Employers Gone Rogue: Explaining Industry Variation in Violations of Workplace Laws

Drawing on an innovative, representative survey of workers in Chicago, Los Angeles, and New York City, the authors analyze minimum wage, overtime, and other workplace violations in the low-wage labor market. They document significant interindustry variation in both the mix and the prevalence of violations, and they show that while differences in workforce composition are important in explaining that variation, differences in job and employer characteristics play the stronger role. The authors suggest that industry noncompliance rates are shaped by both product market and institutional characteristics, which together interact with labor supply and the current weak penalty and enforcement regime in the United States. They close with a research agenda for this still-young field, framing noncompliance as an emerging strategy in the reorganization of work and production at the bottom of the U.S. labor market.

measures continues to be a significant barrier for researchers, an emerging body of evidence suggests that the systematic violation of employment and labor laws is common in a number of low-wage industries (Bernhardt et al. 2009;Weil 2011;Theodore, Bernhardt, and DeFilippis 2012). Examples of violations include disregarding minimum wage and overtime laws; requiring employees to work without pay before or after their shifts; retaliating against workers filing complaints or trying to organize; maintaining unsafe workplaces; and skirting workers' compensation obligations.
The prevalence of workplace violations varies significantly by industry, as we document here. This is especially important for industrial relations researchers because it raises a host of questions about both the patterns and the correlates of employer noncompliance. Are there distinct patterns in which types of laws are violated in different industries? How much industry variation in violations is explained by differences in workforce composition, such as reliance on undocumented workers? How much is explained by differences in employer-level characteristics? And what can we learn about the relative contributions of competitive conditions and institutional arrangements?
Our goal is to begin to answer these questions using the 2008 Unregulated Work Survey (Bernhardt et al. forthcoming), an innovative and representative survey of 4,387 workers employed in low-wage industries and occupations in Chicago, Los Angeles, and New York City. We document substantial interindustry variation across five measures of workplace violations and identify three groups of industries that differ not just in the prevalence but also in the mix of violations. Using multivariate regression, we estimate that worker and employer characteristics each account for a large and statistically significant portion of the interindustry variation in employer noncompliance. But while differences in workforce composition-especially immigration status-are important and should not be neglected, it is differences in job and employer characteristics that play the stronger role in accounting for industry differences in noncompliance.
As with other labor market research, we have a limited number of employer measures in our data set with which to test explanations for these demand-side effects. But based on the benchmark findings in this article, our interpretation is that industry variation in employer noncompliance with workplace laws is shaped by both product market characteristics (that impact ability to pay) and institutional characteristics (that shape managerial strategy), which together interact with labor supply characteristics. At the same time, all industries in the United States currently face a very weak penalty and enforcement regime. In our view, this results in a high overall rate of noncompliance, around which industries then vary based on their particular demand-and supply-side characteristics.

Explanations for Employers' Noncompliance with Workplace Laws
Standard economic theory suggests that employers-as profit-maximizing agents-will comply with legal mandates if the likelihood of detection is great and if penalties for violators are also correspondingly high (Becker 1968). Orley Ashenfelter and Robert Smith (1979) refined this argument by positing that firms will weigh the economic costs of complying with minimum wage laws against the probable costs of noncompliance, expressed in the form of back wages and penalties imposed by government enforcement agencies. Ashenfelter and Smith expect the benefits of noncompliance to rise the more that the market wage falls below the statutory wage rate.
Given this broad framework, what might account for variation across industries in the extent of employer compliance with workplace laws? We will document such variation in more depth, but initial data points come from a series of compliance audits conducted in the late 1990s by the U.S. Department of Labor (USDOL). The agency found that rates of compliance with the Fair Labor Standards Act differed substantially by industry: 22% of garment manufacturers in Los Angeles; 33% of day care providers in the Gulf Coast; 40% of poultry processing plants; 53% of full service restaurants in Tampa; 64% of auto repair shops in Richmond; and 70% of nursing homes nationwide (U.S. Department of Labor 2001).
It is highly unlikely that these industry differences in violation rates are the result of industry differences in either the strength of penalties or the strength of enforcement-that is, in the cost of noncompliance. First, penalties and damages under federal and state laws do not vary by industry, and they are also widely acknowledged to be quite weak (see Estlund 2005). Second, while federal and state departments of labor have made scattered attempts at industry-specific enforcement, these initiatives have lacked both the scale and consistency needed to counter the inadequate and declining level of resources dedicated to wage and hour enforcement. The National Employment Law Project (2008) calculated that between 1980 and 2007, the number of federal wage and hour inspectors declined by 31% and the number of enforcement actions fell by 61%. In contrast, the civilian labor force grew by 52% during this period. While the USDOL has added wage and hour investigators under the Obama administration, current federal staffing levels are still below their 1980 watermark (U.S. Department of Labor 2011), and the number of state investigators is significantly lower than that (Schiller and DeCarlo 2010). As a result, Min Woong Ji and David Weil (2010) put the probability that a restaurant in a national fast-food chain would be inspected by a wage and hour investigator at about 0.8% in a given year.
