2.1. Inspections, fines and compliance: conceptual framework

2.2. Description of the enforcement activities carried out by the SMA in Chile

The SMA is responsible for executing, organizing and coordinating the monitoring of environmental regulation in Chile. The SMA began its activities in 2010 but it was in full operation by the end of 2012.Footnote ⁴ The SMA conducts environmental inspections, promotes compliance and imposes sanctions for violations. Figure A1 in the online appendix illustrates its monitoring and enforcement process. Detected violations allow facilities to submit a plan of action, called a compliance program, within 10 days. Failure to comply with this program may result in fines of up to twice the original amount, while successful programs restore compliance (SMA, 2018). According to the existing environmental law, Chilean facilities must typically comply with a set of different specific environmental regulations. Most of these are related to command-and-control instruments such as emission standards, ambient quality standards where facilities are located, prevention and/or decontamination plans; and economic incentives-instruments such as emissions taxes. The specific environmental regulations are defined in the Environmental Qualification Resolution (Resolución de Calificación Ambiental in Spanish) mentioned earlier, which contains a set of environmental permits that each facility must fulfil (SMA, 2018).

The SMA considers environmental risk, location and facility characteristics in its monitoring actions. Facility complexity depends on the number of environmental instruments and permits held. Non-compliant facilities face sanctions, including warnings, fines or closures. An overview of SMA activities highlights challenges in complying with regulations on wastewater discharges, air emissions, flora and fauna protections and solid waste. Infrastructure and operational issues also present compliance difficulties (SMA, 2018). Concerning facility characteristics, the size of a facility as a measure of sales is especially interesting to the SMA since it is used as input for defining the fines.Footnote ⁵ Regarding the relationship with communities, the SMA has faced an increasing number of complaints from the community against facilities that generate noise and odour emissions, which account for 75 per cent of the total volume of complaints (SMA, 2022). A notable subgroup within the agroindustry is pig farming operations, which face many complaints due to unpleasant odours and are primarily located in rural and lower-income areas across central Chile.

3.1. Data description

The main source of our data is the National Environmental Inspection Information System. It presents a detailed description of each facility included in our study. Each facility is a ‘physical unit in which actions and processes are regulated by one or more instruments of SMA competence’ (SMA, 2018). We include in our analysis facilities that meet at least one of the following three conditions: (1) belong to agroindustry, fishing, aquaculture, mining energy, industrial factories, environmental sanitation, housing and construction, (2) have at least one environmental permit for operation; and (3) are subject to compliance with water emissions.Footnote ⁶ Following these criteria, our study considers a total of 6,670 facilities operating during the period 2013–2019.Footnote ⁷ Figures A2–A10 and tables A1–A8 in the online appendix show details of our data.

Figure A2 presents the total number of facilities and the distribution by sector of activity through the study period. We consider the first year of a facility's operation to be the year it obtained the environmental permit. We assume that the facility remains operational until the end of our analysis, as the SMA does not provide information on whether a facility is still actively operating. In other words, there are no facilities that leave the SMA's oversight in our dataset. Moreover, we also assume that, during the studied period, there is no change in individual characteristics such as size, property or geographic location. At the end of the year 2019, the distribution of facilities in our sample, by sector, is the following: fishing and aquaculture (37 per cent), environmental and sanitation (13 per cent), housing and construction (12 per cent), energy (11 per cent), agroindustry (11 per cent), mining (10 per cent) and industrial factories (6 per cent). Figure A3 illustrates the geographical distribution of facilities across four categories. The first group includes facilities from the energy and mining sectors, primarily located in Chile's northern and central zones. The second map shows the Agroindustry sector, mainly concentrated in the central part of the country. The third group refers to the fishing and aquaculture sectors, which have facilities along the entire coastline of the country. Lastly, we grouped the other sectors, including environmental and sanitation, housing and construction, and industrial factories, which also have facilities primarily in the central zone. Table A1 shows the number of facilities by size per sector. Figures A4 and A5 show the distribution of facilities per year, by size and by zone, respectively.

