Electoral forecasting for 2012 catalan elections: what can go wrong.

Author: Xavier Fernández-i-Marín
November 25, 2012 - 3 minutes
Catalonia Electoral forecast

So finally election day has come. Let me stress again the historical importance of today’s elections, which can be noticed by the interest of international press (The New York Times, The Guardian, Le Monde, The Wall Street Journal, The Times, to name but a few).

The electoral forecasting, updated to the last moment predicts the following percentages of vote share:

<img title="gr-prediccio_avui-caterpillar.png” alt="gr-prediccio_avui-caterpillar.png” src="http://blog.xavier-fim.net/wp-content/uploads/2012/11/gr-prediccio_avui-caterpillar1.png"class="aligncenter” />

It differs from the last prediction in two ways. First, it has more uncertainty, because there have not been polls in the last week. And second, it shows the effect of the tendency of the campaign as computed using the days -20 to -10 to the voting day.

But many things can go wrong. Although the model did quite well two years ago for the same elections and a year ago for spanish elections (see one of the first posts of the blog: ), I have some doubts of its performance for today.

First, the volatility of the supports can be higher this time, due to the repositioning of parties in the space and the fact that the agenda has been almost centered in the referendum for becoming a new European state than on the budget constrains that have been addressed by lower social services.

Second, the percentage of citizens that didn’t know who to vote has been up to a third of them a couple of weeks before the election. Polling houses have had a hard time guessing the distribution of this uncertainty, and again, it does not seem very likely that learning from the past may help too much this time.

Finally, what scares me most is the translation of vote shares into seats. Let me emphasize again that the prediction of shares is based in two strong predictions: equal participation and equal distribution of vote by parties between districts. The evidence right now, coming from postal voting up so far, indicates that participation will increase, which may have an important effect in some districts with regards to the entry of small parties. And as for the distribution of vote by parties, I’m not absolutely sure that such a strong restriction can work, but it is what we have right now.

<img title="gr-prediccio_avui-escons.png” alt="gr-prediccio_avui-escons.png” src="http://blog.xavier-fim.net/wp-content/uploads/2012/11/gr-prediccio_avui-escons.png"class="aligncenter” />

I have to admit, however, that the translation into seats gives a total of 134 (from a total of 135 seats) when the highest probability number of seats by party is taken, which is a relief given the way that the translation has been done.

But I’m sure that many, many more things can be different and ruin the electoral forecasting. I will talk about it tomorrow.

Bavarian state elections 2018 - Evaluation of the forecasts

Author: Xavier Fernández-i-Marín
October 16, 2018 - 3 minutes
Comparison of the electoral results and forecasts for Bavarian elections 2018
Electoral forecast Bayesian Data visualization

Bavarian state elections 2018 - Forecasting

Author: Xavier Fernández-i-Marín
October 13, 2018 - 9 minutes
Electoral forecast for Bavarian elections 2018
Electoral forecast Bayesian Data visualization ggmcmc

Predicció probabilística pel resultat de la consulta d'independència del 2014 a Catalunya

Author: Xavier Fernández-i-Marín
January 13, 2013 - 7 minutes
Electoral forecast
comments powered by Disqus