China Policy Journal Volume 1, Number 1, Fall 2018 | Page 102

Subjective and Objective Air Quality in Urban China
(0.145 mg/m 3 ). PM 10
is one of the most
visible air pollutants that often lead to
citizen complaints.
Multilevel Model Estimations
We report MLM regression
results for both direct and
interaction effects of our
independent variables (see Table 2). In
the first column, the null model with no
predictors suggests that 16.39 percent
(interclass correlation (ICC) = 0.712/
(0.712 + 3.633)) of total variances in air
quality satisfaction could be attributable
to Level 2 predictors. The explanatory
power of Level 2 predictors is remarkably
strong, particularly if we consider
the small sample size (N=32) at Level 2
compared with large sample size at Level
1 (N=25,139) (Raudenbush and Bryk
2002). Actually, the impressive ICC,
0.1639, implies that residents nested
in each sampled city have very similar
perception on the air quality in the city.
This finding implies that subjective air
quality may significantly correlate with
objective air quality. But, whether subjective
and objective air quality are correlated
need further investigated while
controlling other variables. Although
the variance at Level 2 (0.712) is statistically
insignificant, it is substantially different
from the standard error (0.179),
suggesting it is essential to use MLM to
estimate our models.
In the rest columns of Table 2,
we sequentially enter our key independent
variables and moderating
variables. The results show that Level
2 air pollutants do have statistically
significant effects on subjective air
quality. When SO 2
is used as air pollution
measure in Model 2, its regression
coefficient is negative and significant
(β=−18.37, p<0.10) and statistically
significant at the 0.10 level. In Model 5,
NO 2
is used to gauge air pollution, and
we find that its effect is negative albeit
insignificant (β=−12.14, p>0.10). Both
PM 10
in Model 8 and air grade in Model
11 have significantly negative effects
on subjective air quality (β=−19.98,
p<0.01; β=−0.73, p<0.01). In a nutshell,
our results suggest that Hypothesis 1 is
partially supported and objective air
quality is one of the key antecedents of
subjective air quality.
We report the results on the
moderating effects test in Table 2. Most
of our interaction terms are statistically
significant and consistent with our
hypotheses, and therefore, Hypothesis
2 is partially supported. In Model 3,
AQTI has insignificant effect on subjective
air quality and its interaction
term with objective air pollution (herein
SO 2
) also has little effect of substance
(β=0.10, p>0.10). The insignificant effect
of AQTI can be attributable to its
substantial change over the short term
after MEP mandated EPBs to release
air quality information, which shrank
cities’ disparities in transparency (IPE
2012). When we turn to Model 4 with
PITI as our moderator, we find its interaction
term with air pollution have significantly
positive effect on subjective
air quality (β=1.51, p<0.05).
The moderating effect of environmental
transparency on the relationship
between NO 2
and subjective
air quality was not supported by using
99