Category: Current Affairs

In yesterday’s post on Malaysia’s 15th General Election, I showed that constituency-level ethnic structure is a very robust and consistent predictor of which coalition prevailed in elections on the peninsula. But there is a lot more at play than just ethnicity in explaining election results, and new data from Malaysia’s Department of Statistics allows us to be much more comprehensive in our analysis.

This post is a “data-dump” of additional exploratory analysis. The tl;dr version is that urbanization is another strong predictor of which coalition prevailed in peninsular constituencies, but accounting for urbanization mostly does not wipe out the predictive capacity of ethnic structure.

I want to focus on four variables in particular.

1. Constituency size (area in km²)
2. Constituency population density (log of population/km²).
3. Median income
4. Unemployment rate

Taking these into account will give us a sense of how much the relationship between ethnicity and election outcomes is plausibly due to things that happened to be correlated with ethnicity: rural and lower-income constituencies in peninsular Malaysia area also those which tend to be majority bumiputera. This is a major inferential challenge that I have written about on this blog and in this piece in the Journal of East Asian Studies (pdf).

What you see below are sixteen graphs. Let me explain. For each variable (see the four subtitles), we have the correlation among peninsular constituencies between that variable and each coalition’s vote share. We also have the predicted probability of victory, derived from a simple multinomial logistic regression where that variable is the only predictor.

The summary message from this graph is pretty simple. There is indeed a strong correlation between constituency size and constituency population density—each a reasonable measure of urbanization—and coalition vote shares. PN did better in rural constituencies with low population densities, and PH did better in urban constituencies with high population densities. Looking at income and unemployment, though, we find no relationship whatsoever.

Because these analyses are only looking at bivariate relationships between one predictor and each election outcome, though, they don’t allow us to compare the strength of various predictors. For that, I estimate a series of OLS regressions of the following form:

We are predicting each coalition’s vote share as a function of ethnicity, the four variables above, and state fixed effects .* Standard errors are clustered at the state level. Here is what I find.

	BN Vote Share	PN Vote Share	PH Vote Share
Bumiputera Population Share	0.27***	0.59***	-.86***
	(0.04)	(0.03)	(0.03)
Ln(Area)	-2.38	1.00	1.64
	(1.23)	(1.18)	(1.74)
Ln(Population/km2)	-5.09*	2.11	3.12*
	(1.17)	(1.20)	(0.61)
Unemployment Rate	1.06	-1.44	0.01
	(0.72)	(0.94)	(0.50)
Median Income	-0.00	-0.00	0.00*
	(0.00)	(0.00)	(0.00)
N	164	164	164

The results are very clear: even accounting for urbanization and economic factors, ethnic structure is a very strong and consistent predictor of each coalition’s vote share.

To see how these results compare, I’ve estimated a multinomial logistic regression with all the predictors listed above (and the state fixed effects), and plotted the predicted probability of each coalition winning across the deciles of each predictor.

Ethnic structure remains a very strong predictor of which coalition prevails in peninsular constituencies. But we also wee that PH does well in densely populated places. And interestingly, once we account for ethnic structure and population density, PH also seems to do better in larger districts. Score one for multiple regression.

As a final exercise, let us consider the possibility that the relationship between ethnic structure and electoral outcomes depends on how urban or rural a district is. We can test this possibility by dividing constituencies into quartiles by how densely populated they are, and allowing the relationship between ethnicity and the probability of winning to vary across the quartiles. To read the results below, note that the x-axis tells you the relationship between ethnic structure and vote share for each coalition: points to the right of the line at 0 mean that higher bumiputera population share is associated with a greater chance of victory for that coalition, the reverse for points to the left of 0.

In rural and semi-rural districts, PN’s likelihood of victory is positively related to bumiputera population share, and reverse for PH’s likelihood of victory: PH is less likely to win the higher the bumiputera population share. This is the same as what I found above. But if you look at urban districts, you find no relationship between ethnic structure and the probability of victory for any party: confidence intervals are very large and we cannot reject the null hypothesis of no relationship.

Summing up, here is what we have found:

1. Urbanization is a very good predictor of vote share and the probability of victory for each of the three coalitions in peninsular elections
2. Even accounting for urbanization, ethnic structure is still a remarkably strong predictor of election outcomes in the peninsula.
3. When we allow the relationship between ethnic structure and election outcomes to vary according to how urban that constituency is, we find that that relationship disappears in the most urban peninsular constituencies.

