McGill Journal of Political Studies 2014 April, 2014 | Page 16

other words, the smaller p-value is for the
paired t-test, the more confident we can be
to assert that the two groups, high-rankings
and low-rankings, are not treated identically
by the Party, which merits the dichotomous
approach.
Finding
The most noticeable result of the multiple
linear regressions, as seen in Table 1, is that
rank remains statistically significant in all
five models. Model 1 shows that rank alone
explains 25% of the variation in t, and we
can be highly confident (p<0.001) that the
rank has statistically significant relationship
with the duration of party-initiated
corruption investigation. Model 2, 3 and 4
illustrate that the explanatory power of rank
is consistently high while controlling for
xitong, impact and money.
Why, then, does the p-value of rank
increase to p<0.05 in the full model,
model 5? One explanation is that age
captures some characteristics of rank, and
the covariance between the two variables
dilutes explanatory power of rank, and it
subsequently strengthens that of impact. I
ran a regression between age and rank and
found that the two were indeed highly
correlated (p<0.001). This finding is not
surprising, for the mean age for the high &涖