Enforcement spillovers under different networks: The case of quotas for persons with disabilities in Brazil

Berlinski, Samuel G.; Gagete-Miranda, Jessica Working Paper Enforcement spillovers under different networks: The case of quotas for persons with disabilities in Brazil IDB Working Paper Series, No. IDB-WP-1613 Provided in Cooperation with: Inter-American Development Bank (IDB), Washington, DC Suggested Citation: Berlinski, Samuel G.; Gagete-Miranda, Jessica (2024) : Enforcement spillovers under different networks: The case of quotas for persons with disabilities in Brazil, IDB Working Paper Series, No. IDB-WP-1613, Inter-American Development Bank (IDB), Washington, DC, https://doi.org/10.18235/0013112 This Version is available at: https://hdl.handle.net/10419/302219 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/3.0/igo/ Enforcement Spillovers under Different Networks: The Case of Quotas for Persons with Disabilities in Brazil Samuel Berlinski Jessica Gagete-Miranda WORKING PAPER No IDB-WP-1613 InterA merican Development Bank Department of Research and Chief Economist August 2024 * InterA merican Development Bank and IZA ** University of Milano-Bicocca Enforcement Spillovers under Different Networks: The Case of Quotas for Persons with Disabilities in Brazil Samuel Berlinski* Jessica Gagete-Miranda** InterA merican Development Bank Department of Research and Chief Economist August 2024 Cataloging-in-Publication data provided by the Inter-American Development Bank Felipe Herrera Library Berlinski, Samuel, 1970Enforcement spillovers under different networks: the case of quotas for persons with disabilities in Brazil / Samuel Berlinski, Jessica Gagete-Miranda. p. cm. — (IDB Working Paper Series ; 1613) Includes bibliographical references. 1. People with disabilities-Brazil. 2. People with disabilities-Legal status, laws, etc.-Brazil. 3. People with disabilities-Employment-Brazil. I. Gagete-Miranda, Jessica. II. Inter-American Development Bank. Department of Research and Chief Economist. III. Title. IV. Series. IDB-WP-1613 http://www.iadb.org Copyright © 2024 Inter-American Development Bank ("IDB"). This work is subject to a Creative Commons license CC BY 3.0 IGO (https://creativecommons.org/licenses/by/3.0/igo/legalcode). The terms and conditions indicated in the URL link must be met and the respective recognition must be granted to the IDB. Further to section 8 of the above license, any mediation relating to disputes arising under such license shall be conducted in accordance with the WIPO Mediation Rules. Any dispute related to the use of the works of the IDB that cannot be settled amicably shall be submitted to arbitration pursuant to the United Nations Commission on International Trade Law (UNCITRAL) rules. The use of the IDB's name for any purpose other than for attribution, and the use of IDB's logo shall be subject to a separate written license agreement between the IDB and the user and is not authorized as part of this license. Note that the URL link includes terms and conditions that are an integral part of this license. The opinions expressed in this work are those of the authors and do not necessarily reflect the views of the Inter-American Development Bank, its Board of Directors, or the countries they represent. Abstract This study examines labor law enforcement spillovers in Brazil’s highly informal economy, focusing on disability quota enforcement for formal firms. New inspection procedures increased compliance through heightened inspections and fines, boosting disability hiring. We investigate spillover effects across various firm networks: neighborhood, ownership, and human resources specialists. Results show that spillovers can have up to twice the impact on disability employment compared to direct fines. These findings highlight the potential for targeted enforcement strategies to amplify policy effectiveness beyond directly affected firms even in developing economies characterized by low compliance with employment laws. JEL codes: I38, J68, K31 Keywords: Enforcement spillovers, Networks, Persons with disability, Brazil ∗We acknowledge the valuable comments of Matthew S. Johnson, Gabriel Ulyssea, and participants at the 2023 Workshop on Networks and Development, the NEUDC 2023 Conference, and seminars at CAF – Development Bank of Latin America and the Caribbean, the Inter-American Development Bank, University of Milano-Bicocca, University of Pretoria, and the University of Sao Paulo. The opinions expressed in this publication are those of the authors and do not necessarily reflect the views of the Inter-American Development Bank, its Board of Directors, or the countries they represent. 1 Introduction Compliance is a crucial issue policymakers face in virtually all legislative spheres, such as tax (Slemrod,2019), traffic (Lu et al.,2016), environmental (Shimshack,2014) and labor (Almeida and Ronconi,2016) regulations. Deterrence achieved through the inspection and punishment of non-compliers is costly and reaches very few individuals and firms in most environments. To improve general compliance, information about the inspection and punishment of a few must travel between individuals and firms (see, Slemrod,2019;Johnson, 2020). This paper investigates how information on employment fines is transmitted through different firm networks. In particular, we study the case of quotas for persons with disabilities in Brazil (henceforth Quota Law or QL) and explore i) the impact of new inspection and punishment procedures that increased the regulatory enforcement of the QL; and ii) how firms learn from different networks about the QL enforcement and their risk of being punished. The Brazilian Quota Law mandates that companies employing more than 100 workers allocate at least two percent of their employment positions to persons with disabilities. This legal requirement provides an ideal setting for our investigation for several reasons. First, even though the requirement was introduced in 1991, labor regulatory offices only started to effectively enforce it and punish non-compliers about 20 years later. Hence, for a considerable time, firms’ prior was that such a de jure law was not a de facto law, so they would not face the risk of punishment in case of non-compliance. Enforcement became tighter after the introduction of an administrative act in 2012, enabling us to study firms’ adjustment processes once a particular law starts to be enforced. Second, the Brazilian context allows us to combine different types of data to investigate how law enforcement spills over under different firms’ networks. Third, Brazil is a good example of the gap between de jure and de facto laws in developing countries, which reflects endemic problems for governments in enforcing compliance with laws (Acemoglu et al.,2015). We employ a difference in discontinuity design on matched employee-employer data and document that the 2012 Administrative Act led to an increase in inspections and fines issued due to non-compliance with the Quota Law and, consequently, to an increase in the hiring of persons with a disability after that year by about 7%. Furthermore, we show non-causal evidence of what might be interpreted as enforcement spillovers. The impact of the 2012 Administrative Act on issuing QL fines decreases as enforcement capacity decreases. However, the impact on hiring persons with disabilities is independent of the local enforcement capacity. Finally, we present the key contribution of the paper where we test for the presence of 1 enforcement spillovers by employing a stacked differences-in-differences design to understand how the occurrence of a QL fine in a firm’s network impacts the likelihood that