Table A2 summarizes the number of facilities in each sector and those inspected annually from 2013 to 2019.Footnote ⁸ Inspected facilities represent 2.2–2.9 per cent of all SMA-regulated facilities yearly, ranging from 140 in 2015 to 187 in 2019. In 2019, the mining sector had the highest inspection rate (6.4 per cent) with 44 facilities inspected, while housing–construction had the lowest (0.2 per cent) with only 3 inspections. Fishing–aquaculture had 31 inspections, a low 1.3 per cent of its total facilities. Figure A6 shows the number of facilities inspected each year. Table A3 shows the aggregate figures regarding the number of inspections and their compliance outcomes by sector during our study period. From the 1,141 inspections carried out during the period 2013–2019, 632 resulted as compliant, while the other 509 inspections uncovered violations. This suggests that the rate of compliance during the studied period is about 55 per cent, while 45 per cent of the inspections resulted in violations during the same period. These inspections correspond to 754 facilities being inspected; 538 facilities were compliant (71 per cent), and 216 facilities were in violation (29 per cent). We notice that the sectors that show higher compliance are energy, industrial factories, mining and agroindustry, and sectors with higher rates of violations are housing–construction and environmental sanitation.

Table A4 presents the number of implemented fines, the aggregated amount of fines and the average fine by economic sector during the period 2013–2019. These figures also include the results of sanctioning procedures that were started before 2013 and finished with a fine at the end of our studied period. This table focused on analysis of the 104 cases of non-compliance that resulted in fines, offering a detailed view of the fines imposed. Out of the 509 inspections resulting in non-compliance (table A3), 354 compliance programs were initiated. According to data from Sistema Nacional de Información de Fiscalización Ambiental, over 80 per cent of these compliance programs were successfully completed, bringing the facilities into compliance. This highlights that only a subset of non-compliance cases, those that either did not successfully complete a compliance program or did not submit one, proceeded to fines, as captured in table A4. The total amount of implemented fines during the period of study is about US$84 million. The average fine is US$805,000. Sectors of mining and agroindustry show more facilities being fined. Mining and energy sectors present the larger average fine implemented in the period, which is about US$2,700,000 and US$1,060,000, respectively. Each implemented fine is analysed here, which provides detailed information about the severity of the non-compliance incidents. The SMA categorizes each identified incident as a distinct infraction within each fine. On average, each fine in our sample includes six infractions, resulting in a total of 594 infractions across the 104 fines analysed.

Table A5 shows figures regarding the distribution of infractions according to their severity (low, medium, high) by economic sector during the period of our study. From the 598 detected infractions, 386 (65 per cent), 195 (33 per cent) and 17 (3 per cent) correspond to low, medium and high severity, respectively.Footnote ⁹ Tables A4 and A5 show that the mining sector accounts for the largest share, about 70 per cent, of SMA fines, driven by two key factors: (1) the SMA likely considers the economic benefits violators gain from activities in this sector, justifying higher fines, and (2) the severity of infractions significantly influences fine amounts, with the mining sector accounting for a substantial proportion of medium- and high-severity infractions. In contrast, other sectors primarily exhibit low-severity infractions, typically resulting in smaller fines.

Of the 1,141 inspection processes, 632 concluded in compliance, as shown in table A3. However, 17 per cent of these compliant cases (110 out of 632) correspond to facilities with multiple owners, making it interesting to test empirically if these spillover effects exist and how enforcement actions from the SMA contributed to compliance through this indirect channel. In tables A6 and A7, we provide additional descriptive statistics on the submission of compliance programs. Agroindustry and fishing–aquaculture are the sectors that have submitted the most programs, likely due to the high proportion of low-severity infractions in these sectors, as shown in table A5. Additionally, we analyse correlations between factors associated with submitting compliance programs, identifying facility size as a key determinant. Micro and small facilities are less likely to use this alternative, potentially due to limited resources or lack of experience in managing these programs and responding adequately within the first 10 days after being identified as violators. Nevertheless, the predicted probability of submitting a compliance program after being identified as a violator is high, around 80 per cent. Compliance programs are a widely used instrument promoted by the SMA, enabling violators to achieve compliance through process improvements, adoption of decontamination technologies or implementation of better practices.Footnote ¹⁰

