Workplace Compliance With the Law: The Case of the Family and Medical Leave Act

The FMLA

Following years of debate and delay, President Clinton signed the FMLA of 1993 into law. The FMLA was applauded by its supporters as the key employment policy that finally addressed women’s and men’s attempts to balance families and jobs. The FMLA requires employers to offer up to 12 weeks of unpaid leave not only for employees’ personal illness (maternity disability and workers’ own serious medical conditions) but also for their family caregiving responsibilities (care of a newborn or newly adopted child or care of seriously ill children, spouses, or parents).

Two decades after the passage of the FMLA, we examine a very basic question: Do employers comply with the requirements of the act? Estimating and understanding employer compliance with this policy is important in two respects. First, as the primary federal policy that addresses workplace accommodation of women’s and men’s caregiving responsibilities, the consequences of noncompliance are high for affected families, and may be especially high for families with fewer personal resources. More generally, examining employer compliance with this policy has implications for regulatory policy in general. If employer noncompliance is high for the FMLA—a policy with few monetary costs and fairly simple guidelines—this might suggest that more complex or costly policies may face even greater noncompliance, and need greater enforcement efforts.

In this article, we estimate compliance rates using data on the type of leave offered by firms that was reported by human resource (HR) representatives from a sample of private-sector firms. We account for nonresponse in these estimates of overall compliance. Next, we examine the characteristics of organizations in each of three groups (compliant, noncompliant, and nonresponsive) to see whether organizations that are nonresponsive are more similar to those that comply or do not comply with the act.

Employer Compliance and the Meaning of Missing Data

Most research on family leave since the passage of the FMLA has focused on the prevalence of leave taking among individuals. Only a few studies, described in the following, look at compliance from the perspective of organizations and employers. One reason for the inattention to employer compliance is that the existing studies show high rates of compliance with the FMLA. The literature relies on survey-based measures of employer compliance with the FMLA that are generated from responses to questions asking a representative of the firm how much leave is offered to employees in different circumstances.

The first FMLA study, conducted by Westat for the Department of Labor (DOL), surveyed HR directors or the person “responsible for the company’s benefits plans.” The study concluded that FMLA compliance rates among private-sector establishments covered by the act were more than 90% for each type of allowed leave (medical, maternity disability, newborn care, adoption, and family sickness) and were about 88% for overall compliance with all types of leave (Commission on Family and Medical Leave, 1996). These early estimates of compliance were quite high, and subsequent studies relied on these estimates in making assumptions about how to analyze and interpret data. However, in a follow-up DOL survey conducted in 2000, 83.7% of companies reported offering at least 12 weeks of leave for all of the allowed reasons (Cantor et al., 2001) and in the 2012 DOL follow-up survey, 80% of covered worksites report allowing leave for all qualifying FMLA reasons (Klerman, Daley, & Pozniak, 2012). The decline in estimated compliance is unexpected, and raises questions about changes in awareness or enforcement of the FMLA, as well as the credibility of the estimates themselves.

The credibility of estimates based on these data depends on the design of survey questions, the survey and item response rates, and the truthfulness and knowledge of respondents. Kelly (2010) suggests that the DOL estimates of compliance rates may be inflated because of several design features of their survey. The questions cued respondents about the length of leave required by law, and the surveys asked about “job-protection during leave” and “length of leave” in a separate set of questions. In addition, this survey asked about parental leave only in a gender-neutral way, which might obscure important differences in the provision of leave for men versus women. The 2012 version of this survey has the same limitations, and in addition, had a very low response rate (20.9%), likely due to a short field period (Klerman et al., 2012). In her own survey of 389 establishments in 1997, which did not cue respondents about leave length, Kelly found that between 42% and 50% of covered establishments were not compliant with the FMLA. The exact percentage depends on how she treated data on those who did not answer all relevant questions—a set of decisions we return to later.

