.
Intuitively, if the true employment elasticity is 0.5, and the minimum wage
increases by 10%, we need at least 5% of the workforce to earn within 10%
of the minimum wage. On the basis of the evidence in the table, this will be
the case in most of the markets in our data. We address any remaining con-
cerns in the subsection below on results and robustness, where we show that
the bindingness of the minimum wage is not correlated with our measures
of turnover and labor market tightness.
Effects on Labor Market Outcomes

In this section, we present the results of estimating Equation (5) for each of
our employment stock and flow outcomes. We find that increases in the
minimum wage have no effect on the level of teenage employment but sub-
stantially decrease turnover and increase the stability of teenage employ-
ment. Minimum wage increases have no effect on the rate at which teenage
employment is reallocated from firms that are reducing teenage employ-
ment to those that are expanding it. Finally, average teen monthly earnings
increase as well, indicating that despite any reduction in hours at the inten-
sive margin, the overall wage bill for teenage workers increases. After pre-
senting these results, we discuss their magnitude, and then demonstrate
that they are robust to a range of alternative model specifications.
Earnings, Employment, Turnover, and Job Flows

Panel A of Table 4 presents estimates of the relationship between the mini-
mum wage and earnings, employment, turnover, job flows, and job realloca-
tion from the state-level panel data models in Equation (5). The table
includes point estimates for the minimum wage and primary control vari-
ables, and the estimated elasticity pertaining to the point estimate for the
minimum wage. Increases in the minimum wage have a positive and statisti-
cally significant effect on earnings. The employment effect of the minimum
wage in the state data is not statistically different from zero at conventional
levels. A zero employment effect cannot rule out an hours reduction on the
intensive margin, but since we observe higher earnings with minimum wage
increases, any reduction in hours should be relatively small. However, our

Table 4. Minimum Wage Effects on Teenage Employment and Labor Market

Flows (QWI: 1990–2010)
Panel A
Log Employment / Worker Job

Variable (earnings) Population reallocation rate reallocation rate
Log state minimum wage 0.10* 0.00 –0.22*** –0.03

(.054) (.021) (.082) (.017)
CPS: Log average adult wage 0.10*** 0.03* –0.12*** –0.04**
(.042) (.014) (.048) (.019)
CPS: Teen share 0.03 –3.50*** 0.32 –0.03
(.124) (.157) (0.210) (.079)
Prime-male unemployment rate –0.04 –0.11 –0.54*** –0.00
(.111) (.085) (.196) (.031)
Minimum wage elasticity .10* .02 –.20*** –.06
(.054) (.074) (.074) (.040)
R2 .984 .966 .975 .947
N 2, 786 2, 786 2, 737 2, 737
Panel B
Stable hires/ Stable separation/ Nonemployment Nonemployment

Variable Total hires Total separation new hires separation
Log state minimum wage 0.06*** 0.06*** –0.06 0.13*

(.016) (.013) (.126) (.070)
CPS: Log average adult wage 0.05*** 0.01 –.08* –0.02
(.018) (.010) (.038) (.062)
CPS: Teen share –0.11* –0.09* 0.06 0.03
(.061) (.045) (0.204) (.184)
Prime-male unemployment rate 0.05 0.09*** 0.63*** 0.64***
(.053) (.031) (.238) (.257)
Minimum wage elasticity 0.15*** 0.19*** –.02 .07*
(.041) (.041) (.046) (.038)
R2 .977 .972 .957 .953
N 2, 688 2, 688 2, 590 2, 685
Notes: The dependent variable in each model is listed at the top of the column. The models in every
column also contain state effects, period (quarter of sample) effects, state-specific time trends, effects
for state-specific seasonality, state-specific recession effects, and region-period interactions. The
regressions are estimated by weighted least squares where the weights are the teenage population.
Robust standard errors are clustered by state.
*, **, and *** indicate the coefficient is statistically different from zero at the 10, 5, and 2.5% level,
respectively.
zero employment effect also cannot rule out labor-labor substitution in the
market for teenage workers (Ahn, Arcidiacono, and Wessels 2011; Giuliano
2013).
Moving to the flow variables, minimum wages reduce the rate of worker
reallocation and do not have any effect on job reallocation. That is, the flow
of teenage workers into and out of jobs is reduced by the minimum wage.
One direct consequence of a reduction in worker reallocation is that the
duration of jobs and of unemployment spells should increase even if there

20 ILR REVIEW
is no net effect on employment levels. We check these auxiliary predictions

below but defer further interpretation of our results to the section on
employment, turnover, and labor market tightness, where we explore the
link between employment and worker reallocation more formally.
Job Stability and the Duration of Nonemployment

The QWI data also provide insight about the mechanisms driving observed
changes in employment, job flows, and worker flows. In thinking about
labor market dynamics, it is useful to distinguish between stable and
unstable employment. The teenage labor market, in particular, involves sea-
sonal work, with employers using teenagers as a flexible source of cheap
labor in periods of temporarily high demand. The institutional setting raises
the question of whether minimum wages also affect the composition of per-
manent and seasonal jobs and, if so, in what way.
Panel B of Table 4 investigates the effect of minimum wage increases on
the fraction of stable hires and separations and the duration of nonemploy-
ment for new hires and separations. The minimum-wage elasticity for the
fraction of stable hires is 0.15, meaning that a 10% increase in the mini-
mum wage increases the fraction of stable hires by about 1.5% relative to a
baseline average of about 34%. The result is similar for stable separations,
with an elasticity of 0.19 and a baseline average stable separation rate of
27%.
An increase in the fraction of both stable hires and stable separations
indicates a longer tenure on all jobs. This finding is consistent with employ-
ers’ attempts to compensate for higher labor costs by screening for workers
less likely to turn over. Alternatively, workers avoid termination when the re-
employment probability is low. Employers may also put more emphasis on
training and open fewer temporary jobs or reduce seasonal employment.
Regardless of the mechanism, an increase in the fraction of stable hires and
separations is broadly consistent with any decline in turnover.
Given that increased minimum wages reduce the rate at which teenage
workers turn over, minimum wages might also increase nonemployment
durations. Our estimates in the table show that increasing the minimum
wage has no statistically significant effect on the duration of nonemploy-
ment for new hires, but we find a small and statistically significant (at the
10% level) positive effect of increasing the minimum wage on the duration
of nonemployment for workers who separate. That is, when the minimum
wage is higher, workers experience slightly longer spells of nonemployment.
We conclude that the evidence for increased durations of nonemployment
is mixed. Since the data are measured at a quarterly frequency, these vari-
ables may be too coarse to pick up small changes in nonemployment dura-
tions. Furthermore, the number of periods of nonemployment are
measured as averages that record workers who may move job to job and do
not experience a full quarter of nonemployment as ‘‘zeros.’’ Shifts toward