A special characteristic of the observed units in our analysis is that one facility may be controlled by more than one firm. In our data, 12 per cent of the sample (772 facilities) have multiple firms as their owners.Footnote ¹¹ These facilities may receive information about the SMA's enforcement actions at other facilities under the same ownership. Figures A7 and A8 illustrate examples of shared ownership among facilities. Finally, our data also include the annual evolution of the SMA's budget, number of inspectors and employees, as shown in Figures A9 and A10. A general summary of the statistics from our sample is presented in table A8.

3.2. Inspections and implemented fines

3.2.1. SMA's inspections

Our aim is to identify factors correlated to inspections from the SMA. The general panel data specification for our empirical analysis of inspections is given by

(1)\begin{equation}INS{P_{it}} = {\alpha _i} + \beta M{E_{it{^{\prime}}}} + \gamma F{C_i} + \delta L{O_i} + \eta {B_{it}} + \mu {Y_t} + {\varepsilon _{it}}.\,\end{equation}

The dependent variable $INS{P_{it}}$ is a 0/1 binary variable indicating SMA's inspection at facility i in year t. It takes a value of 1 if the facility was inspected in a given year and 0 if the facility was not inspected during that year. The set of variables under $M{E_{it}}$ includes monitoring and enforcement actions faced by facility i during the last 3 years $\left( {t'\, = \,t - 1,\,\,t - 2,\,\,t - 3} \right)$, such as being inspected last year, fined during the last 3 years and being currently operating under a compliance program. $F{C_i}$includes individual facility's characteristics, such as size, economic sector to which it belongs and age. The variable size categorizes facilities by annual sales: Micro and Small (up to US$850,000), Medium (US$850,000–3,400,000) and Large (over US$3,400,000), as defined by the Chilean Tax Authority. ‘Age’ estimates facility age based on the year of its first environmental permit approval. $L{O_i}$ is a set of variables related to the facility's location as a categorical variable that divides the country into five large zones from north to south, a dummy variable if the facility is located in an area that is prioritized by the SMA, and socioeconomic and demographic characteristics of the location where it operates. Finally, to control for the budget, we include ${B_{it}}$ which is an indicator of the SMA's available budget per regulated facility each year calculated as the total budget available for the region where the facility is located, divided by the total number of facilities within that region at a given time. We also add a set of dummy variables ${Y_t}$ to control for years effects.

We address potential endogeneity by using lagged variables for monitoring and enforcement actions for reducing reverse causality concerns. Since fixed effects models cannot estimate the impact of time-invariant facility characteristics (e.g., sector, size), which are central to our study, we adopted the conditional random effects (CRE) logit models, following Shimshack and Ward (Reference Shimshack and Ward2005). We include means of time-varying covariates for the set of variables related to monitoring and enforcement. Including these mean variables also reveals long-term correlations, offering additional insights for analysis.Footnote ¹²

3.2.2. Implemented fines

According to the procedures of the SMA, non-compliant facilities may be sanctioned at the end of a sanctioning procedure. Our purpose is to analyse what determines the size of the implemented fines defined by the SMA. We focus on fines since, during the study period, more than 95 per cent of the sanctions imposed on non-compliant facilities correspond to fines.Footnote ¹³ The cross-sectional empirical model is given by

(2)\begin{equation}FIN{E_i} = \alpha + \beta {^{\prime}}INFRACTIO{N_i} + \delta {^{\prime}}RELAPS\ {E_i} + \gamma {^{\prime}}F{C_i} + \mu {^{\prime}}{Y_t} + \varepsilon _i^{'}.\,\end{equation}