Using a different national survey, Galinsky, Bond, Sakai, Kim, and Giuntoli (2008) estimate that 81% of private employers are compliant with the FMLA. In that study, the survey instrument was administered to HR representatives, did not cue on the length of leave allowed by the FMLA, and asked about leave for maternity, paternity, adoption, and family sick leave. However, the compliance estimates were derived by using only establishments with no missing data. In this paper, we reanalyze data on FMLA compliance from the survey used by Galinsky et al. (2008). The results of our analysis cast doubt on the validity of the assumption that establishments with missing data on FMLA related questions have the same compliance rates as those with complete data, which is an assumption that underlies most of the existing literature.

Like the previous work, our analysis examines responses of HR personnel from a representative sample of firms. Dobbin (2009) argues that HR personnel are influential agents who interpret and translate legal policy into organizational compliance. Even though some argue that HR departments often offer symbolic or ceremonial responses to legal threats while acting as vehicles to circumvent the law (Edelman, 1992; Edelman, Fuller, & Mara-Drita, 2001), it does seem clear that HR departments are a central node in organizational compliance with employment regulations. Furthermore, given the legal requirements, HR departments have an incentive to report compliance, and little incentive to report noncompliance. Thus, we argue that responses to surveys from HR offices are likely to provide the best available evidence on organizational compliance, and the most conservative estimates of noncompliance.

Like civil rights and environmental legislation, the FMLA is an example of “regulatory” legislation, or of “the legal system … taking the initiative directly to modify organizational behavior” (Edelman & Suchman, 1997, p. 483). In principle, regulatory legislation seeks to coerce organizations to act in ways that they would not otherwise. But firms do not always comply with such legal controls. A number of theorists have suggested that organizations have a substantial capacity to resist and make their own varied responses to federal regulations in a variety of policy domains (Edelman, 1992). Oliver (1991) discusses a range of strategic responses—from clear defiance to acquiescence—that organizations use in response to legal mandates. Because HR managers are responsible for interpreting and translating policy into practice, those who lack the knowledge necessary to implement legal mandates may actually be avoiding or defying legal mandates. Based on the DOL’s 2012 Worksite and Employee Survey, Appelbaum (2013) reports that “nearly a quarter of covered work sites employing nearly a tenth of all workers (9 percent) do not know that the FMLA applies to them.” We hypothesize that HR representatives who report that they “don’t know” or who refuse to provide a response to questions about whether their organizations provide family leaves may indicate not only a lack of knowledge but also a type of noncompliance.

Distinguishing lack of knowledge from hidden noncompliance is a difficult task. We begin to make progress on this task by examining those respondents who provide only partial responses to questions about the FMLA and by asking whether they comply at similar rates to full respondents. Similar rates of compliance among partial responders and full responders would suggest that nonresponse does not necessarily imply noncompliance, whereas different rates may suggest nonresponders are more noncompliant. We continue by analyzing those firms with full or incomplete response separately to produce more trustworthy estimates of overall FMLA compliance rates, and also to gain a better understanding of what nonresponse may indicate about firms’ true FMLA compliance status.