If industries effectively are facing the same penalty and enforcement regime, and a weak one at that, then differences in workplace violation rates must be coming from the benefit side of the equation-that is, from differences in how much employers can potentially save by violating the law. Under Ashenfelter and Smith's framework (1979) and as elaborated by Weil (2005), both supply-and demand-side factors can affect the expected benefit from, and therefore the rate of, noncompliance. For example, industries employing unskilled, highly substitutable, and otherwise vulnerable workers should be more likely to pay subminimum wages, all else being equal. On the demand side, Weil (2005) found that small firm size, labor-intensive production, and weak pricing power were correlated with noncompliance in a sample of garment manufacturers. Lack of data, however, has prevented researchers from assessing the relative weight of these explanations for a representative sample of industries.
The related literature on interindustry wage differentials offers insights (see, for example, Krueger and Summers 1988). We look to this literature because, in the absence of strong enforcement and penalties, violations of minimum wage and overtime laws can be treated as a form of wage setting at the bottom of the wage distribution. Our reading of this research is that worker characteristics (including both measured and unmeasured ability) explain roughly half the variation in average wages across industries but that significant and persistent wage differentials remain. How to explain the true industry effect is an unresolved question, although rent sharing and efficiency wages are leading explanations (Krueger and Summers 1988;Gibbons and Katz 1992).
Based on these literatures, we hypothesize that workforce and employer characteristics each will account for a significant portion of the industry variation in workplace violations observed in our data. To the standard list of worker characteristics (education, age, gender, and race/ethnicity), we add immigration status and English-speaking ability because of our focus on the low-wage labor market. Research suggests that lack of legal status has led employers in low-wage industries to prefer to hire undocumented workers, who are perceived as having a better work ethic and less likely to complain about substandard conditions (see Moss and Tilly 2001;Peck and Theodore 2001;Waldinger and Lichter 2003;Gomberg-Muñoz 2011). Employers may seize upon these workers' vulnerabilities by extracting cost savings directly in the form of depressed wages, driving down market wages in the process (Massey, Durand, and Malone 2002). In addition, given our focus on violations of workplace laws, we also include workers' knowledge of the minimum wage in our analysis. Taking into consideration all the variables, our prediction is that industries with higher proportions of less educated and otherwise disadvantaged workers will have higher noncompliance rates because of the higher substitutability, legal vulnerability, and lack of (informal) bargaining power of these workers.
In terms of job and employer characteristics, researchers have identified a number of factors as potentially relevant to explaining industry differences in wage setting. In the literature on interindustry wage differentials, firm size, profit margins, and capital-labor ratios have been found to be key predictors, along with, to a lesser degree, unionization, market concentration, and trade penetration (Dickens and Katz 1987). The industrial relations literature leads us to expect that differences in managerial strategy and the organization of work and production, especially the use of subcontracting and other forms of nonstandard work (Kalleberg 2000;Weil 2011), will also play a role (Beynon et al. 2002;Osterman 2011).
As with other research in this area, we are limited by the job and employer measures available in our data set, though several measures exceed what is available in standard government surveys. At the job level, we measure full-time/part-time status, whether the employment relationship is temporary or short-term, and the number of employers last week (in our sample, this is a measure of decentralized employment relationships). In all cases, our prediction is that industries with higher proportions of nonstandard work arrangements will have higher violation rates because such arrangements can create greater legal distance between employers and workers. At the employer level, we measure firm size, the provision of multiple workplace benefits, and the use of nonstandard pay systems (such as flat weekly wages). 1 Our prediction is that industries with smaller employers who provide few benefits and use nonstandard pay systems will have higher violation rates. The interpretation of these measures (especially firm size) will be less straightforward because they are likely capturing the effects of several related but unmeasured employer characteristics.

Data and Methods: The 2008 Unregulated Work Survey
Existing data are inadequate to assess the current state of employer compliance with U.S. workplace laws. Standard government surveys such as the Current Population Survey and the American Community Survey do not gather the detailed data needed to measure workplace violations. Original surveys of workers or employers are rare, and they must overcome significant challenges to accurately sample the populations of interest and measure violations of intricate legal standards.
In order to fill this data vacuum, the authors designed and conducted the 2008 Unregulated Work Survey (UWS), a representative survey of 4,387 frontline workers in low-wage industries and occupations in the three largest U.S. cities, Chicago, Los Angeles, and New York. 2 We adopted two key methodological innovations to overcome the sampling and measurement challenges. First, we used a recently developed sampling methodology, Respondent-Driven Sampling, that allowed us to reach the full range of workers in the low-wage labor market, including unauthorized immigrants and off-the-books workers. Second, we developed an extensive questionnaire that allowed us to rigorously assess whether employment and labor laws were being violated without relying on workers' own knowledge of these laws. We give a brief description of the UWS here; see Spiller, Bernhardt, Perelshteyn, and Heckathorn (2010) for full documentation of survey design and methodology.

