Sunday, May 8, 2011

Assessing the Generational Shift Effect in GE 2011

This is a technical post, so it will be rather dry. Going by previous pageviews, I doubt many people will be interested to read further, but nonetheless, I will need to perform the analysis and archive the results here for my own records.

The Generational Shift Effect (GSE) was previously defined here. The objective of this post is to assess whether such an effect is in fact present.

Given the differences between each election in terms of candidates, issues and media use, it is a futile exercise to rigorously reduce the results of each election to individually quantifiable covariates. My approach here is to simply look at the data and to check for interesting trends.

To recap, the constituencies that I identified as possibly manifesting the GSE were Radin Mas, Pioneer, Moulmein-Kallang, Sembawang, Sengkang West and Punggol East. This group consists of 4 SMCs and 2 GRCs.

My approach is to compare the PAP's vote share in each constituency in GE 2011 and to compare it against the PAP's vote share in the most recent previous election. However, except for Sembawang (which in fact has had its boundaries redrawn), none of the constituencies in the group above existed in recent history prior to GE 2011. This may seem like a bad thing, but it does have a silver lining.

While previous data may not be directly comparable to current data due to redrawn GRC boundaries, every new constituency in 2011 has a progenitor GRC. We can compare the PAP's vote share for say, Sengkang West SMC in 2011 versus Ang Mo Kio GRC (its progenitor) in 2006. We can also compare the PAP's vote share for Sengkang West SMC in 2011 against Ang Mo Kio GRC in 2011. So, in effect, we have two data points for comparison.

What we are really interested in are differences. Using our example above, we can calculate how well the PAP fared in Sengkang West in 2011 versus 2006 (using Ang Mo Kio as a proxy), taking the difference [1st difference]. We can also calculate how well the PAP fared in Ang Mo Kio between 2011 and 2006, again taking the difference [2nd difference]. It is the divergence between these two differences that may be attributed to the GSE (and other factors).

Now, let's turn to the data. I have excluded Radin Mas SMC from this analysis. Radin Mas was carved out from Tanjong Pagar, which has had walkovers for the longest time, so historical voting data is unavailable. In addition, the 'Lee Kuan Yew effect' in Tanjong Pagar would probably swamp any other effect, so I'm not even going to bother with further analysis here.

Table 1 below shows PAP vote shares for the constituencies of interest in GE 2011 as well as the most recent election. Except for Sembawang GRC, data for the previous election is taken from the progenitor GRC. Also, all previous data relates to GE 2006, except for West Coast GRC, which was last contested in GE 1997.


  Table 1. Comparing PAP vote shares for each constituency across elections. Larger image here.  
  
From Table 1, it is clear that the PAP's vote share has declined for all the constituencies under study. But how does this compare to the progenitor GRCs?


Table 2 below reproduces the rightmost column from Table 1, and includes another column showing the change in the PAP's vote share in the progenitor GRCs. Note that Moulmein-Kallang progenitor data is unavailable since Moulmein-Kallang was carved out of Tanjong Pagar and Jalan Besar. Tanjong Pagar was uncontested and Jalan Besar no longer exists. In addition, I have included a column to show the change in the PAP's average vote share across elections.

Table 2. Comparing changes in PAP vote shares. Larger image here.


In every constituency under study (except Sembawang and Moulmein-Kallang), the drop in the PAP's vote share has exceeded the drop in the PAP's vote share in the progenitor GRC. It has also exceeded the drop in the average PAP vote share across elections. Most striking is the divergence seen in Sengkang West, which experienced a drop in PAP vote share by 8% while in contrast, its progenitor GRC of Ang Mo Kio actually improved in its PAP vote share by 3.2%.


In the case of Sengkang West, the large divergence in PAP vote share between Sengkang West and its progenitor is unlikely to be due to the GSE. It is far more likely due to the fact that Ang Mo Kio is helmed by PM Lee Hsien Loong. Similarly, the divergence for Punggol East may be explained by the presence of heavyweight Teo Chee Hean in progenitor Pasir Ris-Punggol. On an opposite note, Moulmein-Kallang's poor performance for the PAP in 2011 may be more attributable to the presence of Yaacob Ibrahim (Orchard Road floods) and Lui Tuck Yew (hardly a ray of sunshine, see here).