Comparing Organizations: Compliance, Noncompliance, and Nonresponse

Although researchers have not examined nonresponse or looked at its relationship to noncompliance, a series of studies on noncompliance do suggest that social pressures internal and external to organizations prod employers to comply. First, although findings are somewhat mixed, scholars suggest that both female managers and female workers are key constituents for “family-friendly” policies, given that women are still more often responsible for caregiving than are men (Albiston, 2007; Bianchi, Robinson, & Milkie, 2006; Sarkisian & Gerstel, 2012). Second, although the findings are again somewhat mixed, a workforce with higher status or more bargaining power, by virtue of being salaried (rather than hourly) or with union representation (Appelbaum & Milkman, 2008; Deitch & Huffman, 2001; Galinsky, 2001; Glass & Fujimoto, 1995; Milkman, 2007), may compel higher rates of compliance. Third, organizational characteristics—like age and size—may be important for compliance. Just as companies established after the passage of the Civil Rights Act show higher levels of gender integration (Stainback, Tomaskovic-Devey, & Skaggs, 2010), organizations founded after the passage of the FMLA may be more likely to comply with its mandate. The conclusions from research on the effect of employer size on FMLA compliance are also mixed (Kelly, 2010), but overall, it suggests that organizational growth leads to more bureaucratization, which is associated with greater vulnerability to enforcement and standardization of HR practices, leading to greater regulatory compliance (McTague, Stainback, & Tomaskovic-Devey, 2009). Finally, prior scholarship suggests that organizational compliance is subject not only to internal constituencies but also to pressures external to the organization—whether from the state, similar organizations (see Albiston, 2007), or economic conditions (Clinard & Yeager, 1980; Kagan et al., 2003). With this reading of the literature in mind, we might expect that more profitable, larger organizations, founded after the passage of the FMLA, with greater proportions of female managers and workers, and higher status workers, will be more likely to respond to questions about the FMLA as well as to comply with the act. We include these characteristics in our empirical analyses to examine whether nonresponding organizations are more similar to compliant or noncompliant organizations.

Data and Methods

We use data from the 2008 National Study of Employers (NSE) conducted by the Families and Work Institute (Galinsky et al., 2008). Surveys were administered to HR directors or “the person primarily responsible for Human Resources” from a representative sample of 1,100 private-sector employers with 50 or more employees. The survey included only private-sector worksites because all public agencies are covered by the FMLA. ¹ Because the FMLA is limited to employers with 50 or more employees within a 75-mile radius, this survey may include some establishments that are not regulated by the FMLA. However, Galinsky et al. (2008) address this concern by comparing companies that have one worksite to those that have more than one worksite (and therefore are less likely to be covered by the FMLA) and reveal no difference in compliance rates. Employers were selected from Dun & Bradstreet lists using a stratified random sampling procedure in which selection was proportional to the number of people employed by each company to ensure a large enough sample of large organizations. In their comparisons of different sampling frames for organizational studies, Kalleberg, Marsden, Aldrich, and Cassell (1990) rate the Dun & Bradstreet listing as one of the most efficient sources of data, although they do show a higher likelihood than direct enumeration of missing very new businesses (formed in the last couple of years).

Galinsky et al. (2008) report that the firm level response rate for the survey was 44%, which is lower than ideal but not atypical for surveys of this size. All of the analysis we report is weighted to reflect the distribution of firms with more than 50 employers in the United States. It is worth noting that survey nonresponse provides another threat to the validity of the estimates of compliance rates both in this paper and in the existing literature. If the firms that do not participate in any aspect of the survey have different compliance status and item response patterns than the firms that do participate in the survey then the true compliance rates could be different from the estimated rates. Here, we assume that the firms do not systematically opt out of the survey on the basis of their FMLA compliance status (which is a small part of the overall survey). However, we might also reasonably suggest that if there is a bias to the survey nonresponse, it is more likely that those firms who are noncompliant or simply ill equipped to answer questions about their employment practices are more likely to opt out, in which case this survey might systematically overestimate compliance.

Employer Characteristics Related to Compliance, Noncompliance, and Nonresponse

In our analysis of similarities between organizations characterized by compliance, noncompliance, and nonresponse, we include characteristics of the workforce and the organization, and external pressures that may affect compliance. We also include information on other benefits offered by the employer.

Workplace characteristics that may affect compliance and noncompliance include the gender composition of management and the general workforce. We use an item asking whether there are any women in senior management positions (defined as CEO, managing partner, president, chair or vice chair of board, COO, CFO, other officers reporting directly to chair, president, CEO or COO, or board of directors). Respondents were also asked to estimate the percentage of their employees who were (a) women, (b) salaried, and (c) union members. Respondents who had difficulty estimating the precise percentage were asked a series of questions to approximate the percentages. This process resulted in some data heaping at the 0, 25, 50, and 75 percentage points, making it unwise to use these as continuous variables in our analyses. Instead, we looked at bivariate and multivariate distributions of these variables with the compliance variables to pinpoint the most effective ways to dichotomize these variables. In the analysis, we use dichotomous variables indicating whether (a) more than 50% of the firm’s employees were women, (b) more than 25% of the firm’s employees received a salary rather than an hourly wage, and (c) more than 25% of the firm’s employees were unionized.