The UWS Sampling Universe
In order to be included in our study, workers (1) had to be age 18 or older and currently working for an employer within the limits of Los Angeles County, Cook County (Chicago), or the five boroughs of New York; (2) had to be a frontline worker, meaning not a manager, professional, or technical worker; and (3) had to hold at least one job in a low-wage industry or occupation in the previous workweek. We excluded managers and professional and technical workers because many are exempt from the workplace laws examined in our study. Our focus on low-wage industries and occupations was driven by resource constraints. Sampling the entire labor market would have been highly inefficient and cost prohibitive. 3 We estimate that our sample represents about 1.64 million workers or about 31% of the frontline workforce and 15% of the total workforce of the three cities combined. By selecting industries and occupations that constitute approximately the bottom third of the frontline workforce, we struck a balance between scope and depth, a common tradeoff in designing labor market surveys.

Sampling Methodology
In light of the challenges of surveying hard-to-reach and vulnerable workers with standard sampling approaches, we adopted an innovative sampling method that operates through respondents' social networks. All workers have friends, family, or coworkers whom they trust and with whom they come into regular contact. Our approach relied on chain-referral sampling, in which survey respondents recruited into the sample people they knew, and importantly, to whom they could communicate that the survey was safe (that it would not trigger, for example, being reported to the immigration or tax authorities).
The best-known sampling method using this form of recruitment is "snowball" sampling, an approach that yields convenience samples that are not representative of the target population. To overcome this limitation, we used a newer form of chain-referral sampling, Respondent-Driven Sampling (RDS), that was developed by Douglas Heckathorn (1997, 2007 and others (Salganik and Heckathorn 2004;Salganik 2006). Respondent-Driven Sampling is based on a mathematical model of the social networks that connect survey respondents. The model is then used to adjust the sample estimates to reflect respondents' different probabilities of being captured by the survey.
We began the UWS sampling by identifying through our contacts a small set of workers (the "seeds") in the population to be surveyed in each city. After interviewing them, we gave these "seeds" a fixed number of coupons with which to recruit other workers, along with detailed instructions about who was eligible for the survey. The recruited workers then came to one of several survey sites (such as social service organizations, community colleges, and churches), where the coupon was recorded and the respondent was surveyed. The respondent was then given a fixed number of coupons with which, in turn, to recruit other workers. We repeated this process over a period of about six months in each city, with an average of 7.5 waves of recruitment per city; the final sample size was 4,387 workers. As recruitment progressed, the sample became increasingly diverse and eventually became independent of the initial seeds, a key feature of this method. On average, workers in the final sample each recruited two other workers (respondents were compensated for completing the survey and recruiting other workers).

Estimation
RDS uses information collected during the sampling process to quantify features of the social network connecting the respondents and then uses these features to make inferences about population composition (Salganik and Heckathorn 2004). A series of questions in the survey measured each respondent's personal network size, and we tracked the recruitment patterns that link respondents to one another during the survey fielding. This information allowed us to produce estimates that adjusted for each individual's probability of being recruited into the sample and the nonrandom patterning of social networks. (See Heckathorn 2007 for a description of the RDS estimator used for our analysis.) We did not detect strong network clustering on such critical dimensions as industry, occupation, employer, and most important, the workplace violations themselves. Our study, however, did identify strong network clustering within several racial/ethnic groups (the specific groups differed by city). While the RDS estimator adjusted for moderate nonrandom network clustering, the clustering was strong enough to warrant additional poststratification, for which we drew on external data. Specifically, we poststratified the RDS-adjusted estimates on race and nativity using the 2007 American Community Survey (ACS) and adjusted for undercounting of undocumented immigrants (see Spiller et al. 2010).
The final weight we used in our analysis consisted of three components: an RDS weight based on features of the social networks we sampled; a withincity ACS weight based on the relative sizes of race/nativity groups; 4 and a between-city ACS weight that adjusted for the relative size of each city's population. In this article we do not disaggregate results by city but instead analyze the combined sample; for an in-depth treatment of city differences, see Milkman, Gonzalez, and Ikeler (2011).