hiring from unemployment should therefore be picked up as increases in

the number of periods of nonemployment. The evidence suggests that
higher minimum wages are not associated with a change in the composition
of newly hired workers on the basis of their prior employment history.
However, increased minimum wages may be associated with workers’ being
more likely to enter nonemployment (and therefore less likely to be making
job-to-job moves), which would be consistent with a decrease in the rate of
voluntary separations. In our state-level analysis, the point estimate is small
and sensitive to our specification, so we regard the result as being tentative
and suggestive.
Alternative Specifications
In Table 5, we evaluate the robustness of our initial results. Each row corre-
sponds to a different empirical specification, and each cell reports the esti-
mated elasticity of the minimum wage (and standard error) with respect to
the variable in the column heading. Row 1 restates our benchmark specifi-
cation from Table 4. To facilitate direct comparison with related research,
we add results for separation, hiring, job creation, and job destruction rates.
Rows 2–4 relax our preferred specification. Row 2 is identical to row 1
but drops state-specific seasonality effects. Row 3 drops the state-specific
recession effects from row 2; row 4 further removes the region-period
effects. That is, row 4 is left with controls for state effects, period effects,
and state-specific time trends. Row 5 then saturates the model by adding a
state-recession-unemployment rate interaction to our benchmark specifica-
tion in row 1, which allows for the effect of unemployment to vary by reces-
sion and by recessions within states.
Across these five specifications, the results are broadly consistent in
terms of sign and significance, although the magnitudes of the elasticities
vary. Effects on job reallocation are sensitive to model specification.
Furthermore, consistent with a fall in worker reallocation, minimum wages
reduce hiring and separation rates with a magnitude that is nearly identical
across all specifications. This is to be expected, as Figures 2b and 2d show
that hiring rates and separation rates are nearly identical in the aggregate.
In all but the most saturated specification (row 5), higher minimum wages
are associated with lower rates of job creation, but the elasticity is sensitive
to model specification.15
Some research suggests that minimum wage effects occur with some lag
(Neumark and Wascher 1992; Baker, Benjamin, and Stanger 1999;
Burkhauser et al. 2000). Row 6 shows the elasticity on the average of the
15
Meer and West (forthcoming) found that minimum wages negatively affect job creation when one
looks across all workers in the QWI. When we estimate across all workers, we get similar results. The
results for job creation seem particularly sensitive to controls for the state-specific seasonal component,
but results for job growth across all workers are robust.

Table 5. Minimum Wage Effects: Alternative Specifications
Panel A. Full Sample
Stable Stable
hires separation
Log Employment / Worker Job Hire Separation Job Job
(earnings) population reallocation reallocation Total hires Total separation rate rate creation rate destruction rate
(1) Benchmark 0.10* .02 –.20*** –.06 0.15*** 0.19*** –.19*** –.20*** –.08* –0.03
(0.054) (.074) (.074) (.040) (.041) (.041) (.073) (.077) (.058) (.043)
(2) No state cycles 0.12** –.01 –.26*** –.13*** 0.20*** 0.20*** –.27*** –.25*** –.19* 0.21
(0.058) (.066) (.079) (.058) (.055) (.045) (.077) (.104) (.097) (.097)
(3) No cycles/rec. 0.13** –.01 –.20*** –.10** 0.24*** 0.21*** –.23*** –.16*** –.20* 0.09
(0.054) (.065) (.065) (.050) (.060) (.046) (.073) (.069) (.108) (.092)
(4) No cyc/rec/regxP 0.07 –.05 –.27*** –.10 0.35*** 0.29*** –.28*** –.26*** –.16* 0.02
(0.058) (.086) (.100) (.063) (.048) (.066) (.096) (.108) (.086) (.060)
(5) Saturated 0.15*** .01 –.18** –.04 0.14*** 0.17*** –.17** –.19** –.05 –.02
(0.042) (.069) (.082) (.045) (.051) (.058) (.083) (.083) (.065) (.053)
(6) Lagged MW 0.10* .01 –.16*** –.06 0.18*** 0.33*** –.17*** –.15*** –.10 –.00
(0.056) (.057) (.065) (.036) (.052) (.079) (.070) (.061) (.066) (.007)
(7) County level 0.16*** 0.02 –.24*** –.06*** 0.14*** 0.18*** –.21*** –.28*** –.05*** –0.08**
(0.019) (.038) (.031) (.021) (.023) (.021) (.029) (.038) (.000) (.039)
(8) County two-stage 0.16*** –0.02 –.16*** –.05 0.14*** 0.24*** –.16 –.16*** –.08 –0.02
(0.046) (.075) (.062) (.034) (.040) (.054) (.064) (.063) (.052) (.034)
(9) Benchmark: 2000 + 0.10* .02 –.15*** –.07 0.15*** 0.23*** –.16*** –.13*** –.13 0.04
(0.056) (.093) (.053) (.051) (.049) (.061) (.058) (.053) (.089) (.068)

Notes: Each entry is the estimated elasticity of the outcome variable named in the column heading with respect to the minimum wage. Row 1 reports estimates from the
preferred specification. Row 2 drops state-specific seasonality from the benchmark. Row 3 then drops state-specific recession effects from row 2. Row 3 drops region-period
effects from Row 4. Row 5 saturates the row 1 benchmark by adding a state X recession X unemployment interaction (and the corresponding two-way interactions) to our
t1 )
benchmark in row 1. Row 6 reports the elasticity of (lnMWt +2lnMWt1 ) in a model that also includes (lnMWt lnMW
2 . Row 7 is a count-level model analogous to the benchmark in
row 1. Row 8 reports the benchmark model estimated using county-level data in a two-stage procedure. Row 9 reports the benchmark restricted to observations after 1999.
Parenthesized values are robust standard errors clustered by state, except row 7, which clusters at the county level.
*, **, and *** indicate the coefficient is statistically different from zero at the 10, 5, and 2.5% level, respectively.
contemporaneous and lagged log minimum wage in a specification that also

includes the first difference of the lagged minimum. This ‘‘lag operator fil-
ter’’ is motivated by the specification in Equation (2) of Baker et al. (1999).
The reported value picks up the long-run influence of the minimum wage
after netting out short-run variation. Unlike Baker et al. (1999) but consis-
tent with other recent findings (Addison et al. 2009; Dube et al. 2010), we
do not find evidence of a significant negative long-run effect of the mini-
mum wage on employment. For all of the outcome variables, the long-run
filter is very close to the baseline estimate.
In row 7, we report elasticities from the county-level analogue of
Equation (5). This allows us to control for county-specific observables,
trends, and cycles. The results are nearly identical to our state-level results,
with the exception that minimum wages have a small negative elasticity with
respect to job destruction rates. This is our only finding of any effect on the
job destruction rate across any specification.
Row 8 reports estimates from a county-level model that allows for arbi-
trary time-series correlation of common within-state shocks while retaining
the ability to control for county-level observables: specifically adult earnings
and the teen share in the county. We estimate this model using the proce-
dure described in Hansen (2007).16 The point estimates in row 8 are close
to those in row 1 and row 7. The real point is to correct the standard errors
relative to row 7 for correlated state-specific shocks, and indeed we observe
that the estimates in row 8 are less precisely estimated than those in row 7,
and more precise than those in row 1.
Finally, to address concerns about the unbalanced nature of the QWI,
row 9 shows that restricting the benchmark model to the years 2000 to 2010
has little effect on our results.
Employment, Turnover, and Labor Market Tightness

In this section, we present results for our analysis of heterogeneity in the
effects of minimum wages on employment across state-industry pairs with
different labor market conditions, based on the model of Equation (6).
Recall that we use two variables to proxy for differences in market condi-
tions: average turnover and average duration of nonemployment for
16
Specifically, we estimate
ð8Þ yist = zst + Zist d + eist
ð9Þ zst = lt + mt + ths + b ln MW st + Xst g + nst
where i indexes a county in state s, and Zist are county-level log adult earnings and teen share of adult
population. Xst includes the controls from our state-level models. We estimate in two stages. First, we
regress county-level outcomes on time-varying county-level observables to obtain ^zst . Second, we estimate
the state-level model on ^zst . See Bertrand, Duflo, and Mullainathan (2004); Hansen (2007) for details.

