Predicting the election result from the polls

1543 BDT Oct 10: Thoroughly updated for the latest PA poll.

“Had it not been for the protests, now we would all be focusing on next year’s elections and looking at the government’s record in office and the opposition’s pledges,” said Zafar Sobhan, editor of the Dhaka Tribune, an English daily. “Now, all bets are off and elections seem a distant concern. It is hard to see how things will revert to politics as usual after this.”

That’s from Syed Zain-Al-Mahmood’s excellent Guardian report on Shahbag.  That was February.  Now it’s September October.  The protesters are long gone.  And everyone’s focusing on the elections —when will they happen, how they will happen, will they really happen, who will win if they do happen, how big the margin will be.

Zafar was hardly the only one who thought that way about politics as usual.  Across the ideological and political spectrum, there was a general agreement that politics-as-usual would end in the spring and summer of 2013 —the debate really was about what would replace it.  Well, in the autumn of 2013, politics-as-usual is back with vengeance.  And this post is all about politics-as-usual.

I have nothing to say about the when and how or whether of the coming election.  Instead, let me focus on what the polls imply about the results of a hypothetical election held this winter.  Some simple calculations – details over the fold – suggest that while such an election will likely result in an unprecedented comfortable BNP landslide.  victory, AL is still very much in the race.

We have the following publicly available opinion polls:

– published by Prothom Alo in January 2010, January 2011, January 2012, and January 2013, and October 2013;

– published by Daily Star in August 2010, January 2011, January 2012, and January 2013; and

– David Bergman’s analysis of AC Nielsen polls.

Each poll gives one data point for each of the major party’s vote share in a hypothetical election.  This means, we have 143 data points since the last election.  In the chart below, these points are plotted.

 

Of course, each poll is an imperfect guide to the overall political scene.  Trend lines aggregate all that information and allow us to see, well, the trend in political landscape.  And extrapolating the trend line gives us the projected hypothetical election result.

But how to estimate the lines?  While estimating trend lines, one needs to keep two things in mind: does the trend reflect the qualitative history; and how much weight should be given to the latest data.  The first concern rules out a linear trend line.  But non-linear trends put more emphasis on the latest data, and one needs to be careful that this does not drive the projection.

Ultimately, the choice of a trend estimate is quite arbitrary.  And the conclusion of this post is dependent on that arbitrary choice.  So take the whole thing with a fistful of salt.

How did I estimate the trend line?  I asked excel to fit a cubic function through the dots, starting as close as possible from each party’s respective actual vote shares in December 2008.  This, incidentally, has one side effect: it possibly overstates AL’s end point, and underplays BNP — that isn’t perhaps a bad thing given the existing set of election rules, which are tilted to the incumbent’s favour.

And what do the lines show?

Let’s start with the red line, for the AL.  According to the trend line, AL retained its 49% support from December 2008, if not improved on it, throughout 2009.  Does this seem sensible?

Let us think back about 2009.  The government commanded widespread media support.  Its young, fresh-faced ministers were considered to be clean and had good image.  Media gave it a free pass on BDR mutiny.  And there was a massive sympathy wave for the Sheikh family as the executions of August 1975 killers loomed.

Then we have a period of steady decline in AL’s popularity, until reaching a trough around 32% by the end of 2012.  What happened at this period?  Share market scam.  Padma bridge scam.  Several law and order mishaps.  High inflation.  Vendetta against Prof Yunus and Mrs Zia.  Anyone reading the newspapers could come up with plenty of reasons for the dropping popularity.

Interestingly, the red line shows signs of picking up since the beginning of 2013.  Why?  We will come back to this — for now, note that the pick up, should it continue, will mean AL getting 396% of votes in a December 2013 election.