The Survey Instrument
The UWS is unique in that it measures a range of workplace violations using an original battery of detailed questions. Interviews typically lasted between 60 and 90 minutes, and we conducted them in 13 languages. The questionnaire did not rely on workers having any direct knowledge about their legal rights or on whether they had experienced a workplace violation. Instead, our strategy was to gather raw inputs from workers-the necessary data about their hours, earnings, and working conditions. We then used these data to determine whether or not a law had been violated by programming the complex matrix of federal and state statutes that set legal standards for wages, hours, meal and rest breaks, right to organize, workers' compensation, and other dimensions of the employment relationship. 5 Table 1 provides an overview of our sample, combined across the three cities and weighted to estimate the population characteristics of frontline workers in low-wage industries and occupations. We estimate that this workforce consists of more women than men; significant numbers of people of color, especially Latinos; and a full range of age groups and education levels, although about three-quarters attained only a high school degree or less. Also consistent with recent trends in the low-wage labor market, immigrants comprise a disproportionate share of this workforce: 31% documented foreign-born and 39% undocumented foreign-born. (The sizable number of the latter indicates our success in reaching this part of the labor market with the RDS methodology.) Given that we surveyed only workers in low-wage industries and occupations, it is not surprising that this workforce earns very low wages, specifically, a median wage (in 2008 dollars) of $8.00 an hour. Moreover, the wage distribution is quite compressed, with fully three-quarters of the workers earning less than $10.00 an hour.

A First Look at Industry Variation in Workplace Violations
In Table 2, we offer an overview of the industries represented in the UWS sample. We chose the level of industry disaggregation to maximize the number of industries while ensuring sufficient sample sizes for calculating industry-specific violation rates. Reflecting trends in the U.S. economy overall, much of this workforce holds jobs in the service sector-both private industries such as restaurants, retail, hotels, and car washes, and public industries or industries dependent on public funding such as K-12 schools, home health care, and nursing homes. But a sizable segment of the sample is also employed in manufacturing, construction, and warehousing. Most of the industries are low-wage; those that are not appear in the sample because the respondent was employed in a low-wage occupation in the sampling frame. So, for example, primary and secondary schools are not low-wage industries, but teacher assistant is a low-wage occupation and was included in the sampling frame. 6 The UWS collected information on a range of violations of employment and labor laws. We focus on five major violations where there are sufficient numbers of workers at risk of violations in the industries included in our sample. The concept of the risk set is critical. For many of the laws examined, not all workers in the sample were at risk of experiencing a violation. For example, only workers who actually worked overtime hours are at risk of experiencing an overtime violation. In other cases, some groups of workers are not covered by a particular employment or labor law and therefore cannot experience a violation of that law; for example, home health care workers are partially or wholly exempted from a number of federal and state workplace laws.
The five violations we analyze in this paper are as follows: 7 A minimum wage violation occurs when an employee is paid below the federal or state statutory minimum wage, whichever is higher. At the time of the UWS, the three states in our sample all had minimum wage rates higher than the federal standard: $7.15 in New York, $7.50 in Illinois, and $8.00 in California. 8 In our sample, all respondents were at risk of a minimum wage violation, and of these, 25.6% experienced the violation. An overtime violation occurs when an employee works more than 40 hours in a given week but is paid less than one-and-a-half times the regular rate of pay for the overtime hours. We also included in this measure violations of daily overtime laws, which in New York and California require some level of premium pay above a certain number of hours worked per day. In our sample, 24.9% of respondents were at risk of an overtime violation, and of these, 75.3% experienced the violation. An off-the-clock violation occurs when an employee works before or after a regularly scheduled shift but receives no pay for that time. In our sample, 24.8% of respondents were at risk of an off-the-clock violation, and of these, 70.6% experienced the violation. A paystub violation occurs when an employee does not receive documentation of his or her earnings and deductions-this documentation is legally required, regardless of whether the employee is paid in cash or by check.
In our sample, all respondents were at risk of a paystub violation, and of these, 56.7% experienced the violation. A meal break violation occurs when an employee does not receive an uninterrupted meal break during his or her shift-either the meal break was shortened or was not given at all, or the employee was interrupted by the employer or worked during the break. In our sample, 85.2% of respondents were at risk of a meal break violation, and of these, 72.7% experienced the violation.
All five measures show high rates of workplace violations in the UWS sample, and we focus here on the interindustry variation in those violation rates. In Figure 1, we show boxplots of the distribution of industry violation rates, defined as the percentage of at-risk workers who experienced a given violation in each industry. Substantial and statistically significant variation occurs in all five measures at the industry level, raising a host of questions about both the patterning and correlates of employer noncompliance.

Cluster Analysis
We began our analysis by asking whether it is possible to identify distinct patterns in the types of laws violated in different industries. In order to help reduce the complexity of the task, we used clustering methods combined with exploratory data analysis tools to strike a balance between preserving interindustry differences while still reducing the number of industries. 9 As part of that process, we tested various degrees of clustering aggregation for statistically significant differences in violation rates. Table 3 gives a summary of this analysis, which yielded three main clusters of industries: a high-violation cluster, a medium-violation cluster, and a moderate-violation cluster. For each cluster, the table lists its industries and describes its pattern of violations.