24 ILR REVIEW
Table 6. Heterogeneity in Estimated Employment Elasticity by Turnover
Within-state National
Turnover decile (1) (2) (3) (4) (5)
1 20.673*** 20.660*** 20.702*** 0.001 20.847***

(.2304) (.2357) (.2520) (.1054) (.2449)
2 20.426*** 20.405** 20.425** 20.136 20.072
(.1793) (.1858) (.1857) (.2016) (.0919)
3 20.169 20.152 20.159 20.425** 20.731***
(.1180) (.1226) (.1201) (.1845) (.1726)
4 20.176 20.150 20.157 20.497* 0.095
(.1689) (.1797) (.1786) (.2928) (.0740)
5 20.033 20.009 20.014 20.028 0.117
(.0985) (.1040) (.1050) (.2291) (.0825)
6 0.191 0.219* 0.220 0.001 0.085
(.1307) (.1309) (.1358) (.2669) (.0641)
7 0.306*** 0.339*** 0.328*** 0.145* 0.779***
(.0866) (.0911) (.0973) (.0817) (.1835)
8 0.448*** 0.483*** 0.498*** 0.414*** 0.430***
(.1114) (.1159) (.1215) (.0991) (.0954)
9 0.325*** 0.360*** 0.364*** 0.404*** 0.258***
(.0995) (.1042) (.1099) (.1167) (.1086)
10 0.096 0.120 0.101 0.255*** 0.068
(.0690) (.0709) (.0708) (.0698) (.1172)
St.-sec. trend Yes Yes Yes No Yes
St.-sec. cycle No Yes Yes Yes Yes
St.-sec. recession No Yes Yes Yes Yes
Exclude distorted No No Yes No No
St.-ind. trend No No No Yes No
No. of observations 244,332 244,332 224,244 244,332 244,332
Notes: Table entries are elasticities of employment with respect to the minimum wage evaluated at each
decile of the turnover (worker reallocation rate) distribution. The unit of observation is a state-NAICS3-
quarter combination. All models control for state-NAICS3 and region-period effects as well as state-
sector trends. In columns (1)–(3), each observation is assigned into a decile of the within-state
distribution of turnover across 3-digit NAICS industries. Column (1) reports a benchmark specification.
Column (2) adds state-sector-quarter-of-year and state-sector-recession effects. Column (3) drops
observations that were substantially distorted as part of QWI confidentiality protection (flag = 9).
Column (4) uses the specification from column (2) but controls for state-NAICS3-specific trends.
Column (5) uses the specification from column (2) but assigns observations to deciles based on the
position of the 3-digit NAICS code in the national distribution of turnover across industries over the
sample period. The parenthesized values are robust standard errors clustered at the state level.
respectively.
separating workers, which we use as a proxy for labor market tightness.

Table 6 shows that in high-turnover markets, the employment elasticity is
large and positive. In low-turnover markets, the elasticity is large and nega-
tive. This result is mirrored in Table 7 using deciles based on average non-
employment duration of separating workers. In markets where separating
workers have longer nonemployment spells (‘‘slack’’ markets), the mini-
mum wage has a large positive effect on employment. In markets where
workers have shorter nonemployment spells (‘‘tight’’ markets), the

Table 7. Heterogeneity in Estimated Employment Elasticity by Average

Nonemployment Duration for Separating Workers
Decile (1) (2) (3) (4) (5)
1 20.271 20.258 20.281** 20.103 20.318***

(.1032) (.1101) (.1221) (.1404) (.1234)
2 20.455 20.441 20.488* 20.791** 21.369***
(.2613) (.2736) (.2759) (.4054) (.3538)
3 20.587*** 20.557*** 20.580*** 20.977*** 20.018
(.1896) (.1903) (.1942) (.2994) (.1000)
4 0.209 0.241* 0.225 0.108 0.672***
(.1373) (.1414) (.1374) (.1389) (.1255)
5 0.213*** 0.241*** 0.237*** 0.243* 0.181
(.0906) (.0950) (.1001) (.1384) (.0753)
6 0.183** 0.212** 0.208** 0.343*** 0.242***
(.0879) (.0915) (.0970) (.1213) (.0804)
7 0.073 0.102 0.093 0.303*** 0.242***
(.0685) (.0739) (.0763) (.1131) (.0781)
8 0.115 0.143 0.124 0.286*** 0.201***
(.1020) (.1049) (.1148) (.0983) (.0801)
9 0.181** 0.209** 0.202** 0.276*** 0.223*
(.0912) (.0970) (.1043) (.1204) (.1248)
10 0.208*** 0.229*** 0.222** 0.457*** 0.308***
(.0870) (.0918) (.1001) (.1214) (.0758)
St.-sec. trend Yes Yes Yes No Yes
St.-sec. cycle No Yes Yes Yes Yes
St.-sec. recession No Yes Yes Yes Yes
Exclude distorted No No Yes No No
St.-ind. trend No No No Yes No
No. of observations 244,332 244,332 224,244 244,332 244,332
Notes: Table entries are elasticities of employment with respect to the minimum wage evaluated at each
decile of the average nonemployment duration distribution. The unit of observation is a state-NAICS3-
quarter combination. All models control for state-NAICS3 and region-period effects as well as state-
sector trends. In columns (1)–(3), each observation is assigned into a decile of the within-state
distribution of turnover across 3-digit NAICS industries. Column (1) reports a benchmark specification.
Column (2) adds state-sector quarter-of-year and state-sector recession effects. Column (3) drops
observations that were substantially distorted as part of QWI confidentiality protection (flag=9). Column
(4) uses the specification from column (2) but controls for state-NAICS3-specific trends. Column (5)
uses the specification from column (2) but assigns observations to deciles based on the position of the
3-digit NAICS code in the national distribution of average nonemployment durations across industries
over the sample period. The parenthesized values are robust standard errors clustered at the state level.
respectively.
minimum wage has a large negative effect. The remainder of this section
details the motivation, estimation, and interpretation of these results.
Motivational Framework
Our investigation is motivated by search-theoretic wage-posting models pop-
ular in the analysis of minimum wage policy (Manning 2003, 2006). Markets