such a firm will increase its hiring of workers with disabilities. We look at three different networks: neighbor network (i.e., firms located in the same zip code of a specific firm), owner network (i.e., firms that belong to the same owner of a specific firm or an associate of such an owner) and HR workers network (i.e., firms where the HR workers of a specific firm were working before joining such a firm). We compare firms within networks where a Quota Law fine was issued to firms within networks where a Quota Law fine will be issued in the future but have not received one yet. We find strong evidence of enforcement spillovers when a QL fine happens in the neighbor, owner, or HR workers’ networks. If another firm in the network of a firm ireceives a QL fine, this increases the number of workers with a disability present in firm iin the following years by 7.4% in the neighbor network, 7% in the owner network, and 4.6% in the HR workers’ network. A back-of-the-envelope calculation indicates that the total number of workers with disabilities hired due to spillover effects in the neighbor and HR workers’ networks is about twice as large as the direct impact of receiving a QL fine. This figure decreases considerably for the owner networks (only 30% as large as the direct impact) due to the small size of such networks. Spillovers are stronger for firms that were not complying with the Quota Law when the fine was issued. We show that the likelihood of being inspected does not increase after the occurrence of a QL fine in any of the firm’s networks, which suggests that the spread of information and not the local increase in enforcement is causing the emergence of spillovers. This paper contributes to two different streams of the literature. First, it contributes to the literature on regulations to improve the employment opportunities of people with a disability. Title I of the Americans with Disabilities Act (ADA) of 1990 ensures that private employers, state and local governments, employment agencies, and labor unions cannot discriminate against qualified individuals with disabilities during job application procedures, hiring, firing, advancement, compensation, job training, and other aspects of employment. Evidence on its impact is mixed, with some papers showing a negative effect on labor market participation of persons with disabilities (Acemoglu and Angrist,2001;DeLeire,2000a,b), while other studies dispute such findings (Hotchkiss,2004;Jolls and Prescott,2004).1Outside of the United States, quota systems like the one analyzed in this study have been adopted by over two-thirds of OECD countries (OECD,2003). In Austria (Lalive et al.,2013), Hungary (Krekó and Telegdy,2022), and Japan (Mori and Sakamoto,2018) , research has found 1For a review on the impacts of the ADA and other policies targeting the inclusion of persons with disabilities in the United States, see Livermore and Goodman (2009). 2 that firms comply with such regulations. Papers that study the quota system in developing countries, where regulatory compliance is arguably worse, usually leverage the role of law enforcement in improving compliance with the quota to calculate its welfare effects, like Szerman (2022) and de Souza (2023), who also study the Brazilian quota system. In this paper, we provide evidence that enforcing the Quota Law generates spillover effects, which should be taken into consideration in any welfare calculation. Second, this paper contributes to the literature investigating the impacts of law enforcement and, more specifically, the emergence of enforcement spillovers among firms2. Inspections, auditing, and fines have been proven effective in increasing regulatory compliance (see Gray and Shimshack,2011;Levine et al.,2012, for examples on environmental and occupational health and safety regulators). However, developing countries struggle with low enforcement capacity, challenging compliance efforts in many sectors and localities (Almeida and Ronconi,2016;Ponczek and Ulyssea,2022). Hence, such countries could significantly benefit from enforcement spillovers since the emergence of such indirect impact would have a multiplier effect on each atomistic enforcement effort. The evidence on enforcement spillovers so far comes from developed countries. In the United States, for instance, Shimshack and Ward (2005) and Evans et al. (2018) provide evidence that enforcing environmental regulations increases future compliance of other firms located in the same state where enforcement occurs but may create negative externalities in areas that are not inspected. Also in the United States, Johnson (2020) shows that publicizing firms’ health violations led to other firms to comply more with such a regulation. However, the emergence of such spillovers in developing countries is far from obvious due to the weaker enforcement capacity and the larger gap between de jure and de facto laws. Besides being the first paper to present evidence of enforcement spillovers in developing countries, this paper also contributes to this literature by showing how information about law enforcement flows through different networks connected to the firm, which can help inform policymakers on how to improve targeting to leverage the emergence of spillovers. For example, the results of a set of inspections could be disseminated through information letters to companies in a firm’s network. 2The literature has shown that there are enforcement spillovers among individuals’ networks (family, coworkers, and neighbors) in developed countries for dividend and capital taxation, commuter tax allowances, and TV license payments (e.g., Alstadsæter et al.,2019;Drago et al.,2020;Paetzold and Winner,2016; Rincke and Traxler,2011). 3 2 Background The Brazilian Quota Law is an important example of the broader emphasis that both developing and developed countries are placing on diversity and inclusion policies.3This is an important policy goal for ethical, efficiency, and redistributive reasons. Berlinski et al. (2021) estimate that in Latin America and the Caribbean (LAC) countries, 88 million people were living with a disability in 2020 (around 15 percent of the population). By 2050, this figure could rise by 60 million. People living with disabilities have lower educational attendance and school completion rates and large gaps in labor market outcomes with respect to those living without disabilities. For example, Berlinski et al. (2021) report that the employment disability gap for people aged 25–34 in the eight national censuses for LAC countries they analyze is, on average, 18.5 percentage points. Governments often implement employment quotas to promote the hiring of persons with disabilities. (See, Mont et al.,2004;Förster,2007). These quotas are commonly used among OECD and partner countries (Förster,2007;OECD,2003). The specific regulations regarding which firms are subject to quotas and the percentage of vacancies they should reserve for workers with disabilities vary across countries. Major corporations are the primary focus of such measures, with the typical proportion of jobs set aside for individuals with disabilities hovering around 4%. In Brazil, the Quota Law4is the most relevant legislation regarding the employment opportunities of persons with disability. It establishes that, respectively, firms with more than 100, 200, 500, and 1,000 employees must fill at least 2%, 3%, 4%, and 5% of their payroll with people with a certified disability. However, compliance with the Quota Law has historically remained limited since its introduction in 1991. For instance, in 2009, less than 30% of firms with more than 100 workers were employing the minimum number of workers