The dependent variable $FIN \, {E_i}$ is a left-censored variable indicating the size of the fine imposed on the facility i (in thousands of US$, $FIN \, {E_i} \gt 0$). The main explanatory variables are related to the severity of the violation/infraction. We want to evaluate if the implemented fine is affected by the number and severity of the detected infractions. We consider three different ways to evaluate the impact of infractions on the fine implemented. First, $INFRACTIO{N_i}$ is the number of total infractions detected by the SMA at facility i, regardless of the severity of them. Second, $INFRACTIO{N_i}$ is a set of variables that includes the sum of infractions at each severity level (low, middle, high). Third, we consider an impact index related to the infractions of the facility. This index is constructed as a weighted average of the number of infractions by severity level; the weight is equal to 1, 2 or 3 for low, middle and high severity, respectively. Other relevant covariates are $RELAPS{E_i}$, a binary variable that indicates whether the facility has previously received another fine, and $F{C_i}$ a set of individual facility's characteristics such as size and economic sector to which it belongs. We also add a set of dummy variables ${Y_t}$ to control for years effects.

We use a left-censored cross-sectional Tobit model to estimate what determines the size of the fine imposed on the facility, since the panel model is not convenient at this stage due to the time sequence being lost in the several stages that a fine imposition involves, such as the presentation of compliance plans or litigation. Additionally, the number of fines imposed by the SMA is limited, as most facilities that face this process present and fulfil a compliance program. Therefore, the cross-sectional model allows us to study the relationship with the severity of the infractions more clearly.

3.3. Environmental compliance

In this section, we present our empirical strategy to study the impacts of monitoring and enforcement actions on compliance behaviour. Our analysis considers the potential spillover effects on individual compliance of facilities within the same location and sectors, with a novel focus on facilities that share the same firm owner. The general panel data specification for the analysis of individual compliance behaviour is

(3)\begin{equation}COMPLIANC{E_{it}} = {\theta _i} + \lambda M{E_{it{^{\prime}}}} + \rho F{C_i} + \omega L{O_i} + \sigma SPILLOVE{R_{it}} + \tau {Y_t} + {\upsilon _{it}}.\end{equation}

The dependent variable $COMPLIANC{E_{it}}$ is a 0/1 binary variable indicating facility i in year t compliance status revealed by an inspection. It takes value of 1 if the facility was in violation in a given year and 0 otherwise. As before, the set of variables under $ME$ includes monitoring and enforcement actions faced by facility i during the last 3 years ( $t{^{\prime}} = t - 1,\,\,t - 2,\,\,t - 3$), such as being inspected last year, fined during the last 3 years and being currently operated under a compliance program. $F{C_i}$ includes individual facility's characteristics, such as size, economic sector to which it belongs and age. $L{O_i}$ is a set of variables related to the facility's location as a categorical variable that divides the country into five large zones from north to south, a dummy variable if the facility is in an area that is prioritized by the SMA, and socioeconomic and demographic characteristics of the location where it operates. $SPILLOVE{R_{it}}$ is a set of variables that consider different potential spillover effects. These are a set of indicators for four types of relationships: facilities located in the same commune, facilities within the same sector, facilities sharing both the same commune and sector and facilities owned by the same firm. These variables are binary and take the value of 1 if any facility within these relationships has received a fine in the past 3 years. We also add a set of dummy variables ${Y_t}$ to control for years effects.

We address potential endogeneity by using lagged variables for monitoring and enforcement actions to reduce reverse causality concerns. Since fixed effects models cannot estimate the impact of time-invariant facility characteristics (e.g., sector, size), which are central to our study, we adopted the CRE logit models. We include means of time-varying covariates for the set of variables related to monitoring, enforcement and spillovers. The sample for which we know the compliance status is not randomly selected but chosen by the SMA, so we incorporated a prior stage estimation to address sample selection. From equation (1), we use variable B (associated with the SMA budget) as an instrument for the inspection decision. This variable is appropriate as it reflects the resources available for the SMA to conduct inspections but does not directly influence a facility's compliance, except through its impact on the likelihood of inspection. We then include the predicted probability of facing an inspection as a control in equation (3) for environmental compliance. We address potential endogeneity by using lagged variables related to monitoring and enforcement actions, as well as spillovers. The past values of these variables are less likely to be influenced by current compliance behaviour, thus helping to establish a clearer direction of influence and mitigate reverse causality. We do not claim causality but aim to identify the mechanisms relevant to the relationship in compliance behaviour.