We also include employer size and longevity. Employer size is coded in three categories: 50–99 employees, 100–249 employees, and 250 or more employees. Respondents were also asked how long their organization had been in operation, and answers were recorded as years of operation. Since the FMLA became law 15 years prior to the data collection, we defined a dummy variable equal to 1 if a firm was more than 15 years old. This distinguishes firms that have always been subject to the FMLA from firms who experienced a regulatory change. We also tested employer longevity as a continuous variable, to look for nonlinear relationships, but these terms were not significant.

Each employer’s industry was coded as one of the five following categories: goods producing; professional services; wholesale and retail sales; finance, insurance, and real estate; and other services. Ideally, we would use finer distinctions between industry (separating retail and wholesale sales, for example), but this information is not available in the data.

We also include possible external influences on employer behavior. To help account for recent economic conditions, we included an item from the survey that asked whether the company had experienced downsizing in the last 12 months. We also tested a variable that measured relative economic performance, asking whether the employer was doing financially “better than, about the same, or worse than competitors,” but found no difference in results between these two economic conditions variables and thus used only the downsizing measure in our analysis.

We use information on state policies from the U.S. Department of Labor and Women’s Bureau (1993) to code employers in states that had family leave policies comparable to the FMLA in effect for private-sector employees prior to the FMLA. We expect that firms in states that had already been subject to a similar mandate would be more likely to comply than those in states where the FMLA represented a new policy. ⁴

We also include two indices of benefit generosity of the employer. The first is a standardized scale combining 12 benefits related to health insurance and economic security, including offering a group health plan for individual workers, families, and unmarried partners; employer contribution to all or part of the health insurance premium; presence of pensions or other forms of retirement savings; employer contributions to employee retirement funds; and other financial benefits offered to employees (Cronbach’s alpha = .72). The second benefit index measures generosity of family-friendly benefits, including 26 items about various resources and benefits related to child and elder care (Cronbach’s alpha = .81). While these types of benefits differ from the provision of family leave in that they are not required by law, employer generosity on these items may reflect their status as proponents of health, economic, and family benefits for employee well-being and loyalty. Likewise, it may capture unmeasured characteristics about the negotiating power of the employees, which might affect the firm’s FMLA compliance and the type of fringe benefits it provides employees. We have included a list of the specific items in these indices in the Appendix.

To further understand the relationship of noncompliance with the FMLA and nonresponse, we also include a dummy variable indicating whether the respondent did not know or refused to answer any of the items about health and economic security benefits. Given that these are the most prominent types of benefits offered by employers and that they are not required by law, we suggest that an inability to answer these questions may truly indicate a lack of knowledge rather than their defiance of the law.

Given the emphasis in this article on the meaning of nonresponse, it is important to note that none of our independent variables had a missing data rate above 1.5%. In our multivariable analysis, we drop cases that are missing any of our independent variables, which results in dropping 4.5% of our sample from the analysis. Missingness on the independent variables was not significantly related to missingness on the FMLA variables.

Table 1 presents the weighted percentages of employers in each category of these independent variables.

OPEN IN VIEWER

Table 1.

Weighted Percentages of Organizations, by Independent Variables (N = 1,100).

Variable	Percentage
Women in senior/top positions	72.1
More than 50% women employees	43.6
More than 25% salaried employees	54.5
More than 25% unionized employees	8.4
Number of employees
50–99	53.0
100–249	21.9
250+	25.1
More than 15 years in operation	81.0
State had comparable leave policy prior to FMLA	18.8
Downsizing in last 12 months	24.1
Industry
Goods producing	25.1
Professional services	33.6
Wholesale and retail sales	18.2
Finance, insurance, real estate	7.7
Other	15.4
Low benefit knowledge (any health/economic benefits missing)	13.7

Analysis Strategy

Our analysis consists of two parts. In the first part, we estimate FMLA compliance under weak assumptions about the distribution of the missing data. We examine complete and incomplete data on the four types of family leaves and estimate partially identified models of FMLA compliance rates in the U.S. economy.