with disabilities established by the law. Compliance has been an issue in other countries as well (see, OECD,2003) despite some evidence of the effectiveness of quotas at increasing the employment of people with 3For example, the United Nations’ Sustainable Development Goals for the year 2030 aspire to: SDG 4: “Ensure inclusive and equitable quality education and promote lifelong learning opportunities for all”; SDG 5: “Achieve gender equality and empower all women and girls”; SDG 8: “Promote sustained, inclusive and sustainable economic growth, full and productive employment, and decent work for all”; SDG 10: “Reduce inequality within and among countries.”; SDG 11: Make cities and human settlements inclusive, safe, resilient, and sustainable; and SDG 16, “Peace, justice and strong institutions,” which promotes building effective, accountable, and inclusive institutions at all levels to ensure peaceful and inclusive societies for all. Additionally, this SFD is consistent with the social model of disability embedded in the 2008 United Nations Convention on the Rights of Persons with Disabilities, and it is aligned with ILO Convention 169 and articles 3 and 4 of the UN Declaration on the Rights of Indigenous Peoples, which recognize their right to make autonomous decisions regarding their development priorities. 4Art. 93. of Law 8.213/1991. 4 duced by Grembi et al. (2016), who combined the traditional RD design with ideas from the difference-in-differences design. The authors added a second dimension to the regression discontinuity (RD) design, where a structural change happened between two different periods while a discontinuity holds during both periods. In our case, such a change is the introduction of the new inspection procedures in 2012, while the thresholds defined by the Quota Law remained the same during the whole period. The main idea of the differencesin-discontinuities design is that it takes the difference between the pre-2012 and post-2012 discontinuities in the Quota Law threshold to separate the effect of the Quota Law from the effect of enforcing such a policy through the new inspection procedures. We identify the effect of introducing the new inspection procedures through the following equation (estimated within the bandwidth proposed by Calonico et al. (2014a,b)): Yimt =β0+β1Pimt +β2Simt(γ0+γ1Pimt) + T[α0+α1Pimt+ Simt(δ0+δ1Pimt)] + θm+θt+θmt +εimt.(1) Where, Yimt is the outcome of firm ilocated at municipality mat year t,Simt = 1(Firm Size ≥ 100), and Pimt =Firm Size −100. Moreover, T= 1(t≥2012). Our coefficient of interest in equation 1is δ0. It identifies the impact of surpassing the cut-off threshold after 2012. The estimation includes controls for municipal and year fixed effects (θmand θt, respectively) and interactions between these two (θmt). The inclusion of municipality-by-year fixed effects in our estimations addresses two identification threats. First, it ensures that the supply of persons with disability is held constant across municipalities and time. This alleviates concerns that the new inspection procedures might have changed not only firms’ demand for workers with disability but also the willingness of persons with disability to look for jobs. Second, it controls for time-specific stringency of labor inspections – for instance, new openings of labor inspection offices. Three assumptions need to hold for identification. First, as in the traditional RD design, all potential outcomes should be continuous around the discontinuity each year. Second, similar to the difference-in-differences design, the observations just below and above the discontinuity must follow (local) parallel trends in the counterfactual scenario of no new inspection procedures. Third, the effects of the new inspection procedures should be independent of the thresholds of the Quota Law. A further problem in our analysis is measurement error in firms’ size since our variables are aggregated by year. To address this issue, we implement a donut ring strategy (Barreca et al., 2011), where we exclude firms with sizes within one or two workers from the 100 thresholds. Through such exclusion, we avoid inclusion and exclusion errors where we wrongly consider a 11 firm larger or smaller than the QL threshold. The donut strategy also helps in addressing the issue of potentially endogenous manipulation of the running variable around the threshold, even though Figure A1 in the Appendix Ashows that manipulation is not a concern in our analysis. Table 2presents the impact of the 2012 Administrative Act on law enforcement (the likelihood of receiving an inspection in columns (1) and (2) and of being fined due to noncompliance with the Quota Law in columns (3) and (4)15), and on the number of workers with disabilities present in the firm (columns (5) and (6)). Columns (1), (3), and (5) present estimations considering a donut ring of one, while columns (2), (4), and (6) present robustness checks implementing a donut ring of two in the firm-size variable. The introduction of the 2012 Administrative Act increases the likelihood that firms would be inspected by 2.4 percentage points or a 5.3% increase. The increased likelihood of receiving a Quota Law fine, shown in column (3), is much more striking: after the 2012 Administrative Act, such a likelihood increased by 1.2 percentage points or 44.4%. Finally, the result in column (5) shows that firms reacted to the increase in QL enforcement after 2012 by hiring more persons with disabilities. More specifically, we can see that firms larger than 100 workers increased their number of workers with a disability by 6.9%.16 The results are quite similar regardless of the donut ring implemented. Figure 3presents the dynamic results of such estimations, where we substitute the binary variable Tin equation 1for yearly dummies. The figure shows no difference between firms larger and smaller than 100 workers before 2012. However, we observe an increase in inspections, Quota Law fines, and the number of workers with disabilities for firms larger than 100 workers after that year. Recent work by Chen and Roth (2024) shows that one should be careful when interpreting results with log or inverse hyperbolic sine transformations, especially if the treatment affects the extensive margins – in our case, the likelihood of a firm passing from having no workers with a disability to having one or more. We present in Table A2 and Figure A2 two robustness checks to deal with this issue. First, we present extensive margin estimations, calculating the likelihood that firms have at least one employee with a disability. Second, we estimate a linear regression model with the number of workers with disability as the dependent variable. Results are the same for the extensive margin estimation and point in the same direction as 15The likelihood of receiving a QL fine is zero for firms smaller than 100 workers since they do not need to comply with the law. Hence, the results in columns (3) and (4) show the difference across time in the likelihood that firms larger than 100 will receive such a fine. 16As explained in section 3, we use the hyperbolic sine transformation of the number of workers with a disability. We include in Table 2the elasticity of the number of workers with disabilities, using the calculation derived by Bellemare and Wichman (2020). 