3.4. Brief discussion of threats to causal interpretations

While our analysis provides insights into the relationships between monitoring, enforcement and compliance, factors such as omitted variable bias, reverse causality and selection bias challenge the establishment of causal links. As a result, we expect to obtain findings that are associative rather than strictly causal. Below, we discuss the key challenges to drawing causal conclusions in our study. One of the main concerns in identifying causal relationships is the possibility of omitted variable bias. Despite controlling for a range of facility-specific and locational characteristics, there may still be unobserved factors that influence both the likelihood of inspection and compliance. For example, political or community pressure, unmeasured local environmental risks (such as industrial air pollution or water contamination from agricultural runoff) or firm-level characteristics (such as corporate environmental policies) could simultaneously affect both the probability of a facility being inspected and its compliance behaviour. Regarding facility-specific and locational characteristics, we assume they are time-invariant in our model, as we do not have detailed information about potential changes or expansions over time. The absence of these unobserved factors could result in biased estimates of the impact of monitoring and enforcement on compliance outcomes. Similarly, in the fines analysis, unobserved facility-level strategies might influence the size of the fines imposed.

Another significant threat to causal interpretation is reverse causality, particularly in the relationship between past enforcement actions (e.g., fines and inspections) and future compliance behaviour. While our models assume that inspections and fines affect compliance, it is possible that non-compliant facilities are more likely to be targeted for inspections or to receive fines in the first place. To mitigate this concern, we use lagged variables for monitoring and enforcement actions in our models. For example, in the model of inspections, we include inspections and fines from previous years. However, despite these efforts, the possibility of reverse causality remains, especially since facilities that have been non-compliant in the past may receive increased attention from SMA. Selection bias presents another challenge, particularly due to the non-random selection of facilities for inspection. Facilities more likely to be inspected are those already identified as higher risk or non-compliant, meaning that our analysis may over-represent these types of facilities. While we attempt to address this by including controls for facility-specific characteristics and sectors, as well as by using the predicted probability of inspection in our compliance model, this selection process could limit the generalizability of our findings. Facilities that are less visible to regulators or pose a lower risk may behave differently but are underrepresented in our sample of inspected facilities. As a result, the relationships identified in our study could be biased towards those facilities with a history of non-compliance or greater environmental risk.

4.1. Inspections

Table 1 highlights key factors influencing SMA inspections, with coefficient estimates from four CRE logit model specifications. Model 1 includes $ME$ variables, model 2 adds $FC$ variables, model 3 incorporates $LO$ variables and model 4 includes B variables. All models control for time-variant variable means and year fixed effects. Tables A9 and A10 in the online appendix show the detailed mean marginal effects and the conditional fixed effects logit model for inspections, respectively. We estimate that the predicted probability for inspections is 2.6 per cent, indicating that facilities have a relatively low likelihood of being inspected in each period, reflecting limited SMA monitoring efforts relative to the volume of facilities it regulates. Our findings show distinct patterns regarding $ME$ variables: facilities inspected in the prior year are 3 percentage points less likely to be inspected again, suggesting SMA's strategy to rotate inspections and broaden oversight. However, a positive coefficient for the mean past inspections indicates that facilities with a history of frequent inspections are more likely to be inspected again, reflecting SMA's institutional focus on these facilities. These findings emphasize SMA's dual approach of rotating inspections annually while maintaining oversight of historically monitored facilities.

Table 1. Coefficient estimates for inspection for conditional random effects logit models

Note: Standard errors in parentheses.