In the second part of the analysis, we estimate multinomial logistic regression models to examine the effect of internal organizational characteristics and their external conditions on FMLA compliance, noncompliance and nonresponse. Our primary aim in this analysis is to examine whether nonresponders are more similar to those that comply or do not comply in their organizational characteristics. All analyses were conducted using Stata 11.

Modeling Compliance Under Weak Assumptions: Partially Identified Models

Table 2 displays the frequency of responses and missing data for each of the four types of leave: maternity, paternity, adoption, and family sick leave. For each type of leave, we present the weighted percentages of compliance among all respondents (in the second column) and the weighted percentage of compliance among only those who answered the question, with missing data excluded (in the third column). Employers show the highest compliance and lowest missing data rates for the question on maternity leave, and the lowest compliance and highest missing data rates for the question concerning paternity leave. The differences between maternity and paternity leave are statistically significant.

OPEN IN VIEWER

Table 2.

Frequencies and Weighted Percentages of Establishments Reporting at Least 12 Weeks of Leave for Different Reasons (N = 1,100).

Length of leave	Maternity leave			Paternity leave			Family sick leave			Adoption leave
	Frequency	Weighted %	Exclude missing weighted %	Frequency	Weighted %	Exclude missing weighted %	Frequency	Weighted %	Exclude missing weighted %	Frequency	Weighted %	Exclude missing weighted %
Fewer  than  12 weeks	122	12.7	15.0	196	18.9	23.6	122	14.8	16.0	154	12.9	18.6
12 weeks  or more	859	72.5	85.0	736	61.3	76.4	819	64.7	84.0	770	68.2	81.4
Refused/  Don’t  know	119	14.8	168	19.8	159	20.5	176	18.9

Next, we turn to full compliance with the FMLA, to look at those companies which report compliance for all four types of leave versus those which do not. Of the sample of respondents, 69.8% were complete responders (answered all four questions), 19.5% were partial responders (answering between one and three of the four questions), and 10.7% were complete nonresponders (answering none of the questions). Estimating compliance based on the complete responder group alone relies on the assumption that data are missing completely at random (MCAR). If this were true, we would theoretically expect similar compliance rates among all three groups: complete responders, partial responders, and complete nonresponders. Confirming this assumption—which is central to the existing literature on FMLA compliance—is difficult because compliance status is not known for nonresponsive firms. Although we do not know anything about compliance for the complete nonresponders, we can compare complete responders with partial responders.

Recall that full FMLA compliance requires firms to offer at least 12 weeks of leave for each of the four categories of leave. Among complete responders, 19.1% report allowing fewer than 12 weeks of leave for at least one of the four reasons. These firms are noncompliant. Using the data available for partial responders, 60.9% of them report providing fewer than 12 weeks of leave for at least one of the four reasons. We can establish that these firms are noncompliant with the FMLA even though we cannot observe their answers to one or more of the FMLA questions. This is because compliance is an all or nothing concept; one observable instance of noncompliance is enough to establish overall noncompliance. Partial responders are 40 percentage points more likely to be noncompliant with at least one element of the FMLA than complete responders. This is very clear evidence that compliance rates are not the same among complete responders and partial responders. This difference shows that the data do not support the MCAR assumption, and we therefore find it imprudent to simply exclude those cases with missing data. Instead, we use partial identification methods which impose weaker distributional assumptions to estimate population parameters, compliance rates in this case (Manski, 2007). Using these methods, we can take advantage of the information we do have, and consider all possible combinations of the missing data, without imposing strong assumptions about the distribution and independence of missing data. The goal of our analysis is to come up with upper and lower bounds on the level of compliance that we would have observed in the survey data if all of the respondents had answered all of the questions. The width of the bounded range of values provides a direct measure of the degree of uncertainty generated by the missing data. An important point is that the uncertainty characterized by the bounded range has been present in previous studies: Researchers have simply “resolved” the uncertainty using the MCAR assumption. It is also important to note that the partial identification strategy pursued here does not characterize uncertainty associated with other potentially important concerns like the extent to which respondents answer the questions truthfully and accurately.