12 the main estimation in the regression model with the count variable, even though we do not have enough power to reject the hypothesis of null effects in the last case. Table 2: The 2012 Administrative Act, Law Enforcement, and Workers with Disabilities (1) (2) (3) (4) (5) (6) Inspection QL fine Workers w/ a disability (hyp. sine trans.) Year>2012 X Dist. QL threshold>0 0.020∗∗∗ 0.019∗∗∗ 0.013∗∗∗ 0.013∗∗∗ 0.062∗∗∗ 0.068∗∗∗ (0.006) (0.007) (0.003) (0.002) (0.016) (0.018) N 446497 437528 331122 322136 222233 213288 Mean Dep. Var. 0.440 0.441 0.024 0.024 0.841 0.855 Elasticity 0.064 0.070 h (left) 33.305 33.305 19.665 19.665 13.910 13.910 h (right) 118.828 118.828 127.249 127.249 66.489 66.489 R2 0.176 0.177 0.106 0.107 0.208 0.210 Donut ring 1 2 1 2 1 2 Note: This table presents estimations from Equation 1. Elasticities of workers with disabilities are calculated based on Bellemare and Wichman (2020). All estimations include city-by-year fixed effects. Significance levels are indicated by ∗< .1, ** < .05, *** < .01. Standard errors clustered at the city level shown in parentheses. We also investigate whether introducing the new inspection mechanisms led to changes in fines unrelated to the QL (non-QL fines) or to outcomes potentially related to firms’ profitability. Table A3 and Figure A3 in the Appendix show the result of that analysis. We test for the effects of the new inspection mechanisms on non-QL fines as a placebo exercise since the new administrative act should not interfere with the enforcement of other labor regulations. Indeed, we do not observe any impact on the likelihood of receiving fines not related to the Quota Law around the thresholds, which supports our hypothesis that the observed impact for inspections and QL fines is due to the introduction of the 2012 Administrative Act and not to an overall increase in enforcement of labor regulations after 2012. Furthermore, we analyze the impact of higher enforcement around the threshold on firm closure (we proxied for closure by looking at whether a firm present in the data at time tis not found at t+ 1), firm total wage bill, and turnover rate.17 We find no impact on these outcomes suggesting that the enforcement did not lead to major profitability issues. These results are in line with evidence on the enforcement of labor regulation in the United States (Levine et al.,2012) and in the Brazilian context (Szerman,2022), which also does not find that hiring people with disabilities had negative impacts on firms or workers without disabilities.18 17We use separation rate, defined as the number of workers that left the firm at time tdivided by all spells present in the firm at tas our measure of turnover rate.(Pries and Rogerson,2022) 18de Souza (2023), in contrast, finds negative effects from the Quota Law enforcement for the employment and wages of workers not carrying a disability in Brazilian firms. A possible reason for such a difference is that we exploit an institutional change, while de Souza (2023) exploits the timing of firm inspections. Receiving an inspection, however, might lead firms to rush into hiring persons with disability, which might decrease their productivity, at least momentarily. 13 Figure 3: The 2012 Administrative Act, Law Enforcement, and Workers with Disabilities -.05 0 .05 .1 Diff. in Likelihood: Large vs. Small Firms 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 (a) Inspection -.01 0 .01 .02 .03 .04 Diff. in Likelihood: Large vs. Small Firms 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 (b) QL Fines -.1 0 .1 .2 .3 % Growth: Large vs. Small Firms 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 (c) Workers with Disability Note: These graphs present estimations from a model similar to Equation 1, where we substitute the post-2012 dummy T with year dummies, leaving 2011 as the benchmark. We consider a donut ring of two in the firm-size variable. All estimations include city-by-year fixed effects. 90% confidence interval shown in the graphs. 14 A final question we ask is whether enforcing the Quota Law led firms to compete for workers with disability. Hence, we investigate whether the wages of persons with disabilities increased after the 2012 Administrative Act. We present this result in the last column of Table A3 and the last sub-figure of Figure A3 in the Appendix. As we can see, there is no evidence of such a competition for workers with disability. Overall, these results show that the only remarkable difference after 2012 in the prevalence of law enforcement between firms larger and smaller than 100 was the increase in inspections and primarily the increase in issuing of Quota Law fines. The results also show that firms reacted to such an increase in QL enforcement by hiring more workers with disabilities, and this does not seem to impact other firms’ economic outcomes or their competition for the employment of persons with disability. 4.1 Compliance in the Absence of Direct Enforcement Enforcement spillovers are a powerful tool in achieving general compliance due to the spread of information about law enforcement to agents not directly impacted by it. We next present evidence that firms increased their compliance with the QL after the 2012 Administrative Act, even if located in places with lower enforcement capacity or in cases where they never received a QL fine themselves. In the next section, we directly investigate the emergence of enforcement spillovers. Previous research on labor regulations highlights the significant impact of enforcement capacity on compliance (e.g., Almeida and Carneiro,2012;Ponczek and Ulyssea,2022). However, our analysis reveals that while the issuance of QL fines decreases with reduced enforcement capacity, this does not hold for the hiring of workers with disabilities. In particular, we estimate a model that includes interactions between the variable Absence of LO and all elements of equation 1. We define Absence of LO as a binary variable indicating that there is no labor office in the municipality where firm iis located (i.e., the distance between municipality mwhere firm iis located and the nearest labor office is greater than zero). The results are presented in Table 3. The impact of the 2012 Administrative Act on inspections does not change depending on the firm’s distance to the nearest labor office (see columns (1) and (2)). However, the issuance of fines decreases considerably in places with no labor office (see columns (3) and (4)). Even with such a decrease in the issuing of fines in the absence of labor offices, the impact of the 2012 Administrative Act on the hiring of persons with a disability is constant across localities, regardless of whether labor offices are present in their municipality or not (see columns (5) and (6)). 15 Table 3: Heterogeneous Results by Enforcement Capacity Level (1) (2) (3) (4) (5) (6) Inspection QL fine Workers w/ a disability (hyp. sine trans.) Year>2012 X Dist. QL threshold>0 0.014∗0.012 0.016∗∗∗ 0.015∗∗∗ 0.062∗∗∗ 0.063∗∗∗ (0.008) (0.009) (0.003) (0.003) (0.017) (0.020) Year>2012 X Dist. QL threshold>0 X Absence of LO 0.014 0.017 -0.007∗∗ -0.006∗0.001 0.012 (0.013) (0.013) (0.004) (0.004) (0.029) (0.032) N 432081 423407 320222 311529 214932 206282 Mean Dep. Var. 0.437 0.438 0.024 0.024 0.465 0.472 h (left) 33.305 33.305 19.665 19.665 13.910 13.910 h (right) 118.828 118.828 127.249 127.249 66.489 66.489 R2 0.175 0.176 0.107 0.108 0.209 0.211 Donut ring 1 2 1 2 1 2 Note: This table presents estimations from equation 1. All estimations include city-by-year fixed effects. Significance levels are indicated by ∗< .1, ** < .05, *** < .01. Standard errors clustered at the city level shown in parentheses. Figure 4: The 2012 Administrative Act and Workers with Disabilities Sub-sample: Firms that Never Received a QL Fine -.1 0 .1 .2 % Growth: Large vs. Small Firms 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 Note: This figure presents estimation from a model similar to equation 1, where we substitute the post-2012 dummy Twith year dummies, leaving 2011 as the benchmark. We keep in our estimation the sub-sample of firms that never received a QL fine. The estimation includes city-by-year fixed effects. 