Now we wait to see who forms the government.

NOTES

* State fixed effects are important for accounting for historical patterns of party politics (Kelantan is a PAS stronghold, Johor is an UMNO stronghold, etc.).

Malaysians went to the polls two days ago in the country’s 15th General Elections, which for the first time pitted three coalitions against one another. As I described last week, this multipolar political system dramatically increases the complexity of party politics in what is already a highly complex political system. But by taking careful look at how coalitions nominate various parties—given the ethnic structure of Malaysian politics in peninsular Malaysia—you can see the continuing role of ethnicity in shaping political competition.

The preliminary results are in, and they are a devastating blow to the old guard in Malaysian politics. The United Malays National Organisation, long the most dominant party in Malaysian politics and one of the most durable authoritarian political parties in the world, is now only the fifth largest political party in the country.

Party	Coalition	Seats
PAS (Pan-Malaysian Islamic Party	Perikatan Nasional (PN)	44
DAP (Democratic Action Party)	Pakatan Harapan (PH)	40
PKR (People’s Justice Party)	Pakatan Harapan (PH)	31
Bersatu (Malaysian United Indigenous Party)	Perikatan Nasional (PN)	28
UMNO (United Malays National Organisation	Barisan Nasional (BN)	26

There is a lot more to say about these other parties, but the basic result is that the Pakatan Harapan coalition won the most seats, but not enough to form a majority. As of the time that I am writing this, there are rumors that Perikatan Nasional (dominated by an Islamic party and a Malay nationalist party) could form a government with Barisan Nasional (still dominated by UMNO). This outcome, if it comes to pass, would mean that Malaysia’s ruling government would be an Islamist-Malay nationalist coalition, and the opposition will be a multiethnic and multifaith coalition.

I must emphasize that that coalition, if it forms, requires the support of some parties in East Malaysia. This same is true for nearly any governing coalition. East Malaysian party politics is fascinating and complex, but by and large it does not follow the same logic of ethnic political competition found in peninsular Malaysia. That peninsula-style ethnic politics must coexist with East Malaysia-style party politics is a fundamental feature of Malaysian politics, and no account of Malaysian politics can ignore it.

Nonetheless, we can still look through the results on the peninsula to get a sense of how ethnicity structures Malaysian politics. Below, I plot some results based on preliminary data that I have scraped from The Star, analyzed using the data that Seb Dettman and I have pulled together. We can use these data to see how the three coalitions fared across peninsular Malaysia’s 164 parliamentary constituencies (there are 165 in total, but 1 election is delayed for a couple weeks due to the death of a candidate).

What I have done here is to compare the vote share for each coalition with the bumiputera population share within that constituency. Blue points are where the coalition in question lost; red points are where they won. In the bottom left corner, I have predicted the probability that each coalition wins based solely on its bumiputera population share.*

Viewed this way, the results of GE15 are not confusing: the deep structure of peninsular Malaysian politics jumps right out. There is one coalition that fares well in non-Malay majority constituencies. And the other two fare well in Malay-majority constituencies. The battleground is the mixed constituencies on the peninsula with 20-40% non-bumiputera populations. It is true that the BN and UMNO have had a lousy result, but their voters have evidently swung to PN. Put simply, a BN-PN coalition makes perfect sense as the latest coalitional manifestation of ethnic politics in Malaysia. That is why those rumors are swirling.

There is a lot more to be written about these results. But one point that I wish to emphasize is that a BN-PN coalition would indeed represent a lot of continuity in Malaysian politics, but the role of PAS as the largest party in PN is special and important. PAS is an Islamist party; a large majority of Malaysian Muslims are Malays (and all Malaysian Malays are Muslims). It has historically been opposed to UMNO, although not always. But as I have been hinting for about a decade now in various unpublished essays (see e.g. here [PDF]), what I call Malaysia’s “ethnic order” creates incentives for politicians to capture an alignment of Malay and Muslim interests. As I wrote in that piece,

as Malaysia’s ancillary institutions reinforce Islam’s constitutive role in Malayness, they provide an alternative basis for political mobilization

That is my interpretation of is happening in Malaysian politics right now. That said, the latest news is that BN is meeting with Pakatan Harapan tomorrow morning (Malaysia time). Watch this space for more.

NOTE

* These results come from a simple multinomial logistic regression where the outcome is the winning coalition, and bumiputera population share is the sole predictor. More sophisticated analyses with many more predictors produce the same qualitative results.