We begin by defining simple binary measures of compliance associated with each of the four types of leave presented in the survey. To this end, let N m = 1 if the firm provides the mandated maternity leave, N p = 1 if the firm provides the mandated paternity leave, N a = 1 if the firm provides the mandated adoption leave, and N f = 1 if the firm provides the mandated family sick leave. These four indicator variables each take values of 0 if the firm is noncompliant with the specified element of the law. Overall compliance occurs if the firm is compliant with all four elements. Let N fmla = 1 if the firm is compliant with the requirements of the FMLA. Then the overall compliance status of the firm is simply the minimum of the four measures: N fmla = Min ( N m , N p , N a , N f ) . The resulting indicator, N fmla , is a dichotomous variable, with a value of 1 if the firm is compliant and a value of 0 if the firm is noncompliant. In this notation, the compliance rate is ⪻ ( N fmla = 1 ) , which is simply the proportion of firms that comply with all four requirements of the FMLA.

The data suffer from nonresponse when the value of one or more of the four scenario variables is not observed. Let R denote a respondent’s level of response to the four scenario questions, R = Complete if the respondent answered all four questions, R = Partial if the person answered at least one but less than four of the scenario questions, and finally, R = None if the person did not answer any of the four questions.

The problem created by incomplete response is easy to see when the overall compliance rate is expressed as a weighted sum of compliance among complete, partial, and nonrespondents. We use the law of total probability to write

⪻ ( N fmla = 1 ) = ⪻ ( N fmla = 1 | R = Complete ) ⪻ ( R = Complete ) + ⪻ ( N fmla = 1 | R = Partial ) ⪻ ( R = Partial ) + ⪻ ( N fmla = 1 | R = None ) ⪻ ( R = None ) .

Comparing Organizations by Compliance, Noncompliance, and Nonresponse

In this part of our analysis, we examine characteristics of organizations that predict compliance, noncompliance, and nonresponse. The goal here is not only to examine what predicts compliance but also to examine the extent to which nonresponders resemble either compliant or noncompliant organizations, in order to understand the potential meanings of nonresponse.

We use multinomial logistic regression to assess the relationship between our set of covariates and the classification of firms into three categories: (a) FMLA compliant firms, (b) FMLA noncompliant firms, and (c) firms with nonresponse. To construct the three category-dependent variable we coded those companies with complete responses on all four types of leave as known compliant (n = 684) or noncompliant (n = 152) based on those responses. Next, we classified partial responders and complete nonresponders as nonresponders (n = 264). Note that in the prior set of analyses, we coded some partial responders as noncompliant based on available information because we were interested in developing the most accurate estimates of noncompliance. In this set of analyses, we are interested instead in assessing the meanings of nonresponse; therefore, we code partial responders as nonresponders because they did not answer all the relevant questions concerning FMLA compliance. Nonresponse is the base category in our model, and we report the estimated coefficients and robust standard errors predicting the likelihood of being noncompliant or compliant versus nonresponsive.

The results of this analysis are given in Table 3. The first column of results examines the effects of constituency, organizational, and external pressures on the likelihood of noncompliance versus nonresponse. The second column compares compliance with nonresponse on the same set of variables.

OPEN IN VIEWER

Table 3.

Weighted Multinomial Logistic Regression of Noncompliance and Compliance Versus Nonresponse on Independent Variables.