90% confidence interval shown in the graphs. Moreover, even in places where enforcement capacity is higher, the number of QL fines issued is relatively small: after 2012, only 8.4% of firms larger than 100 workers located in municipalities with the presence of labor regulatory offices (i.e., closer to the regulation enforcers) were fined due to non-compliance with the QL, even though compliance was only near 20% for these firms during that period. However, even firms that never received a QL fine reacted to the 2012 Administrative Act by increasing their hiring of people with disabilities. We show this in Figure 4, where we reproduce the estimations shown in Figure 3restricting our sample to firms that never received a QL fine. As we can see, the impact of the 2012 Administrative Act in this sub-sample is remarkably similar to the one considering the whole sample of firms. 16 5 Enforcement Spillovers Having established that introducing new enforcement mechanisms effectively changed firms’ behavior regarding hiring persons with a disability, we now focus on the spillover effects stemming from such an increase in the stringency of law enforcement. If firms learn from their networks about the increase in the likelihood of being punished due to non-compliance with the Quota Law, they might change their behavior, even if not directly exposed to law enforcement. We explore the timing of Quota Law fines and investigate how a firm ireacts when another firm in its network receives such a fine. We look at three different networks from which firm icould learn about the Quota Law enforcement: i) neighbor network, defined as the firms located in the same zip code as firm i;ii) owner network , defined as the firms that belong to the same owner of firm ior firms that belong to a business associate of such an owner;19 and iii) HR workers network, defined as the firms where human resources workers working for firm iat time twere working up to three years before t. We implement an event-study methodology where we analyze trends in the presence of workers with a disability before and after the occurrence of a QL fine in a firm’s network. We use as control group firms that belong to networks that will receive a QL fine in the future (Deshpande and Li,2019;Fadlon and Nielsen,2020). The main identification assumption behind the choice of such a control group is that, while receiving a QL fine might be endogenous to a network, the timing of such a fine can be considered exogenous. Table A4 in the Appendix shows that the occurrence of a fine in firms’ networks already seems quite exogenous, even if we consider networks that have never received a fine. Overall, firms’ previous characteristics, such as their size or their number of workers with a disability, are not able to predict the occurrence of a QL fine in their network. However, the exogeneity is even more evident when we restrict the comparison group to firms whose networks received a QL fine in the future: the little predictive power that we observe in the even columns of Table A4 (i.e., the estimation that included never-treated networks) usually vanishes in the subsequent estimations (odd-columns) when we exclude from the estimation firms belonging to networks that have never received a QL fine. We construct our estimation sample of firms in four steps. First, we take the networks where the QL fine happened at any time after 2012.20 For instance, to look at the impact of a QL fine in the neighbor network, this means restricting the sample to zip codes where any QL fine happened between 2012 and 2018. The same idea applies to the owner network 19We define business associates as individuals who share the ownership of a firm. 20We focus on the period after implementing the new inspection procedures since the number of fines increased considerably after it. 17 and the HR workers network. Over this period, some networks are treated earlier than others. Second, at every year, we label time 0 the first time a network receives the QL fine. At that point, the network is considered treated. Third, at every year, any network that receives a QL fine for the first time at least two years in the future is considered a control network. Fourth, we stack for every year a set of treated and control networks repeating this procedure. Every treated and control network has three years of data before and two after the event. To ensure we investigate spillovers from law enforcement and not its direct effect, we drop from our sample the establishments that received a QL fine at time 0. We estimate the following model using ordinary least squares regressions: Yimct =δ0Treatedimqt + τ=2 X τ6=−1 τ=−3 Dτ t+ τ=2 X τ6=−1 τ=−3 δτ(Treatedimqt ×Dτ t) +θm+θt+θc+θmct +imqt (2) where Yimct is the outcome of interest (for instance, number of workers with disabilities) for firm i, in municipality m, in the year of treatment (or cohort) c, at event-time t, the Dτ tare indicators equal to one for each event-time window (i.e., τ= -3, ..., 0, ...., 2), and imct is a random specification error. As in equation 1, we include controls for time and municipality fixed effects and the interaction between these two terms (θtand θm, and θmt respectively) to control for fixed characteristics of the local labor market which could be correlated with the likelihood of observing a fine, and the presence of labor regulatory offices. Besides, we also include cohort fixed effects θcto control for other shocks happening simultaneously to the QL fine. The coefficients of interest are the estimates of δτ. At every event-time window, they represent the causal impact on the employment of workers with disabilities for firms belonging to a network where another firm received a QL fine. We also present a summary estimate with a post-dummy instead of each event-time dummy. We estimate equation 2first for the sample of all firms larger than 100 workers and second separately for the sub-sample of firms that were at least partially complying with the QL before the QL fine hit their network and the sub-sample of firms that were not complying with the QL at that time.21 We hypothesize that firms already complying with the law should not be affected by the information about enforcing such a law in their networks, while non-compliant firms should be the most impacted by the new information. 21Since fully complying with the QL is a rare event, we consider that a firm partially complies with the law if it has at least 50% of the number of workers with a disability that it should have according to the quota. 