The five violations analyzed here do not move in lockstep. No cluster is high, medium, or low on every violation, and within each cluster there is variation across industries in the patterning of violations. Minimum wage and paystub violations seem to be strongest in differentiating the three clusters-meaning there are generally high scores on these two measures in the first cluster, medium scores in the second, and low scores in the third. By contrast, there is more variability on how industries rank on overtime, offthe-clock, and meal break violations, both within and between clusters. 10 One possible explanation for these patterns is that minimum wage and paystub violations are what one might call the most direct and overt forms of noncompliance-they are difficult to commit and hide in workplaces with computerized payroll systems. Table 3 offers support for this explanation. The high-violation cluster appears to have more informal industries (such as car washes, beauty salons, and private households). Overtime, off-the-clock, and meal break violations, by contrast, are more indirect, in that they are easier to commit and disguise when computerized payroll systems are in place. With this explanation in mind, consider the moderate-violation cluster in Table 3: Off-the-clock violations are above average for home health care, education, and nursing homes, as well as for hotels, convention halls, and stadiums. These industries are more likely to include large employers, 10 Tetrachoric correlations (appropriate for analyzing relationships between binary variables) reflect this patterning. The correlation between paystub violations and minimum wage violations is .56. Other correlations between the five violations are weaker, and in particular, it is clear that off-the-clock and meal break violations capture a somewhat different dimension of employer behavior. They are moderately correlated with only overtime violations. some of them either public sector or publicly funded, and paystub violations are uniformly low, signaling the use of automated payroll systems. In such a setting, it is considerably more difficult to pay workers less than the minimum wage, but having workers come in early or stay late without clocking in is more feasible.
Industry structure also appears to leave its stamp on the patterning of noncompliance. For example, meal break violations are high for caregiving industries such as domestic work and home health care (where often there is only one caregiver on-site), and in industries where production or service delivery is time-sensitive and not easily interrupted, such as restaurants and apparel manufacturing. Similarly, industries with variable scheduling are more likely to have overtime violations (restaurants, retail, beauty salons, car washes, dry cleaning, and home health care).

Modeling Industry Variation in Workplace Violations
Next we used regression models to examine the relationships among workplace violations, industries, and a series of worker, job, and employer characteristics. The outcome variable is a count of how many of five possible violations each worker experienced in the previous week: minimum wage, overtime, off-the-clock, paystub, and meal break violations. Table 4 shows that the average worker in our sample experienced 1.9 violations in the previous week, out of the 3.4 violations for which he or she was at risk. This outcome variable is a count variable, so we would expect it to follow a Poisson distribution where the mean and variance of the distribution are equal. Due to the right-truncation of our counts, the variance (1.7) is slightly smaller than the mean (1.9), but the outcome variable approximates a Poisson distribution well. We therefore estimated Poisson regression models with several controls included to adjust for our sampling methodology and the three-city sampling structure of the study. 11 In particular, our models include city fixed effects, which Milkman, Gonzalez, and Ikeler (2011) have shown are sufficient in the presence of other control variables (that is, city differences in violation rates are almost entirely a function of differences in demographic and industry composition). Our specifications also contain an offset variable that adjusts for the number of violations for which each worker was at risk. This means that the models are effectively estimating the proportion of possible violations a worker experienced. We estimate robust standard errors due to the lack of independence among sample respondents.

Model Results
Our modeling strategy was to regress the outcome variable-the number of violations experienced in the previous workweek-on a set of industry indicators and then to track the industry coefficients as we successively added worker, job, and employer characteristics into the model. We were especially interested in the extent to which the industry variation in violation rates decreases when other variables are included. 12 Table 5 shows the results of three regression models, with additional sets of variables added to the models as the columns progress to the right. All models fit reasonably well, with highly significant log-likelihoods and reasonable levels of approximate R-squared values. 13 11 Because the survey used Respondent-Driven Sampling, the models do not meet the standard regression requirement that sample members be drawn independently of one another (Spiller 2009). To adjust the model's point estimates, we included two adjustment variables: each respondent's self-reported network size and each respondent's recruiter's value for the outcome variable (adjusting for any correlation between the recruiter's and recruit's value on the outcome variable). 12 Ideally in this situation, one would fit hierarchical linear models (with a worker equation at the first level and an industry equation at the second), but our data contain too few industries and significant variation in industry-specific sample sizes.
