12
Local officials' connections with the central government have attracted a lot of attention in the literature of the Chinese political economy (Huang, 1996 and Jin et al., 2000). It is believed that provincial leaders with prior or current central experience tend to have more attachment to the center and to be thereby more compliant with central policies. Alternatively, the central experience may also indicate a lack of local knowledge, which may have a negative effect on economic performance.
13
We are very thankful to an anonymous referee for making this suggestion.
14
Controlling for both cyclical and provincial effects provides us with a difference-in-difference estimate of the effects of performance on the likelihood of turnover. It also helps address the concerns about the effects of performance relative to national or provincial benchmarks on the probability of turnover.
15
Two provincial units, Tibet and Hainan, are excluded from the sample because Tibet has a very special policy status while Hainan is a spin-off from Guangdong province, only becoming a separate province in 1988. In any event, including Tibet and Hainan in the sample does not change the qualitative results.
16
Another necessary condition is that the central government has the ultimate authority to appoint or remove provincial leaders. The authority of the center is beyond any doubt, since the political system in China is highly centralized.
17
Following Denis and Denis (1995), this rate is defined as the total turnover number divided by the total number of province-years in the sample. Since each province has two top leaders, we double the total number of province-years to calculate the overall annual turnover rate.
18
The annual provincial GDP is calculated at 1980 prices.
19
See Murphy and Zimmerman (1993) and Weisbach (1988) on studies of the relationship between CEO turnover and firm performance.
20
We thank Roger Gordon for making this suggestion.
21
We do not report regression results due to space limitations. Interested readers can contact the authors for these results.
22
The likelihood function is also maximized around r=1, lending support to our focus on the average growth rate as the performance measure.
23
We also tried using the age65 indicator interacted with a post-1984 or post-1985 indicator, and the results are similar.
24
We thank an anonymous referee for suggesting these methods.
25
The variable second term indicator does not have enough variation in the second subsample and is therefore dropped in the estimations