18 First, firms react to a QL fine in their network by hiring more workers with disabilities, regardless of the network where such a fine happens. We see such a pattern in Figure 5that presents the results for estimations considering all firms larger than 100 workers for the event of a QL fine in the firm’s neighbor network (Figure 5a), owner network (Figure 5b), and HR workers network (Figure 5c). In all cases, we observe an increase in the number of workers with a disability in the firm after the event of a QL fine in their network. That increase is persistent, and it grows with time in the neighbor and the owner networks while it fades out in the HR workers’ network. Second, only firms that risk receiving a QL fine – that is, those not complying with the law – react to the occurrence of a fine in their networks. Figure 6presents such a result. Overall, we see that the positive effect of a QL fine in the firm’s network on its number of workers with a disability is concentrated in firms that were not complying with the QL before that event (figures 6a,6c, and 6e). The result is imprecisely estimated for the HR workers’ network (Figure 6e), such that they are not significantly different from zero, even though we observe an increase in the coefficients’ size. In turn, firms that were already at least partially complying with the QL do not change their behavior at all when exposed to the event of a QL fine (Figures 6b,6d, and 6f).22 Third, one could hypothesize that, after fining a firm in a particular network, labor inspectors become more likely to inspect other firms in that same network, and this is what drives the increase in the number of workers with disabilities in firms instead of their direct communication with other firms in their network. However, this does not seem to be the case: we can see in Figure 7, which presents results of estimations where we investigate whether the occurrence of a Quota Law fine impacts the likelihood of firms being inspected. We estimate models similar to the one in equation 2, where now our dependent variable is a binary variable indicating whether the firm was inspected. As shown in the figure, having a QL fine in their network does not increase the likelihood that firms will receive an inspection after such an event. We summarize all these results in Table 4, where we substitute the event-time dummies with a binary variable indicating the post-event periods. Overall, these results show that 22We employ the methodology proposed by Rambachan and Roth (2023) to assess the sensitivity to parallel trends violations in Figure A4 (Appendix A). Specifically, we allow for differences in linear trends between treated and not-yet-treated and quantify how large any departures from such linearity should be so that we would have null results. We consider a range of values M, where M= 0 means no difference in linear trends and M > 0allows for deviations in linearity. If we consider all firms, we do not nullify the results for the neighbor network even for values of Mas large as M= 1. The breakdown value of Min the owner network is 0.3, and results are already imprecise in the worker network, even for M= 0. If we consider non-complier firms, we do not nullify the results for the neighbor and owner network even for values of Mand large as M= 1, and the breakdown value of Mis 0.8 in the worker network. 19 Figure 5: Law Enforcement at Firm Networks and the Number of Workers with a Disability -.1 -.05 0 .05 .1 .15 Coefficient -3 -2 -1 0 1 2 Event Time (a) Neighbor Network -.05 0 .05 .1 .15 .2 Coefficient -3 -2 -1 0 1 2 Event Time (b) Owner Network -.2 -.1 0 .1 .2 Coefficient -3 -2 -1 0 1 2 Event Time (c) HR Workers Network Note: These graphs present estimations from equation 2. The dependent variable is the hyperbolic sine transformation of the number of workers with a disability in the firm. The sample comprises firms larger than 100 workers that did not receive a QL fine in t= 0. "Event Time" is the time after the occurrence of the QL fine in the firm’s network. All estimations include cohort and city-by-year fixed effects. 90% confidence interval shown in the graphs. 20 References Acemoglu, D. and J. D. Angrist (2001). Consequences of employment protection? the case of the americans with disabilities act. Journal of Political Economy 109(5), 915–957. Acemoglu, D., C. García-Jimeno, and J. A. Robinson (2015). State capacity and economic development: A network approach. American economic review 105(8), 2364–2409. Almeida, R. and P. Carneiro (2012). Enforcement of labor regulation and informality. American Economic Journal: Applied Economics 4(3), 64–89. Almeida, R. and L. Ronconi (2016). Labor inspections in the developing world: Stylized facts from the enterprise survey. Industrial Relations: A Journal of Economy and Society 55(3), 468–489. Alstadsæter, A., W. Kopczuk, and K. Telle (2019). Social networks and tax avoidance: Evidence from a well-defined norwegian tax shelter. International Tax and Public Finance 26, 1291–1328. Barreca, A. I., M. Guldi, J. M. Lindo, and G. R. Waddell (2011). Saving babies? revisiting the effect of very low birth weight classification. The Quarterly Journal of Economics 126(4), 2117–2123. Bellemare, M. F. and C. J. Wichman (2020). Elasticities and the inverse hyperbolic sine transformation. Oxford Bulletin of Economics and Statistics 82(1), 50–61. Berlinski, S., S. Duryea, and S. M. Perez-Vincent (2021). Prevalence and correlates of disability in latin america and the caribbean: Evidence from 8 national censuses. PloS one 16(10), e0258825. Callaway, B. and P. H. Sant’Anna (2021). Difference-in-differences with multiple time periods. Journal of econometrics 225(2), 200–230. Calonico, S., M. D. Cattaneo, and R. Titiunik (2014a). Robust data-driven inference in the regression-discontinuity design. The Stata Journal 14(4), 909–946. Calonico, S., M. D. Cattaneo, and R. Titiunik (2014b). Robust nonparametric confidence intervals for regression-discontinuity designs. Econometrica 82(6), 2295–2326. Cattaneo, M. D., N. Idrobo, and R. Titiunik (2019). A practical introduction to regression discontinuity designs: Foundations. Cambridge University Press. Cattaneo, M. D., M. Jansson, and X. Ma (2018). Manipulation testing based on density discontinuity. The Stata Journal 18(1), 234–261. Chen, J. and J. Roth (2024). Logs with zeros? some problems and solutions. The Quarterly Journal of Economics 139(2), 891–936. de Souza, G. (2023). Employment and welfare effects of the quota for disabled workers in brazil. 27 DeLeire, T. (2000a). The unintended consequences of the americans with disabilities act. Regulation 23, 21. DeLeire, T. (2000b). The wage and employment effects of the americans with disabilities act. Journal of human resources, 693–715. Deshpande, M. and Y. Li (2019). Who is screened out? application costs and the targeting of disability programs. American Economic Journal: Economic Policy 11(4), 213–48. Drago, F., F. Mengel, and C. Traxler (2020). Compliance behavior in networks: Evidence from a field experiment. American Economic Journal: Applied Economics 12(2), 96–133. Evans, M. F., S. M. Gilpatric, and J. P. Shimshack (2018). Enforcement spillovers: Lessons from strategic interactions in regulation and product markets. The Journal of Law and Economics 61(4), 739–769. Fadlon, I. and T. H. Nielsen (2020). Family labor supply responses to severe health shocks: Evidence from danish administrative records. American Economic Journal: Applied Economics. Förster, M. (2007). From disability to ability. policy challenges and trends in oecd countries. Gray, W. B. and J. P. Shimshack (2011). The effectiveness of environmental monitoring and enforcement: A review of the empirical evidence. Review of Environmental Economics and Policy 5(1), 3–24. Grembi, V., T. Nannicini, and U. Troiano (2016). Do fiscal rules matter? American Economic Journal: Applied Economics, 1–30. Hotchkiss, J. L. (2004). A closer look at the employment impact of the americans with disabilities act. Journal of Human Resources 39(4), 887–911. Johnson, M. S. (2020). Regulation by shaming: Deterrence effects of publicizing violations of workplace safety and health laws. American economic review 110(6), 1866–1904. Jolls, C. and J. J. Prescott (2004). Disaggregating employment protection: The case of disability discrimination. Technical report, National Bureau of Economic Research. Krekó, J. and Á. Telegdy (2022). The effects of a disability employment quota when compliance is cheaper than defiance. Lalive, R., J.-P. Wuellrich, and J. Zweimüller (2013). Do financial incentives affect firms’ demand for disabled workers? Journal of the European Economic Association 11(1), 25–58. Levine, D. I., M. W. Toffel, and M. S. Johnson (2012). Randomized government safety inspections reduce worker injuries with no detectable job loss. Science 336(6083), 907– 911. 