26 ILR REVIEW
are characterized by a parameter governing labor market tightness that has

the intuitive interpretation as a measure of distance from the perfectly com-
petitive neoclassical benchmark. When labor market tightness is very low,
new employment opportunities are relatively difficult to find. Wage disper-
sion arises with firms extracting monopsony rents from relatively immobile
workers. When tightness is high, the distribution of wages collapses to the
competitive outcome where firms pay workers their marginal product.
Intuitively, the effect of the minimum wage should vary across markets
with different levels of labor market tightness. In tighter markets, the
employment effects will be strong and negative. In more slack markets, the
minimum wage can potentially increase employment through a process
akin to moving firms up individual labor supply curves.
Manning (2003) showed tightness is negatively correlated with the frac-
tion of workers who separate to nonemployment (or, equivalently, the frac-
tion of workers that are hired from nonemployment). In our data, for each
market we observe the average number of periods of nonemployment for
workers who separate. A worker who separates and is in a new job within
the same quarter contributes a value of zero to the calculation of average
nonemployment durations. Therefore, the average duration of nonemploy-
ment for separations picks up variation across markets in the share of
separations that involve direct job-to-job transitions. This may be a good
proxy for tightness. When the average nonemployment spell is low, the mar-
ket is tighter either because workers move out of nonemployment quickly
or because a larger fraction of workers are making job-to-job transitions.
The basic wage-posting model leads us to expect that the tightest markets
are the most competitive, those in which we should see the strongest nega-
tive effects of minimum wages on employment.
The average nonemployment duration is measured using quarterly data,
however, and conflates movements to nonemployment and durations of
nonemployment. This conflation is because workers who separate but
accede to employment again in the same quarter will be recorded as having
zero periods of nonemployment. Our other measure of labor market
conditions—the rate of turnover, or worker reallocation rate—does not
have this issue. While turnover is measured with greater accuracy, it is less
clear that it is related to market conditions rather than employer or
industry-specific behavior. On one hand, when markets are tight, workers
may turn over more often because job offers arrive more rapidly, making
high turnover an indicator of labor market health. On the other hand, if
workers also lose employment rapidly for exogenous reasons, because either
demand or supply is volatile, then high turnover may be associated with
large adjustment costs. These two scenarios make different predictions on
heterogeneity in the effect of the minimum wage on employment with
respect to the extent of turnover in the labor market. The question is ulti-
mately empirical.

Estimation Details
To estimate Equation (6), we begin with QWI data that are disaggregated by
state and NAICS 3-digit industry. For the period 1990 to 2010, this yields
244,332 unique state-industry-quarter observations. For each of the 4,269
state-industry combinations that appear in the data, we measure the average
level of turnover and the average nonemployment duration of separations
throughout the sample. For each state separately, we compute deciles from
the state-specific distribution of average industry turnover (or average non-
employment duration) and assign each state-industry-quarter observation in
the original data set to the corresponding (time-invariant) decile. This
method emphasizes variation across industry in our proxy measures for labor
market conditions but is flexible enough to allow, say, the food service indus-
try in New Jersey to occupy a different position in the turnover distribution
than in California (should that be the case). Furthermore, it captures the
fact that workers are more likely to turn over and reallocate within markets
that are geographically close together. Last, these decile assignments are
constant over time. To the extent that turnover may be endogenous to inno-
vations in the minimum wage and employment, our non-time-varying decile
assignments mechanically break that relationship in these data.
This procedure coarsens the data on two dimensions and is effective in
eliminating the part of variation in our measures of turnover and tightness
that is associated with the minimum wage—the pairwise correlation
between the state-industry average turnover and the contemporary mini-
mum wage is 20.0013 and statistically insignificant. However, we show addi-
tional results that coarsen the data even further and assign deciles based on
the national industry distribution of turnover. Here, each industry is
assigned a decile based on the average turnover in that industry over time
and across states. This second method should further mitigate concerns
about endogeneity, but it also has drawbacks. Specifically, it combines geo-
graphic variation in turnover across states with variation across industries
within a state. Also, each particular industry is forced to have the same dec-
ile assignment in every state. If the oil and gas industry in Texas has a differ-
ent level of turnover (or duration of nonemployment) than the oil and gas
industry in Oregon, then this decile assignment rule imposes measurement
error into our estimates. Our qualitative result remains the same when we
use this coarser measure, though the monotonicity in the estimated employ-
ment effect of the minimum wage across deciles is weaker, which would be
consistent with measurement error associated with pooling state-industry
pairs with different levels of turnover.
We have also used two alternative methods to assign deciles that yield the
same pattern of results. One assigns each state-industry a time-invariant dec-
ile based on the average turnover distribution over all state-industry pairs
(not just over industries within the same state as in our primary measure
described above). We have also performed estimates using deciles that vary
over time based on current period turnover. This design is more flexible

28 ILR REVIEW
but raises concerns about endogeneity, since Table 4 shows that minimum
wages affect both the contemporaneous level of turnover and periods of
nonemployment for separations.17
Estimating Equation (6) where all of the unobserved heterogeneity is spe-
cific to each state-3-digit NAICS industry is computationally demanding.
The NAICS collapses 3-digit industries into a set of 20 major sectors, and
most of the time-series within-state variation is common to these major sec-
tors. Our main results fit trends, seasonality, and recession effects for each
state-sector (NAICS major sector) effects rather than state-industry (NAICS
3-digit industries). For robustness, we also estimate a specification with state-
3-digit industry trends with seasonality and recession effects that are state-
sector specific.
Full Results and Robustness

Employment Effects
Tables 6 and 7 report our main results on heterogeneity in the effect of
minimum wages by turnover and labor market tightness. The table entries
are estimated minimum wage elasticities for each decile of either the turn-
over rate or average nonemployment duration for separating workers.
Columns (1) through (4) of Table 6 assign deciles based on the within-state
distribution of turnover, whereas column (5) assigns deciles based on the
national distribution of average industry turnover. Table 7 has the same
structure. Also, for both tables, all the models in columns (1) through (5)
include state-3-digit NAICS industry and region-period effects, and also con-
trol for the adult wage, teen share of the state working-age population, and
the prime-aged male unemployment rate from our state-level models.
Column (1) reports a base model that includes state-sector-specific linear
time trends. Column (2) adds state-sector-specific trends, state-sector-
specific seasonality, and state-sector-specific recession effects. Column (3)
shows that the results in column (2) are not sensitive to dropping observa-
tions reported as being significantly distorted as a result of the QWI confi-
dentiality protection methods. Column (4) replaces the state-sector trends
in column (2) with NAICS 3-digit state-industry trends. Column (5) esti-
mates the same model as column (2) but uses the method to assign deciles
based on the more coarse national industry distribution of average turnover
(or periods of nonemployment).
The magnitude of the elasticities and their statistical significance are very
similar in columns (1) through (3) for both tables. Table 6 shows there are
negative employment effects of the minimum wage where turnover is low
17
The results described in this paragraph are available upon request. We considered assigning deciles
based on measures of turnover and tightness for non-teenage workers as well. However, the non-teen or
adult measures are poor proxies for their teen counterparts for two reasons: 1) adult and teen turnover
are weakly correlated, and 2) non-teen turnover is correlated with the employment of teenage workers,
which generates the same endogeneity concerns as if one would use teen turnover to begin with.

