China Policy Journal Volume 1, Number 1, Fall 2018 | Page 127

China Policy Journal
Assessing prior randomness is complicated
in this case by the distributions of
pollution intensity and effluent concentration.
Both are highly skewed, with
skewness coefficients ranging from 3 to
9. In this case, the traditional student
t test for equality of pre-rating group
means is not appropriate. We employ
the nonparametric Wilcoxon–Mann–
Whitney test for equal means and the
K-sample test for equal medians. Our
results, reported in Table 4, show that
significant differences in means and
medians were common in the sample.
In Zhenjiang, where Green Watch began
in 1999, we find significant differences
in mean and/or median pollution
intensities for waste water, COD, and
dust and smoke as well as significant
differences in mean and/or median effluent
concentrations for TSS, COD,
and dust and smoke. Table 4 reports
similar findings for the other three
sample cities (Huaian, Wuxi, and Yangzhou),
where the first public disclosure
of ratings occurred in early 2001. 2 In
light of these results, it is appropriate
to introduce controls for pre-program
pollution in our estimating equation:
(2) Y it
−Y it-1
= β 0 + α 1 F it + α 2 C it + α 3 R it +
β 1
t + β 2
Y it-1
+ µ it
+ ε it
To determine the appropriate estimator,
we employ Breusch and Pagan
Lagrangian multiplier (BPLM) tests for
random effects. We reject the null hypothesis
in favor of the random effects
model for air pollution intensities, and
for air and water effluent concentra-
tions. We assume that ε it
was correlated
across firms within a city but uncorrelated
across firms in different cities.
Tables 5/6 and 7/8 present estimation
results for changes in pollution
intensity and effluent concentration,
respectively. In Tables 5 and 7, we test
whether a firm reduces pollution simply
because it was rated. A priori, it was
possible that self-scrutiny by a rated
firm resulted in better environmental
management and reduced pollution,
even if the firm had a good rating. Our
results for the regression variable PRD
were consistent with this hypothesis:
PRD rating had a negative impact on
pollution for all equations in Tables 5
and 7, and a statistically significant impact
on TSS and SO 2
for pollution intensity,
and dust and smoke for effluent
concentration.
Tables 6 and 8 provide more insights,
by identifying specific color ratings
for firms. We find strong results
for water pollution intensity (TSS and
COD) and dust-and-smoke intensity
in Table 6, with highly significant reductions
for the poorly rated firms that
were much larger than reductions for
the firms with better ratings. Intensities
generally declined more among the rated
firms for the other pollutants as well,
but without the striking differential for
the poorly rated firms. The same general
pattern holds in Table 8, with generally
declining effluent concentrations across
all rated firms and the largest impacts
among the poorly rated firms. Although
some concentration results were highly
2 We also conducted the equal mean and median tests for each of the three cities. The results are
qualitatively similar.
124