28 Livermore, G. A. and N. Goodman (2009). A review of recent evaluation efforts associated with programs and policies designed to promote the employment of adults with disabilities. report. Rehabilitation Research and Training Center on Employment Policy for Persons with Disabilities. Lu, F., J. Zhang, and J. M. Perloff (2016). General and specific information in deterring traffic violations: evidence from a randomized experiment. Journal of Economic Behavior & Organization 123, 97–107. Mont, D. et al. (2004). Disability employment policy. Social Protection, Labor Markets, Pensions, Social Assistance, World Bank. Mori, Y. and N. Sakamoto (2018). Economic consequences of employment quota system for disabled people: Evidence from a regression discontinuity design in japan. Journal of the Japanese and International Economies 48, 1–14. OECD (2003). Transforming disability into ability: policies to promote work and income security for disabled people. OECD Paris. Paetzold, J. and H. Winner (2016). Taking the high road? compliance with commuter tax allowances and the role of evasion spillovers. Journal of Public Economics 143, 1–14. Ponczek, V. and G. Ulyssea (2022). Enforcement of labour regulation and the labour market effects of trade: Evidence from brazil. The Economic Journal 132(641), 361–390. Pries, M. J. and R. Rogerson (2022). Declining worker turnover: The role of short-duration employment spells. American Economic Journal: Macroeconomics 14(1), 260–300. Rambachan, A. and J. Roth (2023). A more credible approach to parallel trends. Review of Economic Studies 90(5), 2555–2591. Rincke, J. and C. Traxler (2011). Enforcement spillovers. Review of Economics and Statistics 93(4), 1224–1234. Shimshack, J. P. (2014). The economics of environmental monitoring and enforcement. Annu. Rev. Resour. Econ. 6(1), 339–360. Shimshack, J. P. and M. B. Ward (2005). Regulator reputation, enforcement, and environmental compliance. Journal of Environmental Economics and Management 50(3), 519–540. Slemrod, J. (2019). Tax compliance and enforcement. Journal of Economic Literature 57(4), 904–954. Szerman, C. (2022). The labor market effects of disability hiring quotas. Available at SSRN 4267622. 29 A Appendix Table A1: RDD Quota Law Threshold: Different Bandwidth Selection Procedures Dep. var.: workers w/ a disability (hyp. sine trans.) (1) (2) (3) (4) (5) Panel A: Before 2012 RD Estimate 0.016 0.041*** 0.016 0.016 0.016 (0.014) (0.010) (0.012) (0.014) (0.013) N 991510 991510 991510 991510 991510 Mean dep. var. within bandwidth 0.198 0.314 0.195 0.198 0.206 h (left) 16.830 17.820 20.725 16.830 17.820 h (right) 16.830 69.080 20.725 16.830 20.725 Bandwidth selection procedure mserd msetwo msesum msecomb1 msecomb2 Panel B: After 2012 RD Estimate 0.057** 0.080*** 0.059*** 0.057** 0.055*** (0.025) (0.017) (0.018) (0.025) (0.020) N 1593962 1593962 1593962 1593962 1593962 Mean dep. var. within bandwidth 0.343 0.548 0.333 0.343 0.371 h (left) 9.844 10.476 14.293 9.844 10.476 h (right) 9.844 52.741 14.293 9.844 14.293 Bandwidth selection procedure mserd msetwo msesum msecomb1 msecomb2 Note: This table shows results from local polynomial regressions where we estimate firms’ hiring behavior regarding workers with disabilities once they pass the 100 workers threshold established by the Quota Law (see Cattaneo et al., 2019;Calonico et al.,2014a,b, for details on our RDD estimation). The dependent variable is the hyperbolic sine transformation of the number of workers carrying a disability. Panel A shows estimations for the years before 2012, i.e., before the introduction of the new inspection procedures. Panel B shows estimations for the years after 2012, i.e., after the introduction of the new inspection procedures. Each column of the table shows the results of estimations using different bandwidths optimally computed following the algorithm developed by Calonico et al. (2014a,b). Due to measurement errors in the estimation of the firm’s size, we exclude firms within a donut ring of size two from the 100 threshold. Significance levels are indicated by ∗< .1, ** < .05, *** < .01. Standard errors clustered at the city level shown in parentheses. Figure A1: Manipulation test 0 .01 .02 .03 -100 -50 0 50 100 P_100 Manipulation Testing Plot (a) Before 2012: T stat= -0.4719 ; P>|T|=0.6370 0 .01 .02 .03 -100 -50 0 50 100 P_100 Manipulation Testing Plot (b) After 2012: T stat=-2.7434 ; P>|T|=0.0061 See Cattaneo et al. (2018) for details about the implementation of manipulation tests. 30 Table A2: The 2012 Administrative Act and Presence of Workers with Disability: Robustness check One or more worker w/ disability Number of workers w/ disability (1) (2) (3) (4) Year>2012X Distance to QL threshold>0 0.049∗∗∗ 0.062∗∗∗ 0.052 0.042 (0.009) (0.011) (0.070) (0.097) N 282594 265016 282594 265016 Mean Dep. Var. 0.328 0.944 0.919 0.944 h (left) 16.175 16.175 16.175 16.175 h (right) 90.242 90.242 90.242 90.242 R2 0.195 0.198 0.108 0.108 Donut ring 1 2 1 2 Note: This table presents estimations from equation 1. All estimations include city-by-year fixed effects. Significance levels are indicated by ∗< .1, ** < .05, *** < .01. Standard errors clustered at the city level shown in parentheses. Figure A2: The 2012 Administrative Act and Presence of Workers with Disability: Robustness Checks -.05 0 .05 .1 .15 % Growth: Large vs. Small Firms 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 (a) One or more worker w/ disability -.2 0 .2 .4 .6 % Growth: Large vs. Small Firms 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 (b) Number of workers w/ disability Note:These graphs present estimations from a model similar to equation 1, where we substitute the post-2012 dummy Twith year dummies, leaving 2011 as the benchmark. We consider a donut ring of two in the firm-size variable. All estimations include city-by-year fixed effects. 90% confidence interval shown in the graphs. Table A3: The 2012 Administrative Act and Other Firm Outcomes Non-QL Total wage Turnover Firm Wage workers w/ fines bill rate closure disability (ln) (1) (2) (3) (4) (5) Year>2012 X Distance to QL threshold>0 -0.007∗-0.015 -0.002 -0.003 -0.056 (0.003) (0.013) (0.003) (0.005) (0.038) N 641525 298207 426112 357065 113520 Mean Dep. Var. 0.125 12.051 0.314 0.035 7.952 h (left) 45.211 22.223 24.076 13.552 30.213 h (right) 185.839 82.449 246.241 277.317 126.129 R2 0.090 0.440 0.102 0.061 0.267 Donut ring 2 2 2 2 2 Note: This table presents estimations from equation 1. All estimations include city-by-year fixed effects. Significance levels are indicated by ∗< .1, ** < .05, *** < .01. Standard errors clustered at the city level shown in parentheses. 31 Figure A3: The 2012 Administrative Act and other Firm Outcomes -.04 -.02 0 .02 .04 Diff. in Likelihood: Large vs. Small Firms 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 (a) Non-QL Fines -.1 -.05 0 .05 .1 % Growth: Large vs. Small Firms 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 (b) Total Wage Bill -.02 -.01 0 .01 .02 % Growth: Large vs. Small Firms 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 (c) Turnover Rate -.04 -.02 0 .02 .04 % Growth: Large vs. Small Firms 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 (d) Firm Closure -.2 -.1 0 .1 .2 .3 % Growth: Large vs. Small Firms 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 (e) Wage Workers with disability (ln) Note: These graphs present estimations from a model similar to equation 1, where we substitute the post-2012 dummy Twith year dummies, leaving 2011 as the benchmark. The dependent variables are the hyperbolic sine transformation of the variables indicated in each sub-figure. All estimations include city-by-year fixed effects. 