and positive employment effects where turnover is high. Table 7 shows the
same pattern for the duration of nonemployment for separations. When
durations of nonemployment are short, higher minimum wages yield lower
employment; but when durations of nonemployment are high, there are
positive employment effects.
Column (4) of Table 6 controls for trends at the state-industry level. We
find the negative employment effects in the lower two deciles are attenu-
ated relative to the first three columns that control for state-sector trends.
However, the estimated elasticities at deciles 3 and 4 become more negative,
at 2.425 and 2.497, and statistically significant at the 5 and 10% level. The
results in the upper deciles remain roughly the same, and, if anything,
increase in magnitude.
Nearly identical changes occur in column (4) of Table 7, which adds
state-industry trends (rather than state-sector trends) to our model with het-
erogeneity by decile of nonemployment duration. The employment effect
in decile 1 becomes smaller and insignificant. However, the effect in deciles
2 and 4 becomes more negative (20.791 and 20.977) and significant at the
5 and 2.5% level. In this model the estimated positive effects at higher dec-
iles become larger and increase in statistical significance.
Column (5) in Table 6 reports estimates using the national distribution of
average industry turnover. The overall pattern of elasticity estimates is similar
to the corresponding specification in column (2) but less monotonic, and
some statistical significance is lost. As discussed above, this measure likely
introduces additional measurement error in the decile assignments. Two of
the bottom three deciles in this column are negative and statistically signifi-
cant; deciles 4, 5, and 6 are not statistically different from zero; and the esti-
mated elasticities for deciles 7, 8, 9, and 10 are all positive, three of which are
statistically different from zero. One interpretation is that the noise in this
pattern relative to the results in column (2) arises because column (5) pools
state-industry pairs with different levels of turnover. Column (5) in Table 7
similarly shows that using the national distribution of average industry dura-
tions of nonemployment for separations yields slightly noisier estimates but
does not meaningfully change the results or main conclusions.
Earnings Effects
Table 8 reports estimates of the effect of minimum wage increases on aver-
age earnings of workers employed at the end of the quarter. The model in
Table 8 is analogous to column (2) in Tables 6 and 7. In panel A, higher
minimum wages do not appear to have any differential effect on earnings
across markets exhibiting different levels of turnover. However, panel B
shows that increases in the minimum wage reduce earnings in markets with
low average duration on nonemployment and increase earnings in markets
with high durations of nonemployment.
The expected effect on earnings for retained workers is not straightfor-
ward. If a reduction in hours worked (intensive margin) dominates the

30 ILR REVIEW
Table 8. Heterogeneity in Estimated Earnings Elasticity
Panel A Panel B
Turnover decile Average duration decile
1 0.018 1 20.168***
(.0824) (.0616)
2 0.032 2 20.125***
(.0790) (.0612)
3 0.015 3 20.074
(.0758) (.0603)
4 0.035 4 20.032
(.0699) (.0619)
5 0.043 5 0.048
(.0786) (.0697)
6 0.078 6 0.072
(.0646) (.0825)
7 0.048 7 0.104
(.0780) (.0646)
8 0.052 8 0.161**
(.0906) (.0910)
9 0.023 9 0.191***
(.0724) (.0710)
10 20.107 10 0.113
(.0800) (.0863)
No. of observations 237,415 No. of observations 237,415
Notes: Table entries are earnings elasticities with respect to the minimum wage. In Panel A, the
elasticities are evaluated at each decile of the turnover (worker reallocation rate) distribution. In Panel
B, the elasticities are evaluated at deciles of the distribution of average nonemployment duration for
separations. The unit of observation is a state-NAICS3-quarter combination. Parenthesized values are
robust standard errors clustered at the state level. Both models include state-industry and region-period
effects and state-sector trends.
respectively.
extent to which the higher minimum wages increase wages, then we would
expect a reduction in average earnings when minimum wages rise. If the
wage increase due to an increase in the minimum wage dominates any
reduction in hours worked, then we would expect average earnings to
increase. Of course, employers can also keep their wage bill constant by
reducing hours to offset the higher unit labor costs, and there would be no
effect on earnings due to higher minimum wages.
The Relationship between Turnover and Nonemployment Duration

Tables 6 and 7 show that higher minimum wages reduce employment in
markets where turnover is low and in markets where the average duration
of nonemployment for separating workers is low. These results raise a natu-
ral question about the relationship between our mediating variables. In
principle, they seem related. If turnover is high, one might also expect the
duration of nonemployment to be short, since our measure of nonemploy-
ment picks up job-to-job transitions. It is important to note that these

Table 9. Two-Dimensional Classification of Labor Market Tightness:

Employment Elasticities
Market categorization (1) (2) (3) (4)
Low WRR/Low NSEP 20.490*** 20.472*** 20.476*** 20.512***

(.1588) (.1677) (.1655) (.1465)
Low WRR/High NSEP 0.021 0.044 0.031 0.084
(.0777) (.0835) (.0866) (.0750)
High WRR/Low NSEP 0.312*** 0.344*** 0.358*** 0.464***
(.0700) (.0721) (.0769) (.0817)
High WRR/High NSEP 0.236*** 0.265*** 0.254*** 0.280***
(.0871) (.0906) (.0960) (.0863)
St.-sec. trend Yes Yes Yes Yes
St.-sec. cycle No Yes Yes Yes
St.-sec. recession No Yes Yes Yes
Exclude distorted No No Yes No
No. of observations 244,332 244,332 224,244 244,332
Notes: Table entries are employment elasticities with respect to the minimum wage evaluated at cross-
classifications of turnover by duration of nonemployment for separations. For example, Low WRR/Low
NSEP indicates a state-NAICS3 combination that has average turnover and periods of nonemployment
for separating workers below their state-specific or national medians. The unit of observation is a state-
NAICS3-quarter combination. All models include state-NAICS3 and region-period effects.
Parenthesized values are robust standard errors clustered at the state level. The columns are organized
as in Tables 6 and 7.
respectively.
proxies do not necessarily group the same industries and that a low turn-
over subsector does not necessarily also have short periods of nonemploy-
ment. The two mediating variables we construct measure complementary,
but distinct, facets of the labor market.
We can push our proxy measures of labor market conditions a little fur-
ther to get at these questions. We do so by jointly classifying markets based
on both the level of turnover and the duration of nonemployment. In doing
so, we classify a given state industry across two dimensions of heterogeneity
rather than one. Specifically, we group state-industry pairs into ‘‘low’’ and
‘‘high’’ turnover and ‘‘low’’ and ‘‘high’’ nonemployment duration, by com-
bining the bottom five deciles and top five deciles. This yields a four-way
classification of state industries into 1) low turnover and low durations of
nonemployment, 2) low turnover and high durations of nonemployment,
3) high turnover and low duration of nonemployment, and 4) high turn-
over and high duration of nonemployment.
Table 9 presents the results of estimating heterogeneity in the response
of employment to the minimum wage according to our two-dimensional
classification. The models are otherwise identical to those reported in
Tables 6 and 7. We observe that minimum wages are associated with nega-
tive employment effects only in markets where turnover is low and

