The Moorish Wanderer

Predictions for the 2016 Elections, Part.2

So a uniform swing across all 84 marginal constituencies does not change the picture that much: of the 305 seats, the ranking of the 8 largest parties that concentrate 94% of parliamentary, local ballot selected seats did not change but for two: Because it exhibits a larger than usual number of marginal seats, Istiqlal is bound to lose a net 6 seats, thus leaving RNI as second-in-command. PJD on the other hand, can expect to win a net additional 3 seats, still 65 seats shy of an absolute majority.

2011 and 2002 exhibit similar and almost coinciding density curves.

But this is a highly unlikely outcome, precisely because these swings are assumed to be uniformed: indeed, the discrepancies in voting results do point to heterogeneous outcomes, essentially due to the discrepancies in carried votes between the incumbent marginal and the potential challenger. Let me use an example to illustrate my point: the Speaker of the First Chamber, Karim Ghellab, represents a marginal – voted in with 4,789 votes, a long way behind PJD and UC, respectively 14,853 and 8,925 votes. The immediate competitor to Mr Ghellab is the MP list, with 2,735 votes. What is the probability, ceteris paribus, Mr Ghellab would lose his seat to the MP challenger in 2016?

Consider the party’s past electoral performance. Amazingly enough, Istiqlal’s vote distribution did not vary much if not at all between 2002 and 2011 (the density curves for 2002 and 2007 shows it on the graph). Therefore, it makes sense to assume Mr Ghellab’s margin can be computed in terms of probabilities. And this shows us he would keep his seat with a probability of 46.7%, with a vote tally fluctuating most likely between 3,976 and 5,841 votes, regardless of Mr Ghellab’s or the nationwide electoral performance, he needs to ensure a a marginal majority of 813 votes.

(incidentally, voter distribution per party or per seat exhibits very common forms, as various papers from academia[pdf] testify to that)

How about PJD’s six marginal constituencies? According to the averaged PJD past electoral performances since 2002, three of these are more likely to go over the edge to PJD’s local challenger, and one is a pure, statistical toss-up. This is moderate good news for PJD, because they are assured to keep a comfortable margin, provided 2016 Elections come with no exceptional events (the “Black Swan”) In fact, all three marginals have strictly positive likelihoods to swing against PJD incumbents, but three only of these lean toward PJD’s local challengers, and Laayun remains in a tie (a feat for maverick PJD considering the voting pattern down South)

Of the 6 marginals, PJD has a significant chance to lose three, and a fourth is nearly a toss-up

On the other hand, one could also look at the Head of Government’s own seat in Salé, and while it is by no means a marginal (M. Benkirane got 27,000 votes to a second with a little over 8,000) yet there is a positive (very low) likelihood of losing his seat: less than 3% – precisely because it would entail a huge swing, with a third of Salé voters going over to any challenger.

The Istiqlal party is a peculiar contender: though it ranks a distant second in electoral results, it has maintained its initial strength even as the party led a relatively unpopular government, and has been associated with all government coalitions at least since 1997. It is undoubtedly an establishment party. Yet for its formidable 60-members strong caucus, many of those have been elected on razor-thin margins in very competitive districts, and the Istiqlali challengers have not been numerous enough to make up for the endangered seats. All in all, Istiqlal could well expect a net loss of 6 seats.

The party with the most interesting marginal record is undoubtedly Istiqlal, first, because they have the largest number of marginals, and second, because their electoral performance per district did not change over the years (a remarkable stability given its Establishment status) the odds for and against voter swing in PI-held seats is the perfect case study for the proposed method. As one can see, the initial assumption of a uniform swing across all marginals was unrealistic. By that account, Istiqlal was to expect to net a loss of 6 seats. But now, and according to the computed probabilities, Istiqlal can expect – again, ceteris paribus, to lose only 4 seats, with one is a small district where lopsided swings are expected and observed, and four others with moderate loss expectations. Assouerd is a peculiar seat, because a 200 voter switch is a dead heat, 600 is a landslide (respectively 4.3% and 13% swing)

Aousserd is very competitive, the high likelihood of part swing is due to the small constituency to start with.

Can this set-up deliver reliable results for marginals as well as ‘safe’ districts? The Tangier by-election provides good case study (given the fact polls do not curry favour with our esteemed representatives[pdf]) the model tells PJD was strongly favoured to retain all its Tangier seats, since the probability of losing its majority is so small it is not even statistically significant (larger than 99%) but on the other hand, the model predicts it may lose 12,500 votes (but this is probably due to the fact that Tangier district is a PJD stronghold, and their maximum voter reserve has been reached) so even if it retains its majority almost certainly (and it did) it was expected to lose some votes. The actual results show PJD lost 16,200 votes – a figure statistically close (well within the margin of error) to the projected 12,500 synthetic potential loss.

Howe about the other competitors for the Tangier district representatives? According to their respective results, PAM was the marginal representative, with a potential risk of losing it to UC, with a probability of 17%. So actually, PAM’s marginal Tangier seat was relatively safe, with expected losses of no more than 1,730 votes – it turns out PAM lost only 300-odd votes. The actual loser in this Tangier recall election was RNI, even though it was not a marginal (yet exhibited a loss probability of 9.7%) yet it failed to carry any vote. The model predicted PAM could lose to UC, but since RNI did not get any votes, UC candidate list simply filled in.