13 Pseudo-R 2 values for Poisson models are highly unreliable, so the "approximate" R 2 values presented are from linear models that are identical to the Poisson models, except that the outcome variable is the square root of the count instead of the raw count.  We began in Model 1 by simply fitting the industry indicators. The table shows exponentiated coefficients, which are ratios of mean violation rates. For example, the mean violation rate for education, nursing home, and hospital industries is estimated to be 56.3% of the violation rate for the grocery store industry (chosen to be the contrast category because its raw violation rate is roughly the average across the sample). More generally, the results in Model 1 largely confirm the clustering analysis above by identifying the same industries whose violation rates are significantly higher than average and those whose violation rates are significantly lower than average. The one exception is food and furniture manufacturing, which in the model has a below-average violation rate but in the cluster analysis is placed with the medium-violation group. That there would be some differences is not surprising, since the cluster analysis groups industries on their violation patterns across five different measures, while the Poisson regression model predicts a single outcome, the total number of violations last week. Most important, Model 1 confirms the presence of significant interindustry differences in workplace violations, with an approximate R-squared of .21.
In Model 2, we added a group of worker characteristics measures. The raw correlations (available on request) show that gender, race/ethnicity, education, and English-speaking ability are each individually correlated with the violation outcome variable. But in the results for Model 2, those relationships are either insignificant or attenuated, which indicates high levels of collinearity. By contrast, immigration status is highly significant, even with other demographic controls in place. Estimated violation rates for U.S.-born workers and documented immigrants are significantly lower than those of undocumented workers. Moreover, the two estimates are very similar to one another, and this indicates that nativity alone does not lead to greater vulnerability to workplace violations; rather, it is lack of authorization to work in the U.S. that is key. Finally, workers' knowledge of the value of the statutory minimum wage plays a significant role, even with related controls, such as education and English-speaking ability, in place.
In Model 3, we added a group of job and employer measures. The first three variables added in this model are measures of the organization of work. In the results, hours worked last week is significant, but not in the expected direction: The estimated violation rate for part-time workers is moderately lower than for full-time workers. Further analysis suggests that full-time workers are more vulnerable to violations than part-time workers because they are more likely to have flat pay arrangements, even as their hours are frequently adjusted from day to day and week to week. By contrast, part-time workers are more likely to have a set hourly wage and a fixed schedule. It is also surprising that there is no significant effect of working in a short-term job (i.e., the respondent had a short-term job, was a temp worker, or found the job as a day laborer), both in this model and without controls. We speculate that the absence of a significant effect here is due to the overall ratcheting down of labor standards in low-wage industries, which flattens out differences that would be more marked if the full labor market had been sampled. More in line with expectations, respondents had significantly higher violation rates when they had two or more employers in the previous workweek. In our sample, this variable is a measure of the organization of work; that is, respondents with more than one employer worked in jobs distinguished by decentralized employment relationships. Almost all were housekeepers, childcare workers, home health care workers, and residential construction workers. Note that this is not simply an occupation effect because many other workers in these occupations had only one employer.
The final three variables we added in Model 3 were employer measures. Not surprisingly, employer size has a strong effect. Large employers have an estimated violation rate that is two-thirds of that of small or medium employers. 14 The workplace benefits variable is an index that counts whether the employer offered the respondent health insurance, gave the respondent paid sick leave or paid vacation leave, or ever gave the respondent an annual raise. Workers whose employers provided two or more of these benefits had a violation rate significantly lower than those whose employers provided none or only one. The final variable we added in this model was pay type, a key measure because many workers in the sample were paid in nontraditional ways: flat weekly or daily pay, project-based pay, piece-rates, and so forth. Compared to hourly pay rates, such nonstandard pay systems are associated with significantly higher violation rates and enable the shrouding of actual wage rates (as with flat weekly wages for long hours of work).

Decomposing the Variance in Industry Effects
Having analyzed the correlates of workplace violations across a range of worker, job, and employer characteristics, we now return to our original question: What explains the significant interindustry variation in violations? We answer this question in two ways.
First, we return to Table 5 and scan across Models 1 to 3 in order to assess how the size and significance of the industry indicators change with the addition of successive sets of covariates. Going from Model 1 to Model 2 (where worker characteristics are added), we see generally mild attenuation of the industry effects. The main change is that the construction and apparel manufacturing indicators become insignificant, which means that their higher-than-average violation rates were a function of demographic composition (in the case of construction, this is almost entirely the result of the industry's high percentage of undocumented workers).
Going from Model 2 to Model 3 (where job and employer characteristics were added), we see more marked changes. Many of the estimated industry effects become weaker, and several industries lose significance altogether. For example, the home health care industry indicator becomes insignificant after the addition of employer size, which means that the lower-thanaverage violation rate for home health care is primarily a function of large firm size. The loss of significance for private households and the personal services industries is driven largely by the addition of the pay type variable, indicating that part of what distinguishes high-violation industries is their reliance on nonstandard pay systems.
Our initial observation, then, is that job and employer characteristics appear to account for a greater portion of the interindustry variation in violations than do worker characteristics. In Table 6, we conducted a formal decomposition by estimating the marginal strength of the industry indicators in the presence of other covariates.
