The Moorish Wanderer

By-elections: a Case Study

This is a great opportunity for me to apply some of the computations I have elaborated on when I started to lay out the 2016 electoral map.

Al Ahdath Newspaper mentioned 12 seats (from 10 districts) where up for a by-election by next week, and I cannot find all of these, as I have managed only 7, and 2 others I am not sure are indeed contested as well (both PJD). Or perhaps the 12 seats mentioned comprise also those in Tangier-Assilah and an additional seat in Marrakesh-Menara:

909/2012 PPS – Youssoufia

908/2012 PI – Sidi Kacem

907/2012 PA – Azilal-Demnate

906/2012 UC – Settat

905/2012 MP – Moulay Yacoub

888/2012 MDS – Chichaoua

878/2012 USFP – Inzegane Aït Melloul

858/2012 PJD – Hay Hassani

898/2012 PJD – Menara

As I have mentioned several times before, many parties have ‘marginal seats’, i.e seats with barely enough votes to cross the minimum electoral coefficient. And these are more likely than others to fall next election. And as it happens, a couple of these are in play: UC, PA, USFP and PPS could lose their seats with a few hundred votes. But, having a marginal doesn’t translate into a higher probability of losing the seat. Indeed, the following probabilities computed on the basis of each party’s past electoral performance show this is not systematic.

Union Constitutionelle member of parliament Abdellatif Mirdas  for Sidi Kacem, whose seat will be contested is a good example to illustrate my point. He has managed to get a little under 6,000 votes and was subsequently the bottom of the list, very close indeed to the 6% threshold (about 4,800 votes) and his local electoral performance was by any measure a very good one for his party. Yet for this by-election, his majority is very slim indeed. Assuming a turnout similar to 2012, he has an 85.58% chance of losing his seat; roughly the same likelihood of not making it to the 6% threshold. This is not a very informative probability because it is unconditional on the district itself; so there is a need to normalize this probability with local turnout, and how close the party candidate is to the 6% threshold. Adjusting for these elements, UC has therefore a little under 70% chance of losing the seat.

I would suggest this method is naïve Bayesian actually: the probability \mathbb{P}(V_t<V) denotes the probability of getting at most the same number of votes they have got in 2011. It is then finessed by taking into account local factors, and then conditioned (rather than just observed) on past electoral performance, local 6% threshold, turnout, etc: \mathbb{P}_j(V_t|V_{t-1})=\prod_{i=1}^{n}\mathbb{P}_{i,j}(N_{t,i}|\{N_{t-k,i}\}_{k=K}^{0})

The table below uses these computations for the three other parties, although the results for PA and MDS are not as statistically significant as I would like them to be, but it seems many of these seats are going to be very competitive, and many are very likely to change hands. I have also used the same method to compute the likelihood of the other parties whose seats are not marginals, with some tweaking, mainly by normalizing probability to present turnout and when applicable, threshold effect per seat.


Whatever the end result, there will be no big change in parliamentary caucuses – not the largest caucuses, any way. On the other hand, there is good evidence to suggest competitive by-elections in perhaps all but one district (MP- Moulay Yacoub) I will fire off a next blogpost offering some insight as to which parties enjoy the largest probabilities of carrying these seats.