Our results show an interesting dynamic between compliance programs and inspections. Having a compliance program in progress shows no short-term effect on the likelihood of inspection, reflecting the nature of these programs, which are primarily managed online and do not require frequent inspections during implementation. In contrast, the positive and significant effect of 1.4 percentage points for the variable means of compliance program suggested that facilities with a history of compliance programs are more likely to be inspected over time (see table A9). This likely reflects increased monitoring by the SMA as these programs approach finalization, typically around 3 years after initiation, to ensure full compliance is achieved. Economic sectors also play a crucial role of facility's characteristics ( $FC$) in influencing SMA inspections. Facilities in agroindustry and mining are more likely to be inspected, with marginal effects of 0.92 and 0.6 percentage points, respectively, compared to the fishing–aquaculture sector. This prioritization aligns with sectors posing greater environmental impacts or economic importance. Conversely, sectors like housing and construction show lower inspection likelihood, reflecting reduced environmental risk or regulatory focus. Additionally, newer facilities are more likely to be inspected, highlighting the SMA's focus on early compliance or previously unmonitored sites, consistent with lag inspection findings.

Location characteristics also play a role, as facilities in prioritized areas, such as those near national parks, have a 0.35 percentage point higher probability of inspection. Facilities in poorer regions are less likely to be inspected, which may reflect the SMA's focus on wealthier regions with greater economic activity or larger facilities that pose higher perceived environmental risks. Logistical challenges in poorer regions may further limit inspection frequency. We also show that the (log) SMA budget is a significant driver of inspections. A 1 per cent increase in the regional SMA budget per facility leads to a 0.38 percentage point increase in inspection probability. This highlights the crucial role of resource allocation in regulatory monitoring and underscores the importance of adequately funding SMA to improve compliance enforcement. Finally, we observe a downward trend in inspections during 2015 and 2016, likely reflecting the SMA's initial years of operation, focusing on prioritization and strategy development following a high inspection rate in 2014. This trend aligns with the descriptive statistics shown in Figure A6, demonstrating that our model captures the SMA's specific inspection patterns.

4.2. Implemented fines

Table 2 presents the coefficient estimates for the (log) size of implemented fines. Our results show that both the number and severity of infractions substantially affect fine sizes. Model 1 shows that an additional infraction increases the fine by approximately 15.9 per cent.Footnote ¹⁵ In Model 2, infractions of low and middle severity raise fines by approximately 12.5 and 32.2 per cent, respectively, while high-severity infractions do not have a significant impact. This is mostly related to the fact that high-severity infractions are only 3 per cent of the total amount of fines imposed (please see table A5). Model 3 confirms these findings, indicating that a one-unit increase in the impact index raises the fine by approximately 13.1 per cent. This result aligns with expectations, as our constructed impact index has a weight equal to 1 for low severity infractions, making the coefficient similar to the one for low severity infractions in Model 2. Our results show in all the models that facilities with a history of infractions, indicated by the variable Relapse, are fined more severely by the SMA, with fines approximately 4–5 times higher. Regarding facility size, micro and small facilities are associated with fines that are approximately 70–73 per cent lower compared to medium-sized facilities.

Table 2. Coefficient estimates for the (log) size of fine for Tobit models

Note: Standard errors in parentheses.

Sector analysis reveals that mining and energy facilities face significantly higher fines, equivalent to 10 and 20 times higher than the fishing–aquaculture sector, respectively. This suggests that the SMA considers the negative impacts (damage) generated by violations in the mining and energy sectors to be more relevant, or that the economic benefits obtained from the violations are higher. Yearly trends show a significant increase in fines in 2017 compared to 2013. This can be explained by three main factors. First, sanction processes, whether resulting in compliance programs or fines, are lengthy, often taking around 3 years to complete, meaning many cases initiated in 2013 and 2014 likely concluded in 2017. Second, the SMA's budget significantly increased in 2016 (see Figure A10), enabling more enforcement actions and fine impositions. Lastly, the end of Michelle Bachelet's left-wing presidency in 2017 and the transition to Sebastián Piñera's right-wing administration may have expedited the resolution of pending cases, including changes in SMA leadership. In summary, the size of fines is significantly influenced by the number and severity of infractions, relapse history, facility size and sector, with the mining and energy sectors showing particularly higher fines.