	Noncompliance (vs. nonresponse)		Compliance (vs. nonresponse)
	Coefficient	Robust std error	Coefficient	Robust std error
Women in senior/top management	−0.45	0.28	−0.19	0.22
More than 50% women employees	0.15	0.30	0.65**	0.23
More than 25% salaried employees	0.38	0.25	0.92***	0.20
More than 25% unionized  employees	−0.51	0.47	0.01	0.33
Number of employees
50–99	(excl)
100–249	0.76**	0.29	0.85***	0.22
250+	0.70*	0.29	1.00***	0.22
More than 15 years in operation	0.25	0.33	0.43	0.24
State had comparable leave  policy prior to FMLA	0.67*	0.33	0.57*	0.27
Downsizing in last 12 months	0.08	0.29	−0.08	0.22
Industry
Goods producing	(excl)
Professional services	0.05	0.36	0.19	0.28
Wholesale and retail sales	0.34	0.37	−0.15	0.28
Finance, insurance, real estate	0.77	0.56	0.27	0.46
Other	0.06	0.39	−0.36	0.29
Health & economic  benefits generosity	0.30	0.27	0.61**	0.20
Family-friendly benefits  generosity	−0.05	0.31	−0.28	0.26
Any health/economic  benefits missing	−1.27***	0.40	−1.34***	0.25
Constant	−1.20*	.49	−0.51	.34
N = 1,047
Pseudo R2 = 0.105

Overall, our analysis suggests that there are more and greater differences between companies reporting compliance versus nonresponse than between noncompliant versus nonresponse. That is, there are four variables that distinguish nonresponse from both compliance and noncompliance. In addition, there are another three characteristics on which compliant firms differ only from nonresponding firms.

More specifically, in the first column, we see that employer size distinguishes nonresponding firms from those which comply as well as those that do not comply. Larger employers are more likely to either comply or not than to not respond to the questions. This result may reflect greater bureaucratization and standardization of HR policy as well as knowledgeable personnel in larger companies. Similarly, employers in a state with an FMLA-comparable policy prior to the FMLA are also significantly less likely to be nonresponding, compared with both compliers and noncompliers. Finally, employers who are also missing data on one of the basic health or economic benefits are also more likely to be missing data on family leave. All of these variables might indicate that for some respondents, missing data reflect a lack of knowledge rather than defiance, since these characteristics differentiate nonresponders from responders (whether compliant or not). There are additional characteristics, however, that distinguish nonresponders from compliers, suggesting that nonresponders may be more similar to noncompliers than to compliers. Employers with more than 50% women employees and more than 25% salaried employees are significantly more likely to comply than to not respond. And employers who provide more health and economic benefits are more likely to be FMLA compliant than are nonresponders.

We argue that this analysis of nonresponse versus compliance and noncompliance in a multinomial logistic model provides insight into the meaning underlying missing data. Because there are more and greater differences between compliers and nonresponders than between noncompliers and nonresponders, the model provides some evidence that nonrespondents are likely to be noncompliant with the FMLA. These results add further evidence that organizational compliance rates with the FMLA are lower than much prior research led us to believe. Although caution is certainly warranted, these findings also offer partial support for our hypothesis that nonresponse may indicate some level of resistance to legal mandates. Some key measures—especially constituent characteristics—predict both noncompliance and nonresponse.

Conclusion

Using more sophisticated analysis of missing values than prior research has utilized, we find that overall FMLA compliance rates are lower than previously thought. Prior research on coverage suggests that about half the workforce is covered by the law; our findings imply that, in practice, an even larger fraction of workers is unable to take advantage of the benefits afforded by the FMLA. Furthermore, we show compliance and nonresponse are both patterned: Although the law was intended to be gender neutral, organizations are more likely to comply with the mandated maternity leaves than with the other components of the act. Our bivariate analyses suggest that patterns of organizational compliance with the FMLA reinforce gender difference in the care of families. Given space limitations and the broad aims of this article, we did not include detailed analyses of these gendered patterns of compliance. But future research could profitably do so.