Let’s now turn to the green line, for BNP.  Well into 2010, it went backwards from its 33% vote share from December 2008, with popularity falling far below the 30% mark.  Then the turnaround started from late 2010.  Again, this seems to tally well with what actually happened in 2009 and 2010.  In 2009, BNP was completely missing in action from the public debate about, well, everything.  The party was still recovering from 1/11 and the election defeat.  By late 2010, it found its feet.  Plus, the start of its recovery seems to coincide with the government’s assault on Ziaur Rahman’s residence.  Coincidence?  Or public’s innate sense of justice?

The turnaround notwithstanding, BNP seems to have been behind AL throughout 2011.  And it appears that AL was losing popularity more quickly, rather than BNP becoming popular rapidly.  Again, both seem to gel well with the conventional wisdom of that year — see this Asif Nazrul op ed for example.

But its popularity has been rising steadily, and BNP seems to be on track to be around 3947% support by the end of 2013 — should this come to pass, it will be BNP’s best ever performance.somewhat less than its 41% share in 2001, but far stronger than what it used to win in the 1990s.

A straightforward reading of the trend is that BNP has not only unified its core supporters, it has also attracted a plurality of the ‘middle’, while the AL has lost not only a lot of the people who swung its way in 2008 (49%), but its 2001 base has returned to the fold.  also many who came to the party in 2001 (40%).

This all seems fairly straightforward and plausible to me.  AL’s manifold mishaps are self-evident.  They explain why AL’s support dropped below 35%.  And BNP gradually recovering to somewhere close to 40% is also quite understandable.

What about Shahbag and all that?

David Bergman and Syeed Ahamed have debated about the electoral fall out of Shahbag.  It seems to me that whatever Shahbag’s impact might have been, BNP was already on the come back trail before this February.  But what explains the ‘middle’ swinging to BNP?  Could David be right about BNP winning the ‘ideological war’?  

Or could the uptick in AL’s support in recent months have anything to do with Shahbag and aftermath?  Perhaps the death sentences and prospective hanging, or the spectre of another BNP government, has brought some folks back to the Awami fold?

Perhaps.  But if so, it won’t be enough to avoid a defeat.

As of summer October 2013, the trend lines suggest that BNP was expected to get 4538% votes against the AL’s 364%.  Extrapolating this to December, and the numbers are 3947% BNP vs 396% AL.

Of course, given our first past the post voting system, with 3947% votes, BNP will likely win many more than 3947% of seats.  How many more?  Looking at the two three elections where BNP defeated AL in a landslide may provide some clue.

In 1979, BNP won 207 seats with 44% of votes —a seat to vote percentage ratio of 4.7.  In 1991, it won 140 seats with 30.8% vote —a ratio of 4.5.  In 2001, 41.4% vote gave it 193 seats —again seat to vote percentage ratio of 4.7.  Applying these ratios to 3947%vote would imply more than 220 something around 180 seats —its best ever showa comfortable majority.

What about AL?  In 1979, it won 39 seats with 24.5% votes — a ratio of 1.6.  In 1991, it captured 92 seats with 30.2% vote —a ratio of 3.  In 2001, it got 62 seats with 40% votes —ratio of 1.5.  Applying these to 39% vote suggest around 60 seats.

Why is AL’s seat-to-vote ratio so disparate?  Essentially, both parties have geographical pockets of support.  Outside these areas, in both 1979 and 2001, BNP was uniformly more popular than AL.  However, in 1991, AL was more competitive in many areas outside of its base districts.  Why this was the case, I don’t know.

The polls don’t give us any clue about whether the 1991 or 1979/2001 are likely to be repeated for AL.  If AL’s support is like that of 1991, it may well end up with 110 seats.  This would, however, imply that smaller parties will be virtually non-existent.  On the other hand, should the other two elections prove to better guides for AL’s fortune, then the party seems to heading for around 55-60 seats.  In this scenario, smaller parties will receive as many as 60 or so seats.

That is, based on the opinion polls, we are looking at BNP getting around 180 220 seats, against AL’s maximum likely tally of 1160.

How confident am I about these numbers?  Not a lot.  There are a lot of caveats, including: alliance factors, voter turnout, new voters, and geographic differences.  We will cover these factors in a follow up post sometime soon.