4.3. Compliance and spillover effects

Table 3 identifies key factors influencing facility compliance using CRE logit models, with detailed marginal effects in online appendix table A11. Table A12 (online appendix) shows the conditional fixed effects logit model for compliance. Model 1 includes $ME$variables, Model 2 adds $FC$variables, Model 3 incorporates $LO$variables and Model 4 includes $Spillover$variables. All models control for time-variant variable means and year fixed effects, following the CRE approach. Estimation occurs in two stages, using the predicted probability of past inspections based on prior SMA decisions. Results show a high predicted compliance probability, about 70 per cent for previously inspected facilities. First, we analyse the results regarding $ME$variables. We show that correlation between being inspected last year (predicted) on compliance highlights a clear distinction between short-term and long-term impacts. In the short term, the variable shows no significant effect on compliance, indicating that inspections conducted in the previous year do not immediately influence a facility's likelihood of compliance. However, in the long term, the positive and significant effect of the mean of this variable suggests that repeated or consistent inspections over time foster higher compliance.

Table 3. Coefficient estimates for compliance for conditional random effects logit models

Note: Standard errors in parentheses.

Results show that lagged fines positively impact compliance in the short term, supporting the SMA's enforcement goal by prompting facilities to adjust their behaviour and adhere to standards. This demonstrates the effectiveness of fines as a specific deterrent. However, the negative correlation with mean lagged fines suggests that for a small subset of systematically fined facilities, fines may lack deterrence, potentially due to high abatement costs. Table A8 reveals that repeat violations among fined facilities represent less than 5 per cent of the sample, indicating these cases are a minority. The results for the compliance program variable show a negative but not statistically significant effect on short-term compliance. This aligns with the nature of compliance programs, which are part of a longer sanctioning procedure initiated when a facility is found in non-compliance. In the long term, the negative effect observed in the mean of the compliance program variable likely reflects the extended nature of these programs. Facilities remain categorized as non-compliant until the program is successfully completed, which often takes up to 3 years. Therefore, compliance records are expected to still reflect non-compliance for facilities systematically engaging in compliance programs, as these are still in progress.

The results for $FC$variables highlight significant differences in compliance across sectors. Compared to the reference sector, fishing–aquaculture, most sectors exhibit lower compliance probabilities. Notably, sectors such as housing–construction and environmental sanitation show the largest negative effects on compliance, with substantial deviations from the baseline, but these are not the focus of monitoring of SMA as we saw on results of inspections. The mining sector also shows significantly lower compliance, potentially due to the complexity and scale of operations in this sector, which may make achieving compliance more challenging. The results for the $LO$variables highlight the influence of community characteristics on compliance. Variable (log) density shows a positive and significant effect, with a 1 per cent increase in density associated with a 1.87 percentage point increase in compliance. This suggests that facilities in densely populated areas face greater oversight from both the SMA and local communities, incentivizing better compliance and emphasizing the role of community and demographic factors.

The analysis of $Spillover$effects reveals interesting dynamics. Imposing fines on facilities with the same owner in the past 3 years has shown a significant positive impact on compliance, increasing the probability of compliance by about 13.3 percentage points. This finding indicates a deterrence effect across facilities with the same owner, demonstrating the effectiveness of fine impositions beyond the violating facility, transmitted to their closely related firms with the same owner. Additionally, we observe a long-term positive spillover effect, as indicated by the mean variable for facilities in the same sector. Finally, our results for the year variables show a general stability in compliance over time, with a notable exception in 2016, where a positive effect is observed at the 10 per cent significance level. This slight increase in compliance during 2016 may reflect temporary enforcement efforts or specific regulatory actions taken by the SMA during that year, although the effect is not strong enough to suggest a consistent trend. Overall, these findings emphasize the multifaceted nature of compliance behaviour, influenced by past enforcement actions, facility characteristics, location and spillover effects across facilities with the same ownership.