Surrounding the Cities from the Countryside: An Empirical Assessment of the Electoral Effects

Abstract

In this chapter, the actual electoral effect of the relationship vote-oriented grassroots strategy of Beijing-sponsored parties and the pandering of the pan-democrats to the protest vote are examined. Did the Beijing-sponsored parties’ grassroots strategy really achieve their intended effect? The chapter not only presents how their grassroots strategy led to their electoral successes in the District Councils but also how occupying the District Councils allowed them to marginalize the pan-democrats in the major battlefield, the Legislative Council. The empirical evidence challenges the conventional wisdom that the District Councils are politically insignificant.

Appendix: Details on the Regression Discontinuity Design

In a laboratory experiment, a researcher randomly assigns subjects to the treatment and control groups. Because randomization tends to produce relatively balanced control and treatment groups, the researcher can significantly minimize the risk of omitted variable bias. As such, the identified effect is more likely due to the treatment effect rather than the effects of other confounding factors. In the current context, an ideal research design would be to randomly assign District Council seats to political parties. For instance, some pro-establishment parties would land on districts that are ideologically predisposed to the pro-establishment camp, and some would land on districts in favor of the pan-democrats. In other words, we can avoid the situation where only districts that are ideologically inclined to Beijing-sponsored parties would self-select to be led by pro-establishment parties. We can then examine the effect of occupying a District Council seat on LegCo elections by comparing the vote shares obtained by Beijing-sponsored parties and those obtained by the pan-democrats.

In reality, I cannot affect the data-generating process of the District Council elections, but if I can apply regression discontinuity, a quasi-experimental research design, I would be able to find out the causal effect. The idea of regression discontinuity is simple.¹³ If occupying a District Council Constituency has some effect on the outcome of the legislative elections, the relationship between the pan-democrats’ LegCo vote shares and the Beijing-sponsored parties’ District Council vote shares, for example, should be best characterized by a function discontinuous at a certain threshold (such as 50 % in a two-party vote of the District Council elections). The discontinuous jump is the causal effect of capturing a District Council seat on the outcome of the LegCo elections. The reason is that no matter how close a pro-establishment’s District Council vote share gets to the threshold, the party would not get elected and hence cannot occupy the District Council office until its vote share just surpasses the threshold. The validity of regression discontinuity hinges upon the assumption that districts are very similar to each other in the neighborhood of the discontinuity. The only difference that sets them apart is whether or not they happen to receive the “treatment” by chance; that is, whether the Beijing-sponsored party obtains, due to random uncontrollable factors, barely sufficient votes to carry the districts. In other words, we have balanced treatment and control groups in the neighborhood of the discontinuity as if in a randomized experiment, so that the causal effect that is identified is more likely due to the effect of the treatment rather than the effect of other factors such as the district’s ideological predisposition. This argument is formalized in the study by Lee (2008).

To estimate the electoral effect of the District Council office, I examine the relationship between the pan-democratic camp’s LegCo vote share in each District Council Constituency (the dependent variable) and a Beijing-sponsored party’s (or a pro-establishment District Councillor’s) margin of victory in the same District Council Constituency (the independent variable or the forcing variable).

More formally, we can express the regression discontinuity design in the following way:

$$\displaystyle{D_{i} = \left \{\begin{array}{l l} 1&\quad \mbox{ if $x_{i} > 0$} \\ 0&\quad \mbox{ if $x_{i} \leq 0$}\\ \end{array} \right.}$$

where D _i is the treatment status of District Council Constituency i, with the value “1” denoting the constituency controlled by a Beijing-sponsored party (or a pro-establishment District Councillor) and “0” otherwise, while x _i is the pan-democratic camp’s margin of victory in the District Council Constituency i.

This leads to the main regression specification:

$$\displaystyle{ y_{i} = f(x_{i}) +\delta D_{i} +\mu _{i} }$$

(6.1)

where y _i is the pan-democratic camp’s LegCo vote share in constituency i, f(x _i ) is a polynomial function, δ is the causal effect of interest, and μ _i is an error term assumed to be independent and identically distributed. The polynomial function is intended to provide a flexible functional form to model the relationship between y _i and x _i , which is not necessarily linear, in order to avoid mistaking nonlinearity for discontinuity.¹⁴