32 ILR REVIEW
durations of nonemployment are short, consistent with the results for both
proxies separately. Minimum wages have no statistically or quantitatively
meaningful employment effect in markets with both low turnover and long
durations of nonemployment. However, high-turnover markets with short
durations of nonemployment exhibit positive employment effects, perhaps
indicating that positive employment effects associated with high turnover
dominate negative effects associated with low employment duration.
Markets with high turnover and long durations of nonemployment have
positive employment effects (also consistent with the results for both proxies
separately).
Within state industries that have low turnover (low WRR), the employ-
ment effect of the minimum wage is more negative where nonemployment
durations are low (20.490 in low nonemployment duration markets versus
0.021 in high-duration markets). Conditional on short average nonemploy-
ment durations (low NSEP), the employment effect of the minimum wage
is higher where turnover is higher. The one case in which the basic pattern
fails to hold is in high-turnover markets. Conditional on high turnover, the
employment effect is positive and statistically significant in low NSEP mar-
kets (0.311) but slightly lower in high NSEP markets (0.236).
Our previous results suggest turnover may primarily represent a cost
channel for firms in a particular sector, whereas average nonemployment
duration is a measure of labor market conditions facing the industry in a
particular state. This analysis, which jointly classifies markets based on both
turnover and nonemployment duration, indicates that negative employ-
ment effects most likely occur when the outside market is extremely com-
petitive and adjustment costs are low. That is, minimum wages induce
unemployment where labor market conditions are most like a ‘‘spot mar-
ket.’’ Conversely, minimum wages appear to generate positive employment
effects only when turnover is high, re-emphasizing our interpretation of that
measure as reflecting the costs of adjusting labor.
Variation in Speed of Adjustment

Table 7 suggests that in markets where nonemployment durations are long,
increases in the minimum wage have a positive effect on aggregate employ-
ment and earnings. One concern is that in these markets, the speed of
adjustment may be relatively slow. If it is, part of the heterogeneity we have
measured may be coming from differences in the speed of response to the
minimum wage.
To check whether this might be the case, we estimate an extension of
Equation (6) that incorporates five quarterly leads and five quarterly lags of
the minimum wage. Figure 3a displays the estimated long-run effects for the
top three deciles (deciles 8, 9, and 10).18 The horizontal axis measures the
number of quarters relative to a change in the minimum wage. The lines
18
Specifically, the leads and lags are added to the model represented by column (2) in Table 7.

plot the cumulative effect of the minimum wage on employment, expressed

as an elasticity. The figure shows no evidence of a long-run negative effect
up to five quarters out.19 Furthermore, there is no evidence of a systematic
trend in any of the five quarters prior to the minimum wage change.
For completeness, we display the same information for deciles 4, 5, 6,
and 7 (Figure 3b) and deciles 1, 2, and 3 (Figure 3c). Figure 3c indicates
there may be a residual trend at deciles 1, 2, and 3 in the quarters prior to
the minimum wage change under our benchmark specification with state-
sector trends, seasonality, and recession effects. We address this by control-
ling for the most disaggregated state-industry trends in column (4) of
Table 7. When we do, the effect at deciles 2 and 3 becomes more negative
and statistically significant, and the results at higher deciles become more
positive.20
Turnover and the Bindingness of the Minimum Wage

We saw above in the section on empirical strategy that the minimum wage
binds for teenagers across industries and states. Nevertheless, there may be
specific state-industry combinations for which the minimum wage is not
binding, or binds little. A remaining concern is that our results are driven
by variation in the extent to which the minimum wage binds on teenage
workers across markets with different levels of turnover.
Table 10 presents evidence that this is not likely to be the case. To build
on the evidence from Table 3, we create two measures of the bindingness
of the minimum wage within state-industry cells and show these measures
are very high for teen workers across deciles of the turnover and non-
employment duration. To construct these measures, we return to the CPS
and compute, for each teen, an indicator equal to 1 if their wage is within
10% of the prevailing minimum wage in the same state and month.21 We
then calculate the share of teens for which the minimum wage binds in
each state-industry cell (using the CPS weights).
The rows of Table 10 correspond to state-industry deciles of turnover
(columns 1 and 2) and nonemployment duration (columns 3 and 4). The
entries in columns (1) and (3) report the share of state-industry markets in
each decile in which at least 10% of the teenagers face a binding minimum
wage. The intuition underlying this measure is that if 10% of workers earn
within 10% of the minimum wage, then at least 10% of workers would be
affected by the average minimum wage increase. The fraction of markets
that have at least 10% of teenagers with wages within a 10% increase in the
minimum wage is substantial. Turnover deciles 1 through 9 all reflect hav-
ing at least 93% of markets with binding minimum wages. Decile 10 has a
somewhat lower fraction of markets with binding minimum wages, but the
19
Sorkin (forthcoming) argued that true long-run effects are hard to identify using U.S. data.
20
Controlling for state-sector quadratic trends does not substantially change our results relative to the
benchmark specification with state-sector linear trends.
21
We choose 10% because it is the size of the average minimum wage increase in our sample.

34 ILR REVIEW
Figure 3. Cumulative Employment Elasticity of the Minimum Wage by Decile of the

Average Duration of Nonemployment for Separations
minimum wage still binds in 84.6% of these markets. The results are simi-
larly strong for deciles of nonemployment duration.
Columns (2) and (4) report the average share of teens for whom the
minimum wage binds in each state industry. This is a measure of the num-
ber of teenagers we expect to be affected by a minimum wage increase of

Table 10. Bindingness of the Minimum Wage by State-Industry Turnover and

Nonemployment Duration Decile
Turnover Nonemployment duration
Share . 0.1 Average Share . 0.1 Average

Decile (1) (2) (3) (4)
Decile 1 0.931 0.351 0.959 0.346

Decile 2 0.964 0.468 0.980 0.478
Decile 3 0.973 0.391 0.972 0.418
Decile 4 0.970 0.432 0.932 0.351
Decile 5 0.970 0.397 0.838 0.284
Decile 6 0.961 0.413 0.910 0.314
Decile 7 0.967 0.376 0.852 0.284
Decile 8 0.982 0.344 0.916 0.320
Decile 9 0.963 0.329 0.976 0.367
Decile 10 0.846 0.264 0.999 0.381
Source: CPS-MORG and QWI, authors’ calculations.

