X hits on this document

64 views

0 shares

0 downloads

0 comments

9 / 25

regression parameter is to capture the “individuality” of each facility without actually requiring data on all of the differences between facilities. In other words, αi exploits the panel nature of the data to account for facility-specific confounding factors (like size, age, industrial sub-category, and profitability) without actually requiring data on these confounders.

The fixed effect empirical model holds the slope coefficient (representing the impact of an additional regulatory action on future pollution or compliance) constant for all facilities, but allows each facility to have its own regression intercept αi. This approach accounts for the “individuality” of each facility and implicitly controls for all facility-specific confounding factors that are approximately constant across time, like size, profitability, and industrial sub-category. Intuitively, the identification of the fixed effects model can be interpreted as a difference-in-differences estimator. Here, the specific deterrence impact of a marginal (additional) regulator action on compliance or pollution is the difference between (a) the difference between post-action pollution or compliance and the average pollution or compliance levels for all facilities that had received an action in the recent past and (b) the difference between the same time periods and the average pollution or compliance levels for all facilities that did not receive an regulatory action in the recent past. The general deterrence impact of a marginal (additional) regulator action is similar, except that the regulator action is on other facilities in the same state and sector, rather than on plant i itself.

Random effect models also attempt to capture the “individuality” of facilities while holding the slope coefficient (representing the impact of an additional regulatory action on future pollution or compliance) constant. However, the modeling approach for αi differs from the fixed effect approach. Instead of allowing each facility its own intercept as in the fixed effects model, the random effects model assumes a statistical distribution for these parameters around a common mean value. Intuitively, the identification of the random effects model can be interpreted like an ordinary least squares regression. Here, the specific deterrence impact of a marginal (additional) regulator action on compliance or pollution is the pollution or compliance difference between observations in which there was an agency action in the recent past to observations in which there was no agency action in the recent past, after controlling for confounding factors. The general deterrence impact of a marginal (additional) regulator action is the pollution or compliance difference between observations in which there was an agency action in the recent past directed towards others in the same state and sector to observations in which there was no agency action in the recent past directed towards others in the same state and sector.

Finally, the conditional random effects model also attempts to capture the “individuality” of facilities while holding the key slope coefficient constant. The intuition is identical to that of fixed effects, and the aim is still to control for missing variables potentially correlated with the key explanatory variables. Conditional random effects are persistent effects at the plant-level, like fixed effects, but they condition on the sample average of a few observed variables rather than all variables (as in fixed effects). Since the intuition is identical, and fixed effects are more comprehensive, one might wonder

8

Document info
Document views64
Page views64
Page last viewedTue Dec 06 13:09:23 UTC 2016
Pages25
Paragraphs327
Words9837

Comments