and time periods are characterized by significant difference in production processes, treatment technologies, environmental impact, and regulator attention.
Sectors for future consideration should be selected on the basis of data availability, environmental impact, and agency priorities. The external validity of the simplified models is also strongest for sectors with salient characteristics similar to the pulp and paper and iron and steel sectors used to calibrate the presented models. The common characteristics of these industries are large industrial sources, relatively similar production processes, relatively similar pollution treatment technologies, and geographic diversity. Several core program sectors in the completed Sector Facility Indexing Project are particularly good candidates for replication, since they have significant environmental impacts, significant data availability, and relatively large and homogeneous industrial facilities.
Recommendation 2: In the longer run, OECA should consider applying the simplified deterrence measurement models to datasets created from the extensive data available to the EPA (facilitated by Task 5 of the Statement of Work). Extensive Permit Compliance System water pollution discharges and violations data, Continuous Emissions Monitoring System air pollution discharges and violations data, Toxic Releases Inventory toxics data, RCRA Biennial Reporting System hazardous waste violations data, and Compliance Data System/IDEA air pollution violations data are available for analysis across a wide range of industries and time periods. In many cases, near-censuses of major facilities can be obtained.
Recommendation 3: As work continues in the Compliance and Deterrence Research project, particular care should be paid to the issue of reverse causality in the estimation of specific deterrence effects. Future work should allow for alterative lag specifications and additional conditional random effects corrections. Lagged monitoring and enforcement variables serve two important purposes. First, lags reduce statistical simultaneity (endogeneity) and help isolate the direction of causality. If contemporaneous monitoring and enforcement variables are included in the analysis, statistically detected correlations between these factors and compliance or emissions may reflect the causal effect of compliance or emissions on monitoring and enforcement due to regulator targeting. This reverse causality is mitigated using lags. Second, lags allow time for firms to alter their environmental behavior in response to regulatory actions. Alternative lagged specifications for the key explanatory variables should be considered for all future explorations, since firm response times and regulator targeting regimes may differ across sectors. Alternative lag specifications, like inspections 1 year ago, or 2 years ago, or 3 years ago should be considered for different sector analyses. In other words, future research in the compliance and deterrence project should not necessarily be bound to the exact lag variable specifications included in this