Overall, our analyses of these data suggest that we should reconsider standard social science techniques used to estimate compliance which we argue mistakenly excludes missing data. Our multivariate analysis raises the possibility that at least some of the survey nonresponses observed in the data indicate a kind of organizational defiance. While we suggest considerable caution about interpreting the reasons behind the missing data, the finding that nonresponsive firms resemble the noncompliant more than the compliant could be evidence that some organizations camouflage their noncompliance as nonresponse. It is important to separate empirically those for whom a lack of knowledge creates nonresponsiveness from those whose nonresponsiveness indicates defiance of the law. These two distinctive processes have the same effect—denying important policy resources to employees—but require different responses, with the first requiring better mechanisms to inform employers and employees about existing law, while the second requires greater enforcement mechanisms and penalties for illegal acts.

More generally, our analysis provides further evidence for the weakness of U.S. enforcement of regulatory legislation which as Edelman and Suchman (1997) argued over a decade and a half ago was already a “picture of non-compliance, subversion and evasion” (p. 487). The DOL’s Wage and Hour Division, which is the key body responsible for the enforcement of this act, has lost much of its ability to enforce federal labor laws in recent administrations (Matejkovic & Matejkovic, 2006). Dobbin (2009) suggests that while “government regulations got the ball rolling,” HR personnel were “the champions of women’s rights in newly feminized personnel departments” who “took the ball and ran with it” (p. 188). Our analysis suggests that many of these personnel do not run very far.

Much recent scholarship and activism is directed to revising current legislation by replacing unpaid with paid family leaves. However, the persistence of inequality in the utilization of leaves found in prior research and the low rates of compliance with existing law found in our work suggest that progress depends on revising, monitoring, and enforcing the current law. These initiatives can, and should, include efforts to make sure employers and employees are informed of their rights and responsibilities under the FMLA. Given the weakness of regulatory bodies, this is a difficult task. But passing another law without supporting existing law is inadequate as a route for building family-friendly organizations.

Although our findings go considerably beyond prior research, the data preclude analyses of other potentially fruitful comparisons and explanations. First, all employers included in the NSE (like those in DOL surveys) are private. Prior research suggests private-sector employers have lower levels of compliance than public-sector employers (Kelly, 2010). It would be useful if future research compared public-sector and private-sector employment and assessed whether they have different rates of compliance, noncompliance, and nonresponse as well as whether different or similar pressures explain these rates in both sectors. Second, more detailed data related to management may also help us theorize and empirically specify organizational characteristics associated with compliance. The presence of a formal HR function, which we cannot discern in these data, would seem to be an important factor in explaining compliance and nonresponse. Additionally, although we do include data on whether women are part of senior management, our measure may not allow adequate specification of women’s managerial presence. We do not have information on the presence of women in different levels of management; it is possible that the gender composition of lower and middle management has an effect on compliance with “family-friendly” policies, as a growing body of evidence suggests (Lambert, 2008; Swanberg & Simmons, 2008). Third, our available measures of current economic conditions may be inadequate. Kagan et al. (2003) found that lagged economic conditions had an effect on organizational compliance with environmental regulations, and similarly, we might expect FMLA compliance to be tied to economic pressures at the time the law passed or in the interim. Fourth, although we began to disentangle the role of a lack of knowledge of the law as opposed to outright rejection of it as a cause of noncompliance, we cannot fully disentangle these. Because these may require different policy responses—whether education or sanctions—it is important to be able to distinguish them.

Finally, as this research makes clear, policy—at the levels of the state and the organization—does not always equal practice. Our results rely on respondents’ reports about what their company policies allow; these are unlikely to perfectly correspond with what happens when employees need or even ask for leaves. For instance, firms that fail to publicize the law, or formally offer leaves but informally penalize those who request them, are counted here as “compliant.” Researchers have noted that men and women needing leave may experience pressure from supervisors and colleagues not to take leave, or to come back to work earlier than they feel appropriate (Albiston, 2005; Appelbaum, 2013; Fried, 1998; Gerstel & McGonagle, 1999; Williams, 2010). Consequently, even the figures we present here may involve an inflated estimate of the actual provision of family leaves by U.S. organizations.