In the first row, when only the industry indicators were included in the model, the industry chi-square statistic is 471.63. This is the statistic from a Wald test of the null hypothesis that all industries have equal coefficients in the model, which is rejected. In the second row, we added the industry indicators after the worker characteristics variables were introduced; the industry chi-square statistic falls to 313.97, a 33% drop in the direct-industry effect. In the third row, we added the industry indicators after the job and employer variables were introduced. In this case, the industry chi-square statistic falls to 83.27, a much bigger drop of 82% in the direct-industry effect, which is roughly 2.5 times the drop in the industry effect after worker characteristics were added. 15 In the final row, we added the industry indicators after all worker, job, and employer variables were in the model. Here, the industry chi-square statistic falls to 75.62, which represents an 84% drop in the direct-industry effect.
In sum, we have been able to account for a significant portion of the interindustry variation in violations observed in the UWS sample. Moreover, our decomposition suggests that differences in job and employer characteristics are a stronger source of industry variation in violation rates than are differences in worker characteristics-on the order of 2.5 times as strong. Note also that some of the industry variation remains unexplained; we discuss additional, unexamined sources of variation in the next section.

Summary and Discussion
Historically, employers' violations of minimum wage, overtime, and other workplace laws have been understudied because of lack of data and accurate measures. But an emerging field of research is beginning to use administrative data and original surveys to document high and persistent rates of noncompliance in the low-wage labor market. In this article, we analyzed the 2008 Unregulated Work Survey to identify a new dimension for empirical and theoretical study: the substantial variation at the industry level in the extent of employer noncompliance with workplace laws.
Using clustering techniques, we identified three main groups of industries that differ in both the prevalence and the mix of workplace violations. While some industries stand out as committing the most overt violations (paying less than the minimum wage and paying under the table), other industries commit more indirect violations that are easier to disguise (overtime, off-the-clock, and meal break violations). We also found that industry structure leaves its mark on the pattern of violations; for example, meal break violations are high in caregiving industries such as domestic work and home health care, where often there is only one worker in the workplace.
Using multivariate regression, we then found support for our initial hypothesis that worker and employer characteristics each account for a large and statistically significant portion of the interindustry variation in workplace violations. But while differences in workforce composition (especially immigration status) are important and should not be neglected, it is differences in job and employer characteristics that play the stronger role in accounting for industry differences in noncompliance-on the order of 2.5 times as strong as workforce composition. This is an important empirical benchmark for future research on the drivers of wage theft and other workplace violations, because in this still-young field a first instinct might be to attribute industry effects to supply-side explanations (similar to the initial waves of research on interindustry wage differentials in the late 1980s and early 1990s). And to be clear, workforce substitutability, legal vulnerability, and lack of (informal) bargaining power are crucial to understanding industry effects. But our findings suggest a rich agenda for exploring demand-side explanations, including such industry-level and firm-level dynamics as the state of industrial relations, product market structure, managerial strategy in the organization of work, and, more generally, prevailing modes of economic restructuring.
Like other labor market researchers, we were limited in our exploration of demand-side explanations by the job and employer measures in our data set, though our findings do provide baseline estimates. Surprisingly, we found only partial confirmation for the predicted role of nonstandard work arrangements. Part-time, temporary, and short-term jobs did not show significantly higher violation rates and therefore did not help to explain variation across industries in noncompliance. We suspect this finding results from the fact that our sampling universe was restricted to the low-wage labor market, which flattened out differences in work structures that may be more marked across the entire labor market. That said, our measure of decentralized employment relationships was a strong predictor of higher violation rates. This of course raises the question of the role of subcontracting, which was not measured in our survey.
By contrast, three employer measures were all highly significant in our models, as expected, and accounted for much of the demand-side portion of the interindustry variance in noncompliance. It should come as no surprise that industries with larger firms had lower violation rates than industries with a preponderance of small firms. In addition to a true firm-size effect, however, this variable likely also captures the effect of such related but unmeasured firm characteristics as the size of profit margins, pricesetting power, capital-labor ratios, and organizational pay equity concerns, all of which we would expect to correlate with industry-level compliance rates (Katz and Summers 1989;Weil 2005;Osterman 2011). The provision of workplace benefits was strongly correlated with lower industry violation rates; we interpret this variable as signaling the presence of a business model that favors competition on the basis of quality goods or services and high productivity over cutting labor costs (Cappelli et al. 1997). Finally, the use of nonstandard pay systems was a strong correlate of higher industry violation rates. We interpret this variable as measuring informality and, in particular, the probable absence of centralized human resource departments and computerized payroll systems. The use of nonstandard pay systems may also be capturing actual intent to violate the law on the part of the employer, because pay arrangements such as flat weekly wages easily obscure the true wage rate when hours vary.