For the remaining 2 constituencies, it is more difficult to reject the presence of the GSE. Khaw Boon Wan helms Sembawang GRC. Notwithstanding his $8 gaffe, he is still one of the PAP's more popular and stalwart politicians. Yet Sembawang fared poorly relative to the fall in the PAP's average vote share in this election (-12.8% vs -6.5%). This is an unexpected result which I interpret as the GSE exerting itself. For Pioneer SMC, I can't think of major reasons for the larger than expected fall in vote share except for the possibility that Pioneer is demographically much different from its progenitor, West Coast GRC. Again, I think the GSE could have exerted itself here.

Table 3 below summarizes the foregoing discussion in assessing the GSE. I have included a column showing which opposition parties contested in each constituency for GE 2011. Note that Sembawang and Pioneer, which I flagged as having a high likelihood of GSE, were contested by SDP and NSP respectively. The WP was the big opposition party winner in this election, presumably due to their branding, as surmised by Yawning Bread. So, we can exclude the "WP effect" from explaining the large falls in PAP vote share these two constituencies. In addition, neither SDP nor NSP fielded their 'A' teams or top candidates in these constituencies as well, so candidate differences should be muted.


Table 3. Likelihood of GSE and possible explanations. Larger image here.

5 comments:

eagle29 said...

Your analysis is very interesting. So much feedback on the fledgling popularity of the ruling party can be easily found everywhere, so I still wonder why the leaders keep on repeating that they have to analyse what when wrong for them in the GE.

mjuse said...

Whether or not the PAP feels comfortable with their victory margin, they *have* to declare that they will analyze how to make things better for Singaporeans and themselves.

To not do so would be to invite criticism. Whether they actually mend their ways, however, remains to be seen.

And the truth is, the PAP victory margins were sub 10% in many constituencies. Even though the PAP was returned to power, I would not be surprised if they felt antsy over their diminished vote share.

The GRC system is such that massive losses can be sustained over a single election. Vote shares move linearly, but change of seats is a non-linear phenomenon.

Ray said...

"This is a technical post, so it will be rather dry.", I ONLY read your technical posts. Less sugar more crunch!

Anyway, some questions.

1) Where are the election boundary sources for 2006?

2) I don't quite get the way you define GRC/SMC progenitors. The way I would do it would be to take the GE 2011 map and overlay the GE 2006 map. Lets say we are interested in Ang Mo Kio in 2011. We would look at all of the GRC/SMCs which cover Ang Mo Kio in 2006 and take the as progenitors of Ang Mo Kio. But this would give Ang Mo Kio multiple progenitors which doesn't seem to apply in your analysis.

3) Have u tried different generational partitions? Instead of 29 say 34 or 39 (according to census data).

4) Since you are comparing distributions of possible very difference sizes did u control for large changes in variance in the data?

Anyhow, still AWESOME!

mjuse said...

To Ray:

1)Detailed information on electoral boundaries is difficult to find. Most maps are large scale and the boundaries are not resolved to the street level. I have not been successful in finding good information in this regard.

2)You need to read my earlier post on the GSE to understand how I arrived at the progenitor GRCs. Census data is based on the 2008 URA Masterplan, while election data is based on the electoral map. I would *love* to overlay the maps, but seeing as how I lack the requisite technical skills, I'm reduced to eyeballing.

3)Too much work! (I'm lazy). Besides which, the group aged 29 onwards were already able to vote in 2001 and 2006 (assuming they weren't in a walkover constituency). I was most interested in the 20-29 age group since they are voting for the very first time.

4)Nope. Frankly, I'm not sure how I would do this given the poor resolution in the data. Apply a variance stabilizing transform? Some other statistical tricks? The problem is that the data is already at a poor level of resolution, so additional rigor might simply lead to spurious accuracy. I have always stated upfront that my goal is simply to explore the possibility of interesting trends in the data, rather than arrive at hard, scientifically backed conclusions.

Lastly, there was an article in the paper today where someone spoke of using multiple regression. Interesting results, however, correlation does not necessarily imply causation. And further more, even if causation can be definitively shown, cause and effect still need to be distinguished from each other. Not a trivial task given the poor quality of data available.

Ray said...

1) Not unexpected, considering who's in charge.

2) I did! The overlay stuff was background for the point about multiple progenitors. I still don't get how you only have one progenitor per GRC/SMC.

3) Sadly, life intrudes.

4) Was thinking more about t test for different means in hypothesis testing but I realized that we are dealing with the population mean (compulsory voting) not the sample mean. So direct comparison makes sense and my question doesn't.

Anyhow, whatever method used to analyze the data so that policy can be formulated to fix the opposition (or counter the incumbent) is good enough. No need for scientific rigor.