Notes: The table entries report the share of state-NAICS3 pairs for which the minimum wage is binding
in each turnover (nonemployment duration) decile. We define bindingness of the minimum wage in
two stages. The raw CPS microdata are merged to the minimum wage based on the observed state and
month. The minimum wage binds for a worker if the difference between the natural log of his or her
reported wage and the log minimum wage is less than 0.1. We then define the bindingness of the
minimum wage in a state-NAICS3 market as the (weighted) share of workers for whom the minimum
wage binds. Column (1) reports, for each turnover decile, the share of state-NAICS3 markets in which
the minimum wage binds for at least 10% of teen workers. Column (2) reports, for each turnover
decile, the weighted mean bindingness across state-NAICS3 markets. Columns (3) and (4) report the
same statistics by deciles of the distribution of average nonemployment duration. In all cases, we omit
cells with fewer than 10 (unweighted) observations.
10%. For turnover (column 2), these estimates range from 26.4 to 46.8% of
teenagers. For nonemployment duration, they range from 28.4 to 47.8%.
While there is variation in these numbers, economically speaking they illus-
trate that minimum wage increases will have a bite in all deciles.
Furthermore, what variation exists in bindingness across our measures of
tightness is not strongly associated with the pattern of employment effects
we observe in the previous tables.
Finally, some caution is required in interpreting Table 10. The number
of teens in the CPS in a particular state and a particular industry is very
small from month to month and year to year. To construct Table 10, we
must pool information across years. Hence, our statistics may conflate peri-
ods in which the minimum wage is very binding and those in which it is not
binding at all.
Conclusions
A troubling gap exists between theoretical models of the effects of the mini-
mum wage and empirical evidence. We suspect this is because many aspects
of the labor market response have been hidden from view. New data from
the Census Bureau’s Quarterly Workforce Indicators combine geographic,

36 ILR REVIEW
industry, and demographic detail on the stock and flow of employment and
earnings. The detail in these data facilitates a much finer analysis of the
many margins along which labor markets respond to changes in the mini-
mum wage. The primary contributions of this article have been to use the
QWI 1) to further characterize the effect of minimum wage policy on labor
market dynamics, particularly turnover and job reallocation, and 2) to docu-
ment heterogeneity in the effects of the minimum wage across industries
characterized by different labor market conditions.
Theoretically, the effect of the minimum wage on employment is a func-
tion of the degree of competition in the market and the ability of the firm
to easily adjust its labor input. We use two variables in the QWI that proxy
for these market and employer characteristics: turnover and the average
duration of nonemployment for separating workers. Both measures capture
heterogeneity across industries in the response of employment to the mini-
mum wage. The employment effects of an increase in the minimum wage
are strong and negative in markets with low turnover and in markets with
short durations of nonemployment for workers who separate from their
jobs. Conversely, increases in the minimum wage are associated with
increases in employment in high-turnover markets and in markets with high
nonemployment durations. These results are robust to various specifications
and are not driven by variation in bindingness of the minimum wage.
Our results are consistent with the view that shorter durations of non-
employment correspond to more competitive markets. In the most competi-
tive markets, we expect the data to behave similarly to the way the
neoclassical demand model predicts—increases in the minimum wage bring
decreased employment. In the least competitive markets, conditions may be
closer to dynamic oligopsony, in which it is possible that higher minimum
wages may increase employment. Our results that estimate the effect of the
minimum wage on employment, conditional on the duration of nonemploy-
ment, conform to these stylized predictions.
The mechanism by which turnover mediates the effect of the minimum
wage on employment is less clear. The pattern we observe—negative
employment effects in low-turnover markets—runs counter to a wage-
posting model in which high-turnover firms are at the bottom of the job lad-
der. However, our results by turnover are consistent with a model in which
turnover represents a cost channel for firms. If turnover is very costly, then
minimum wages hikes can ‘‘pay for themselves’’ by reducing turnover. This
implication emerges from a simple model of adjustment costs based on
Manning (2006). In this environment, our observed pattern would be
expected—negative employment effects where turnover is low and positive
effects where turnover is high.
Additional work is needed to better explain the link between turnover,
nonemployment durations, minimum wage policy, and employment. Our
results indicate that heterogeneity in the effects of the minimum wage are
meaningfully summarized by turnover and durations of nonemployment,

but it remains an open question whether these variables proxy for variation
in conditions within, or across, local labor markets, and whether they proxy
for variation in market frictions, supply conditions, or production
technology.
References
Aaronson, Daniel, Eric French, and James MacDonald. 2008. The minimum wage, restaurant
prices, and labor market structure. Journal of Human Resources 43(3): 688–720.
Abowd, John M., Kaj Gittings, Kevin L. McKinney, Bryce Stephens, Lars Vilhuber, and Simon
Woodcock. 2012. Dynamically consistent noise infusion and partially synthetic data as con-
fidentiality protection measures for related time-series. 2012 Research Conference Papers.
Federal Committee on Statistical Methodology, Office of Management and Budget.
Abowd, John M., Bryce E. Stephens, Lars Vilhuber, Fredrik Andersson, Kevin L. McKinney,
Marc Roemer, and Simon Woodcock. 2009. The LEHD infrastructure files and the cre-
ation of the Quarterly Workforce Indicators. In Timothy Dunne, J. Bradford Jensen, and
Mark J. Robers (Eds.), Producer Dynamics: New Evidence from Micro Data, pp. 149–230. Chi-
cago: University of Chicago Press.
Abraham, Katharine G., John Haltiwanger, Kristin Sandusky, and James Spletzer. 2009.
Exploring differences in employment between household and establishment data. Work-
ing Paper No. 09-09. Center for Economic Studies, U.S. Census Bureau.
Addison, John T., McKinley L. Blackburn, and Chad D. Cotti. 2009. Do minimum wages raise
employment? Evidence from the U.S. retail-trade sector. Labour Economics 16(4): 397–408.
Ahn, Tom, Peter Arcidiacono, and Walter Wessels. 2011. The distributional impacts of mini-
mum wage increases when both labor supply and labor demand are endogenous. Journal
of Business & Economic Statistics 29(1): 12–23.
Allegretto, Sylvia, Arindrajit Dube, and Michael Reich. 2011. Do minimum wages really
reduce teen employment? Accounting for heterogeneity and selectivity in state panel
data. Industrial Relations 50(2): 205–40.
Allegretto, Sylvia, Arindrajit Dube, Michael Reich, and Ben Zipperer. 2013. Credible research
designs for minimum wage studies. Working Paper. Institute for Research on Labor and
Employment, Institute of Industrial Relations, University of California Berkeley.
Autor, David H., John J. Donohue III, and Stewart J. Schwab. 2006. The costs of wrongful-
discharge laws. Review of Economics and Statistics 88(2): 211–31.
Baily, Martin N. 1982. Comment on ‘‘The minimum wage and job turnover in markets for
young workers.’’ In Richard B. Freeman and David A. Wise (Eds.), The Youth Labor Market
Problem: Its Nature, Causes, and Consequences, pp. 494–97. Chicago: University of Chicago
Press.
Baker, Michael, Dwayne Benjamin, and Shuchita Stanger. 1999. The highs and lows of the
minimum wage effect: A time-series cross-section study of the Canadian law. Journal of
Labor Economics 17(2): 318–50.
Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan. 2004. How much should we
trust differences-in-differences estimates? Quarterly Journal of Economics 119(1): 249–75.
Blanchard, Olivier, and Pedro Portugal. 2001. What hides behind an unemployment rate:
Comparing Portuguese and U.S. labor markets. American Economic Review 91(1): 187–207.
Brochu, Pierre, and David A. Green. 2013. The impact of minimum wages on labour market
transitions. Economic Journal 123(573): 1203–35.
Burkhauser, Richard V., Kenneth A. Couch, and David C. Wittenburg. 2000. A reassessment
of the new economics of the minimum wage literature with monthly data from the Cur-
rent Population Survey. Journal of Labor Economics 18(4): 653–80.
Dube, Arindrajit, T. William Lester, and Michael Reich. 2010. Minimum wage effects across
state borders: Estimates using contiguous counties. Review of Economics and Statistics 92(4):
945–64.
——. 2016. Minimum wage shocks, employment flows and labor market frictions. Journal of
Labor Economics 34(3).