If we step back, our assessment is that in future research on workplace violations, researchers will need to conduct more original surveys and develop better measures, especially at the firm and industry level. Otherwise, it will not be possible to sort through the validity of competitive and institutional explanations (Osterman 2011). But based on the benchmark findings in this article, our interpretation is that industry variation in employers' compliance with workplace laws is shaped by both product market characteristics, which affect ability to pay, and institutional characteristics that shape managerial strategy. Together these factors interact with labor-supply characteristics to produce a risk matrix of noncompliance. Research on how industry-level norms about violating the law are generated and diffuse over time will be of particular interest (Jones, Ram, and Edwards 2006;Bernhardt et al. 2007). At the same time, it is important to note that all industries are currently facing a very weak penalty and enforcement regime, and in our view this results in an overall high rate of noncompliance, around which industries vary based on their demand-and supply-side characteristics.
We close by suggesting several topics that require particular attention going forward. Perhaps most important, researchers need to devise strategies for measuring the use of subcontracting at the industry level. 16 Janitorial and security services, food services, industrial laundries, warehousing contractors, and waste management companies are industries that have grown significantly over the last several decades, as large firms and public institutions have externalized those functions to lower-wage subcontractors-thereby increasing the risk of wage and hour violations (Zatz 2008;Dube and Kaplan 2010). In a similar vein, Weil (2011) has analyzed data on franchising in industries such as fast-food restaurants and finds higher wage and hour violations among franchises compared to company-owned establishments. To be clear, subcontracting is not always driven by the desire to cut labor costs (Abraham and Taylor 1996), and the outcomes for workers are not predetermined (Erickcek, Houseman, and Kalleberg 2003). But the economy-wide spread of subcontracting is undoubtedly central to the story of workplace violations in the twenty-first century, precisely because the practice can be used to create legal distance between employers and workers.
Researchers will also need to give attention to analyzing care work industries that depend on public funding streams, such as home health care, nursing homes, and childcare. Analysts have documented that public funding for these services (in the form of Medicaid, Medicare, and Child Care Development Block Grants) falls far short of need and therefore places significant constraints on the wages of caregiver jobs (Dresser 2008;McGrath and DeFilippis 2009). Further complicating the picture is that some forms of care work are either partially or wholly exempted from employment and labor laws (National Employment Law Project 2011). As a result, workers in 16 We attempted to identify subcontracted jobs in our survey but abandoned the effort because of workers' inability to accurately identify whether their employer was a contractor or not. Using similar logic as Dey, Houseman, and Polivka (2009), however, we were able to conduct an indirect test for janitors. In the building services industry (where the employer is likely a contractor company), we found that the average violation rate for janitors was 48% (using the measure defined in Table 4). By contrast, for janitors employed in the education, finance, and hotel industries (where they are more likely direct employees), the average violation rate was 28%. See Nissen (2004) for similar findings. these industries are some of the lowest paid in the U.S. economy and are at greater risk of wage and hour violations. In addition, the chronic underfunding of these public goods generates unregulated markets. The "gray market" in home health care, for example, is by some accounts as large as the publicly funded market. Families who have exhausted or are not eligible for Medicaid coverage turn to the gray market to hire workers directly, often at substandard wages (Dawson and Surpin 2001).
Finally, a range of industries, including beauty and nail salons, dry cleaning, car washes, ethnic retail, and independent restaurants are highly informalized, with disproportionate numbers of small firms that employ some or all of their workers off the books, face razor-thin profit margins, and may not themselves be registered businesses. The drivers of this informality are manifold. Researchers have noted that the growth of high-income households in large U.S. cities has increased the demand for labor-intensive personal services (Sassen 2001). But there is also a flip side of this dynamic that requires research and analysis, wherein growing numbers of workers earning very low wages need inexpensive goods and services. What emerges is an entire subeconomy of food vendors, dollar vans, restaurants, 99-cent stores, and informal childcare providers, who earn less than the minimum wage because their customers or relatives are too poor to pay more (see Wilson 2011). The payment of subminimum wages thus increases the demand for below-market goods and services, which can be delivered only through substandard working conditions.
In sum, our findings call for a deeper investigation of how noncompliance with employment and labor laws is becoming a key feature of employers' competitive strategy at the bottom of the U.S. labor market. In a range of industries, the evasion and outright violation of minimum wage, overtime, and other laws is creating new industry conventions that normalize substandard jobs. These practices are not confined to marginal businesses. Instead, violations appear to be occurring in industries where economic restructuring over the past three decades has been most intense, resulting in the dominance of low-wage, cost-cutting business models over high-wage, high-productivity strategies. Employers increasingly are organizing work in ways that obscure the lines of legal accountability in the employment relationship, and weak penalties and public enforcement contribute to, rather than counteract, this trend. If, as we suspect, these practices are becoming more entrenched, further development of new data sets and analytic tools will be of the utmost importance.