used (Train, 1999). The program that estimates the RPL model is based on GAUSS
code developed by Train, Revelt and Ruud (1999).
Estimation
Conditional logit model results
Results for three models are presented in Table 3. To estimate standard errors, the
robust asymptotic covariance matrix estimator is used (Mc Fadden and Train). The
first column corresponds to the most restrictive, benchmark CL model in which coef-
ficient values are not permitted to vary across plant managers. All of the technology
specific fixed effects are negative, all but the low NOx burner (LNB) fixed effect are
significant at the 1% level. This suggests that, relative to the baseline option of no
control technology retrofit, the average manager was biased against adopting these
technologies (controlling for costs).
The variable operating cost and capital cost coefficients are also significant at the
1% level and have the expected negative sign, suggesting that an increase in either
capital or operating costs significantly reduces the probability that a given compliance
alternative will be chosen. The ratio of the variable cost and fixed cost coefficients is
3.75, suggesting that plant managers are, on average, willing to pay an additional $1
in capital costs so as to reduce annual ozone season operating costs by $3.75.
The second column of Table 4 presents the results from a nested likelihood ratio
test of this benchmark specification. The test statistic is larger than the χ2 statistic
with 2 degrees of freedom and a p-value of 0.001. This indicates that variable op-
erating cost and capital cost variables significantly improve the fit of the model (as
compared to a model that includes only technology fixed effects).
The second CL model (CLII) accounts for systematic differences in responsiveness
to variation in capital and variable compliance costs. The second column of Table 3
reports results for the second CL model. To account for the possibility that firms in
21