During the LegCo by-election last February, Alvin Yeung and Edward Leung are often described as archenemies whose voter bases largely overlap. Yu-Xi Chau looks into the statistics and suggests otherwise.
Politics, like economics, happens at the margins. Nowhere is this more true than in Hong Kong’s convoluted electoral system for the Legislative Council. However, it seems that the political operators are not the only people playing the margins – a great many Hong Kong voters are skillful at hedging their bets too.
The fog of electoral war
Predicting election results in Hong Kong is much more difficult compared to predicting, for example, the two candidate American Presidential or Senate races for two reasons:
- Hong Kong Legislative council uses a proportional representative system. Legislative councilors typically need only 7-8% of votes to win in large constituencies. Error in polls are typically 2-3%, so there can be large fluctuations.
- Polls are expensive and there are not many polls available in Hong Kong. Nate Silver, the American election guru, is able to liberally combine hundreds of polls to formulate a single prediction. In Hong Kong, arguably the only reliable poll comes from HKUPOP.
The intelligent crowd: how voters maximize their power
The bad news is, even the HKUPOP polls should be used with caution when predicting final results, as is evident from the plot below. We show the last-minute swing from the actual election results and the final poll rating of candidates on the horizontal axis, against the changes upon the actual result on the vertical axis.
In other words, this plot shows the last minute swing for all candidates from the last day of the polls (figures are actually corrected by a small factor based on constituency size, details available from the author). The two lines represent the smoothed average for both the pan-democracy camp and the pro-Beijing camps, while the grey area represents the 90% confidence interval.
This graph shows that the higher the candidate’s support in the polls, the more likely voters would not vote for them on Election Day.
Why is that happening? The idea is that in a proportional representative system, extra votes are wasted on the same candidate. Seeing one candidate had gained sufficient support already, voters would change minds and vote for a struggling candidate at the last minute.
Pan-democrats have always lamented the precision of pro-Beijing camp’s campaign machine. This is achieved by flawless use exit polls to mobilise a small percentage of voters to vote for struggling candidates. But according to the graph above, a similar effect is observed for pan-democrat candidates, as conscious voters will always attempt to maximize their vote’s utility. Typically a candidate with 5% final day polling result can be expected to have a boost of 2-3% at the expense of poll leaders polls in the same camp.
Surprisingly, pan-democrats’ ability to get the votes out is arguably on par with pro-establishment candidates. Crowds are intelligent, after all, and intelligent voters will always attempt to maximize the power of their vote.
Voting by faction
This brings us to a conclusion: voters vote by faction, and many voters will maximize the utility of their vote by switching to support a candidate on the same camp with lower poll ratings at the last minute. This explains a lot about LegCo elections.In the New Territories East constituency, the support rating across traditional ideological boundaries are largely stable, as the table below shows:
The distribution of political division has not changed much in the past 4 years. Even though candidates come and go, the total vote share of each camp is stable. What is interesting, however, is the increasing competition within the pan-dem camp.
If we break down the support of pro-Beijing and pan-Democrats to two sub-camps, we can see a 12% swing towards the extreme wing for the democrats (from 12.4% to 24.7%):
The swing was grossly underestimated in the 2016 by-election. In the first-past-the-post system, radical voters would naturally converge to a compromise candidate that has a higher chance of winning, that is, Alvin Yeung Ngok-kiu (楊岳橋) instead of Edward Leung (梁天琦). Note that the hard evidence for why the 2016 by-election underestimated the progressive democrats vote can be found in the 2015 district council elections, but this is beyond the scope of this article.
We expect to see the progressive democrats’ vote share to maintain or even be higher than the level achieved in 2012. The interesting topic now, naturally, is the competition between the moderate and progressive democrats.
They took my votes!
Correlation is a useful tool to study if two candidates have the same voter base. Looking at past results, we show the correlation relationships below. The correlation plot is a metric that measures how similar two distributions are, and in this case, the distribution of votes that each candidate has for each constituency in New Territories East. This figure compares candidates in the 2012 New Territories East election (at the bottom) and the 2016 by-election (on the right). In simple terms: the darker the colour, the more similar the voter base is.
We can see that Edward Leung, the Localist superstar, has a lot of voter base overlap with the conventional radical wings of the pan-Democrats (indicated by the three dark blocks on the top left corner) but not other moderate democrats (the lighter shades of blue at the top centre).
This is a very interesting development, as progressive democrats tend to be left-wing liberals (e.g. Civic Party) who detest what Edward Leung stands for. It also shows that, at least during an election, Localists are not moderate Democrat’s enemies. Moderate Democrats and progressive Democrats all face competition from within their respective camps.
It is unclear whether progressive democrat voters have strong ideological beliefs and changed their minds to vote Localists in the 2016 by-election or they simply wanted to vote for the most radical representative. My speculation leans to the latter, but we can only find out after the coming election.
Examining our diagram, it shows that, given the right candidate, voters on the radical spectrum of the pan-democrats are happy to back a Localist. This would mean either ‘Long Hair’ Leung Kwok-hung (梁國雄) or Ray Chan Chi-chuen (陳志全) would be threatened by a localist candidate.
Rolling polls – too soon to comment?
It is still a bit too early to comment on the HKUPOP rolling-polls, because we are still at the initial stages and initial polls always favor incumbents. Yet Long Hair, the socialist veteran, has seen his support drop by half (7%) compared to the same period in the last election and Ray Chan (6%) is not looking like a safe bet.
Looking to the future of this election, there is currently a high level of volatility in the polls as voters are still not familiar with some of the candidates. But after the televised debate, we expect Alvin Yeung’s poll scores to drop to around 15% and the polls will stabilize.
Based on previous observations, elections tend to equilibrate at approximately 15 days before the election date. From that point onwards, candidates with ~6.5% score in the polls are likely to be elected, as there will be a late swing of 2-3% for them.
Given current polling results and having prior knowledge of local strongholds for certain candidates, the two tickets of DAB, Alvin Yeung, Cheng Ka-fu (鄭家富), Leung Kwok-hung, Gary Fan Kwok-wai (范國威) and Christine Fong Kwok-shan (方國珊) are likely to be elected. The last two seats are likely to fall to a moderate Democrat (Fernando Cheung Chiu-hung; 張超雄) and a moderate pro-Beijing candidate (FTU).
But of course, it only took an evening in Mong Kok to change everything in the last election.
In politics, it is always too soon to comment.
[starbox desc=”Yu-xi Chau
Chau is a data analyst and political commentator. He has a PhD in Complexity Science obtained from the University of Warwick and co-founded the Hong Kong Public Affairs and Social Service Society there. He also writes on social media platform Medium Raw.