Sungai Limau #GE13 Analysis

During the 13th General Election (GE13) we came up with a methodology for estimating individual voter support based on polling lane results from the 12th General Election (GE12). The results of these calculations were used to design a program to predict election results based on ‘swing’ parameters fed in e.g. ethnic support; age group; increasing partisanship etc. You can read a summary of this method at the end of this post.

Using the polling lane results of the 13th General Election we compared our previous calculations and come up with an estimated swing. This can help inform analysts and campaigners. For the upcoming by-election in Sungai Limau we found that:

1. Support for PR has decreased by 6.5% since GE12, from 63% to 57%
2. Support decreased from voters aged 21 – 64, particularly ages 25 – 50
3. PR is likely to retain the seat this by-election

Seat Statistics

Sungai Limau is a DUN (State Assembly seat) located in a rural parliamentary seat, Jerai, located in Kedah.

Based on our GE13 data there were 27,287 voters of which 1 is a postal voter and the rest are regular voters. 93.03% of the voters are ethnic Malays while 6.75% are ethnic Chinese. The chart below illustrates the breakdown of voters by age group. Youth (ages 21-30) make-up 20.45% of the regular voters. 48.22% of the electorate are men. As we do not have the electoral roll being used for this by-election, expect some differences in our statistics compared to the current roll.

As our previous calculations were based on Parliament seat results, we will use the same seat (P012.Jerai) for this DUN. Voting patterns for DUN and Parliament were similar for Sungai Limau in GE13.

Predicted GE13 Support (based on GE12 results)

Leaning BN: 7,969 (29.20%)

Fence-sitters: 2,053 (7.5%)

Leaning PR: 17,265 (63.27%)

Estimated GE13 Support (based on GE13 results)

Leaning BN: 9,018 (33.04%)

Fence-sitters: 2,792 (10.23%)

Leaning PR: 15,477 (56.72%)

‘Leaning BN’ means these voters have a >50% probability of voting for BN. In GE13 BN obtained 10,520 votes from the 27,287 voters, close to the combined total of ‘Leaning BN’ and ‘Fence-sitters’ group.

By comparing the 2 datasets we can estimate the swing in support by the voters. For this analysis we will focus on the swing to and away from PR.

The following histograms illustrate the distribution of probabilities among voters. The blue bars represent ‘Leaning BN’ voters. The 50% bar represents the fence-sitters. The red bars represent fence-sitters and ‘Leaning PR’ voters – we combine the 2 to show PR’s potential base.

While the majority of support for PR is still above 50%, the number of fence-sitters and ‘Leaning BN’ voters has increased. Support for PR has moved closer to the middle. The average support for PR for each voter has decreased by an average of -2.86%. For women the decrease was -2.83%, while for men the decrease was -2.89%.

When we group the voters by their est.swing, there is a clear pattern of movement away from PR. This histogram illustrates the number of voters by estimated swing to/from PR. 19,202 voters (70.38%) experienced a drop in support. However this decrease was too small to affect the outcome, as the average support from voters was 57%. The majority of the decrease was between 1% – 7%, meaning that most supporters either became fence-sitters, or remained supporters.

This chart shows the swing by age. Support for PR decreased from ages 21 – 64, particularly ages 25 – 50. Ages 65 and above had an average swing of +0.31%.

When we combine this graph with the number of regular voters by age, we can see the impact of the swing.

This chart illustrates the estimated level of support from voters. Despite increasing their share of support to 33% of the electorate, BN faces a difficult challenge to win the seat.

Polling District Swing

The following is a list of polling districts (daerah mengundi) with their average estimated swing to PR and estimated support for PR based on GE13 results. The list is sorted by support levels and only covers averages of regular voters.

1. Bukit Choras (83.03% support, -6.95% swing)
2. Bukit Besar (77.58% support, -4.62% swing)
3. Gelam Tiga (66.44% support, -3.57% swing)
4. Sungai Dedap (61.55% support, -4.79% swing)
5. Batu Enam Belas (58.93% support, -2.46% swing)
6. Kampung Titi Batu (58.63% support, +2.07% swing)
7. Ulu Sedaka (56.30% support, -3.96% swing)
8. Simpang Tiga Luar (56.13% support, -4.56% swing)
9. Matang Buloh (55.67% support, -5.30% swing)
10. Kabu Sepuloh (51.11% support, -0.58% swing)
11. Sedaka (50.61% support, -0.53% swing)
12. Padang Lumat (49.02% support, -4.35% swing)
13. Sungai Limau (48.99% support, +0.09% swing)
14. Simpang Tiga (46.33% support, +6.33% swing)
15. Sungai Daun (44.86% support, -4.24% swing)
16. Selengkoh (40.79% support, -6.20% swing)
17. Dulang Besar (37.21% support, -3.47% swing)
18. Dulang Kechil (30.89% support, -5.75% swing)
19. Sungai Kering (20.89% support, +1.77% swing)

Summary of Methodology

Voters are assigned to polling lanes based on their age and polling district (the area they reside in). A lane can hold 350-700+ voters, with a common size of 400-650 voters. From the results of the lane, we can assign a probability value to each voter in that lane.

For example if 400/600 votes went to BN, then the average probability of voters in that lane voting for BN is 66.67%. We can take the analysis a step further by adjusting for ethnicity, because each lane will have a fixed maximum number of ethnic voters. If there are 500 Malay voters in our example that means BN obtained at most 400 Malay votes and at least 300 Malay votes. This changes the probability of Malays voting for BN to a value between (300/500) to (400/500), or 60% – 80%. We take the average of this, giving a probability of 70%.

Note that this calculation does assume that turnout is evenly distributed by race. While that is not true, high voter turnout rates and ethnically dominant lanes reduce the impact of error. We tested both the average probability and the ethnicity-adjusted probabilities in our GE13 simulations and the ethnicity-adjusted probabilities yielded greater accuracy. We also had to take into account that voters migrate between constituencies. When that happens they bring their individual probabilities with them. This helps avoid inaccuracies that are introduced by focusing on polling district results from the previous election.