38 ILR REVIEW
Giuliano, Laura. 2013. Minimum wage effects on employment, substitution, and the teenage
labor supply: Evidence from personnel data. Journal of Labor Economics 31(1): 155–94.
Gorry, Aspen. 2013. Minimum wages and youth unemployment. European Economic Review 64:
57–75.
Hansen, Christian B. 2007. Generalized least squares inference in panel and multilevel mod-
els with serial correlation and fixed effects. Journal of Econometrics 140(2): 670–94.
Hirsch, Barry T., Bruce Kaufman, and Tetyana Zelenska. 2015. Channels of labor market
adjustment: Analysis of the 2007–2009 federal minimum wage increases. Industrial Rela-
tions: A Journal of Economy and Society 54(2): 199–239.
Kiyotaki, Nobuhiro, and Ricardo Lagos. 2007. A model of job and worker flows. Journal of
Political Economy 115(5): 770–819.
Kugler, Adriana, and Giovanni Pica. 2008. Effects of employment protection on worker and
job flows: Evidence from the 1990 Italian reform. Labour Economics 15(1): 78–95.
Lazear, Edward P., and James R. Spletzer. 2012. Hiring, churn, and the business cycle. Ameri-
can Economic Review 102(3): 575–79.
Manning, Alan. 2003. Monopsony in Motion: Imperfect Competition in Labor Markets. Princeton,
NJ: Princeton University Press.
——. 2006. A generalised model of monopsony. Economic Journal 116(508): 84–100.
Meer, Jonathan, and Jeremy West. Forthcoming. Effects of the minimum wage on employ-
ment dynamics. Journal of Human Resources.
Mortensen, Dale T., and Christopher A. Pissarides. 1994. Job creation and job destruction in
the theory of unemployment. Review of Economic Studies 61(3): 397–415.
Neumark, David, J. M. Ian Salas, and William Wascher. 2014. Revisiting the minimum wage-
employment debate: Throwing out the baby with the bathwater? ILR Review 67: 608–48.
Neumark, David, Mark Schweitzer, and William Wascher. 2004. Minimum wage effects
throughout the wage distribution. Journal of Human Resources 39(2).
——. 2005. The effects of minimum wages on the distribution of family incomes: A nonpara-
metric analysis. Journal of Human Resources 40(4): 867–94.
Neumark, David, and William Wascher. 1992. Employment effects of minimum and submini-
mum wages: Panel data on state minimum wage laws. Industrial and Labor Relations Review
46(1): 55–81.
——. 2002. State-level estimates of minimum wage effects: New evidence and interpretations
from disequilibrium methods. Journal of Human Resources 37(1): 35–62.
Portugal, Pedro, and Ana R. Cardoso. 2006. Disentangling the minimum wage puzzle: An
analysis of worker accessions and separations from a longitudinal matched employer-
employee data set. Journal of the European Economic Association 4(5): 988–1013.
Pries, Michael, and Richard Rogerson. 2005. Hiring policies, labor market institutions, and
labor market flows. Journal of Political Economy 113(41): 811–39.
Sabia, Joseph J. 2009. The effects of minimum wage increases on retail employment and
hours: New evidence from monthly CPS data. Journal of Labor Research 30(1): 75–97.
Singell, Larry D., and James R. Terborg. 2007. Employment effects of two Northwest mini-
mum wage initiatives. Economic Inquiry 45(1): 40–55.
Sorkin, Isaac. 2015. Are there long-run effects of the minimum wage? Review of Economic
Dynamics 18(2): 306–33.
Thompson, Jeffrey P. 2009. Using local labor market data to re-examine the employment
effects of the minimum wage. Industrial and Labor Relations Review 62(3): 343–66.
U.S. Census Bureau. 2006. Current Population Survey: Design and methodology. Technical
Paper No. 66. U.S. Census Bureau.
Yuen, Terence. 2003. The effect of minimum wages on youth employment in Canada: A
panel study. Journal of Human Resources 38(3): 647–72.

gittings2015

Uploaded by

Copyright:

Available Formats

gittings2015

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

gittings2015

Uploaded by

Copyright:

Available Formats

GETTING HANDCUFFS ON AN OCTOPUS: MINIMUM

WAGES, EMPLOYMENT, AND TURNOVER

Theoretical work on minimum wage policy emphasizes labor mar-

Trying to understand the nature of unemployment is like trying to put handcuffs

M artin Baily’s colorful description of an economist-strangling octopus

KEYWORDs: minimum wage, job flows, turnover, unemployment

ILR Review, XX(X), December 201X, pp. 1–38

Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

the research in empirical microeconomics on minimum wage policy contin-

Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

effect on employment, while in high-turnover markets, there is a statistically

Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

finding that minimum wages reduce worker flows is becoming a stylized

Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

The Quarterly Workforce Indicators

Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

a specific individual by a particular employer. The LEHD program inte-

B: employment at the start of the quarter

In the process of aggregating the raw job-level records to geography-

JF: net job flows, with JF = E2B = A2S = C2D

Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

Choice of Dependent Variables

Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

Noise Infusion and Cell Suppression

Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

Minimum Wage Data

Current Population Survey

Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

EPOP: end-of-quarter teenage QWI employment to population ratio (E/POP)

where POP is teenage population, and EMP = B + 2

Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

Table 1. Summary Statistics

Variable State level

QWI: (EPOP) Teen end-of-quarter employment-population ratio 0.28

Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

Empirical Models and Identification

yst = a + lt + ms + ths + vr ðs Þ, t + xs, q(t)

Heterogeneity by Labor Market Conditions

Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

average number of periods of nonemployment for separating workers. We

ykst = a + lt + mks + thks + vr ðsk Þ, t + xks, q ðt Þ + pks RECESSIONt

Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

as well as to more relaxed specifications. We also compare our results

Does the Minimum Wage Bind on Teenage Workers?

Variation in Bindingness of the Minimum Wage by State and Industry

Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

Table 3. Bindingness of the Minimum Wage by State and Industry

Industry State-Industry State

Decile Teen Adult Teen Adult Teen Adult

Decile 1 .00 .23 –.02 .21 –.02 .42

Source: CPS-MORG 2000–2010, authors’ calculations.

If so, minimum wage effects that vary by state-industry might be an artifact of

Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

Effects on Labor Market Outcomes

Earnings, Employment, Turnover, and Job Flows

Downloaded from ilr.sagepub.com at University of New England on June 4, 2016

Table 4. Minimum Wage Effects on Teenage Employment and Labor Market

Log Employment / Worker Job

Log state minimum wage 0.06* 0.06* –0.06 0.13*

1 20.673* 20.660* 20.702* 0.001 20.847*

1 20.271 20.258 20.281 20.103 20.318*

Low WRR/Low NSEP 20.490* 20.472* 20.476* 20.512*