90% confidence interval shown in the graphs. 32 Table A4: Firms’ Characteristics in t−1and QL Fine in Network in t 2012 2013 2014 2015 2016 2017 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Panel A: Neighbor network L.hyp_disabled -0.000 0.000 0.001 0.001 -0.002∗-0.004∗0.001 0.000 -0.001 -0.003 0.000 -0.001 (0.001) (0.003) (0.002) (0.004) (0.001) (0.003) (0.001) (0.002) (0.001) (0.002) (0.001) (0.002) L.hyp_firm_size_QL 0.001 -0.002 -0.000 -0.002 0.002 0.003 -0.000 -0.003 0.001 0.002 -0.001 -0.002 (0.001) (0.003) (0.002) (0.004) (0.002) (0.004) (0.002) (0.004) (0.001) (0.003) (0.001) (0.003) Never treated included Yes No Yes No Yes No Yes No Yes No Yes No N 22744 9926 23222 10033 22962 9927 21249 9286 19892 8717 19822 8145 R2 0.429 0.484 0.357 0.417 0.329 0.403 0.367 0.439 0.407 0.473 0.213 0.329 Panel B: Owner network L.hyp_disabled 0.002∗∗ 0.003 0.000 -0.002 0.002 -0.001 0.001∗-0.003 -0.001 -0.008∗∗ 0.001 -0.004 (0.001) (0.004) (0.001) (0.004) (0.001) (0.005) (0.001) (0.004) (0.001) (0.004) (0.001) (0.003) L.hyp_firm_size_QL 0.001 -0.012∗∗ 0.004∗∗∗ 0.000 0.001 -0.009 0.001 -0.004 0.002 0.002 0.001 -0.007 (0.001) (0.005) (0.001) (0.006) (0.002) (0.008) (0.002) (0.009) (0.001) (0.006) (0.001) (0.007) Never treated included Yes No Yes No Yes No Yes No Yes No Yes No N 22744 3719 23222 3654 22962 3564 21249 3361 19892 3166 19822 3089 R2 0.055 0.124 0.064 0.125 0.061 0.137 0.071 0.149 0.067 0.138 0.061 0.138 Panel C: HR workers network L.hyp_disabled 0.001 -0.002 0.001 -0.002 -0.000 -0.003 0.001 0.001 0.000 -0.002 0.000 -0.002 (0.001) (0.004) (0.001) (0.003) (0.001) (0.003) (0.001) (0.003) (0.001) (0.002) (0.001) (0.002) L.hyp_firm_size_QL 0.006∗∗∗ -0.004 0.005∗∗∗ -0.004 0.002 -0.017∗∗∗ 0.002 -0.014∗∗∗ 0.000 -0.006∗0.002∗-0.002 (0.001) (0.005) (0.001) (0.004) (0.001) (0.004) (0.002) (0.005) (0.001) (0.003) (0.001) (0.004) Never treated included Yes No Yes No Yes No Yes No Yes No Yes No N 22744 4332 23222 4559 22962 4652 21249 4426 19892 4277 19822 4194 R2 0.051 0.118 0.054 0.134 0.041 0.117 0.046 0.121 0.062 0.121 0.045 0.110 Note: This table shows estimations of the likelihood that a network receives a Quota Law fine each year, depending on the characteristics of firms belonging to such a network in the previous year. Odd-numbered columns include firms from never-treated networks, that is, that never received a Quota Law fine, while even-numbered columns exclude such firms, restricting the sample to networks used in our analysis. All estimations include city-by-year fixed effects. Significance levels are indicated by ∗< .1, ** < .05, *** < .01. Standard errors clustered at the city level shown in parentheses. 33 Table A5: Law Enforcement at Firm’s Networks: Robustness Checks One or more worker w/ disability Number of workers w/ disability (1) (2) (3) (4) (5) (6) All firms Non-compliers Partially compliers All firms Non-compliers Partially compliers Panel A: Neighbor network Post-event X Treated 0.023*** 0.038*** -0.001 0.432 0.719*** 0.046 (0.007) (0.010) (0.010) (0.508) (0.275) (1.399) N 118160 70940 42603 118160 70940 42603 N (firms) 12248.000 7861.000 5610.000 12248.000 7861.000 5610.000 Avg. firm size 368.404 385.448 351.608 368.404 385.448 351.608 Mean Dep. Var. 5.579 2.607 10.739 5.579 2.607 10.739 R2 0.172 0.193 0.193 0.044 0.101 0.082 Panel B: Owner network Post-event X Treated 0.014 0.032* -0.019 1.279** 1.063** 1.145 (0.013) (0.019) (0.011) (0.643) (0.434) (1.621) N 37049 20846 13948 37049 20846 13948 N (firms) 3794.000 2322.000 1729.000 3794.000 2322.000 1729.000 Avg. firm size 485.848 499.083 492.306 485.848 499.083 492.306 Mean Dep. Var. 8.649 3.983 16.025 8.649 3.983 16.025 R2 0.254 0.287 0.216 0.140 0.195 0.173 Panel C: HR workers network Post-event X Treated 0.003 0.011 -0.003 0.552 1.106** 0.078 (0.010) (0.015) (0.013) (0.442) (0.523) (0.760) N 37237 22122 13385 37237 22122 13385 N (firms) 3671.000 2415.000 1615.000 3671.000 2415.000 1615.000 Avg. firm size 513.923 575.726 419.657 513.923 575.726 419.657 Mean Dep. Var. 7.698 4.692 12.547 7.698 4.692 12.547 R2 0.218 0.252 0.203 0.218 0.301 0.273 Note: This table presents estimations from a model similar to equation 2, where we substitute the time dummies with a post-event dummy. The sample is composed of firms larger than 100 workers, which did not receive an inspection or any type of fine in t= 0. "Treated" is an indicator that some type of law enforcement took place in the firm owner’s network at t= 0. Elasticities of workers with disabilities are calculated based on Bellemare and Wichman (2020). All estimations include cohort and city-by-year fixed effects. Significance levels are indicated by ∗< .1, ** < .05, *** < .01. Standard errors clustered at the city level shown in parentheses. 34 Table A6: QL Fine in Firm’s Network: Robustness Check Using Callaway and Sant’Anna (2021) Neighbor network Onwer network HR workers network (1) (2) (3) Pre_avg -0.022 -0.038 0.011 (0.030) (0.030) (0.022) Post_avg 0.104∗∗∗ 0.154∗∗∗ 0.102∗∗∗ (0.027) (0.038) (0.031) Tm3 -0.067 -0.090 0.011 (0.067) (0.081) (0.066) Tm2 -0.017 -0.032 0.008 (0.032) (0.047) (0.039) Tm1 0.019 0.008 0.013 (0.023) (0.032) (0.028) Tp0 0.050∗∗∗ 0.073∗∗∗ 0.047∗∗ (0.019) (0.027) (0.023) Tp1 0.125∗∗∗ 0.158∗∗∗ 0.108∗∗∗ (0.028) (0.044) (0.034) Tp2 0.138∗∗∗ 0.233∗∗∗ 0.150∗∗∗ (0.046) (0.057) (0.050) N 24477 9887 10439 Note: This table presents estimations from equation 2, but using the method proposed by Callaway and Sant’Anna (2021) instead of the stacked differences-in-differences used in our main estimations. All estimations include city-by-year fixed effects. Significance levels are indicated by ∗< .1, ** < .05, *** < .01. Standard errors clustered at the city level shown in parentheses. Table A7: Law Enforcement at Firm’s Networks: Placebo with Firms Smaller than 100 Neighbor network Owner network HR workers network (1) (2) (3) Post-event X Treated 0.000 -0.006 -0.014 (0.002) (0.010) (0.012) N 599867 69379 29689 N (firms) 70133 7952 3614 Avg. firm size 41.717 49.181 54.135 Mean Dep. Var. 0.097 0.167 0.227 Elasticity 0.000 -0.006 -0.013 R2 0.045 0.164 0.194 Note: This table presents estimations from equation 1focusing on a sample of firms smaller than 100 workers, instead of firms larger than 100, as in our main estimations. All estimations include city-by-year fixed effects. Significance levels are indicated by ∗< .1, ** < .05, *** < .01. Standard errors clustered at the city level shown in parentheses. 35 Figure A4: Pre-trend Robustness .02 .04 .06 .08 .1 .12 90% Robust CI Original .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Mbar (a) Neighbor NW: All Firms .05 .1 .15 .2 .25 90% Robust CI Original .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Mbar (b) Neighbor NW: Non-compliers -.05 0 .05 .1 .15 90% Robust CI Original .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Mbar (c) Owner NW: All Firms 0 .05 .1 .15 .2 90% Robust CI Original .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Mbar (d) Owner NW: Non-compliers -.05 0 .05 .1 .15 90% Robust CI Original .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Mbar (e) HR workers NW: All Firms 0 .1 .2 .3 90% Robust CI Original .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Mbar (f) HR workers NW: Non-compliers Note: These figures report 90% confidence intervals for different deviations from linear trends between treated and not-yettreated firms, employing the methodology proposed by Rambachan and Roth (2023) to assess the sensitivity to parallel trends violations. Mbar = 0 means no difference in linear trends and Mbar > 0allows for deviations in linearity. 36