lundi 19 mars 2018

Russia: The day after


Finally, how did the polls fare in the Russian presidential election? First, let us revisit the graphs that I presented on Friday, based on all the polls conducted since the end of December. On average, the polls estimated Putin's support one week before the election -- there was a ban on publication of the polls during the last week -- slightly under 80%. The results published by the electoral Commission according to the newspaper Le Monde show that Putin received around 76.6% of the vote. Therefore,  the polls estimates are within the margin of error. The support of the two main opponents, Grudinin from the Communist Party, and Zhirinovski, an ultranationalist candidate, was generally estimated at around 6%. Zhirinovski's support was well estimated: he received 5.6% of the vote. However, Grudinin finished second, with 11.8% of the vote. He managed to get twice the support he was estimated to have.

I do not have the sample size for these surveys but if they had 1000 respondents, the margin of error for Putin would have been 2.5% and for Grudinin or Zhirinovski, around 1.5%. Therefore, we may conclude that the only support that was not estimated accurately is Grudinin's. It reminds of the difficulties pollster had to estimate the support for the Communist Party -- and extreme left in general -- in some European countries during the Cold War, a difficulty that has since been replaced by a similar difficulty to estimate the extreme-right. Only one poll had put Grudinin at more than 10%, the one conducted by CIPKR at the end of February.

The support for all the other candidates, including Xenia Sobchak, was generally estimated  to be under 2%. Sobchak had two higher estimates, one at 2.3% by WCIOM, and one at 4.3% by CIPKR. She got 1.5%. Navalny's support was estimated at 1% at the end of December, just before he was banned from running. Of course, we do not know how he would have fared if he had run, but from the beginning, in this election, there were  two "main opponents" to Putin according to the polls, i.e., Grudinin and Zhirinovski.

What about pollsters?

The following graph compares the estimates from the different pollsters. It shows that, while FOM estimated Putin's support at around 75%, WCIOM tended to slightly overestimate the support, putting it at more than 80%.

However, in the exit polls, WCIOM had a slightly lower estimate of Putin's support, at 73.9% compared to 76.3% for FOM (see below the estimate from WCIOM). Both estimates were within the margin of error or close. And the estimates for Grudinin and Zhirinovski are quite accurate. It is rather impressive given the difficulty to conduct exit polls with a population scattered over such a large territory. In Canada, with a similar scattered population over a large territory, pollsters do not conduct exit polls.

Finally, the turnout was 67.5%. It is quite close to the estimates that I saw from WCIOM, i.e., 70% of respondents reporting that they would go vote for sure. As elsewhere, there is a tendency for respondents to over report their intention to vote or their vote -- or else, people who do not vote tend not to answer polls.


Finally, the pollsters fared rather well in their estimation of the vote, except for Grudinin, whom they clearly underestimated. However, since the polls could not be published during the last week, one could speculate that there was some movement during that week. Such an eventuality is unlikely however since there was no huge movement during the whole campaign.The exit polls were very good. However the pollsters will have to try to understand how come they underestimated voting intention for the Grudinin.

vendredi 16 mars 2018

Russia: The Day before


Although the results of the Russian presidential election are quite known in advance, it is  nevertheless interesting to see what the polls tell and how they will fare.

I took all the results of all the polls referred to on Wikipedia since the end of December 2017. I added more recent polls that I found on the pollsters' web sites. There is an embargo on the publication of polls in Russia during the last week of the Campaign. The last poll I found was conducted by FOM and ended on March 11.

There are two "serious pollsters" who conducted polls almost every week, i.e. WCIOM and FOM. In addition two polls were conducted by the "Center of research on political culture of Russia" (CIPKR) and three polls were conducted by people around Navalny. These polls seem to have been conducted by Navalny's partisans and it is not sure at all that the methodology is scientific. However, since the results of these polls are not different from the results of other polls, there is no serious justification not to keep them in.The statistical procedure that I use is the same as the one I used in many recent elections, i.e., local regression (loess).

The first graph shows voting intentions for the main candidates, including non-disclosers, i.e., respondents who say that they will abstain, that they will void their ballot or that they do not know whom they will vote for. It shows that voting intentions are quite stable with Putin always close to 70%. The two other most important candidates, Grudinin and Zhirinovski get less than 5% and the rest of the candidates together get also less than 5%. The non-disclosers are also quite stable, at a level somewhat lower than 20%.

However, since the proportion of non-disclosers varies with pollsters and with polls, it is necessary to exclude them in order to examine the proportion of support for each candidate after excluding the non-disclosers. In this way the total support for all the candidates is 100% for all polls. The next figure shows the support for the same candidates excluding the non-disclosers. It shows Putin's support around 80%, perhaps decreasing a bit in the last weeks. The support for Grudinin and Zhirinovski is stable at 5% as is the support for the total of the other candidates.

Finally, one interesting question is whether the pollsters give the same estimation of the support for Putin. The following figure presents the estimates from WCIOM, FOM, CIPKR and FBK. There are trend lines only for WCIOM and FOM because they are the only two who conducted enough polls. In the following graph, the blue line is the trend line from the WCIOM polls and the green line, from the FOM polls, We also see two violet larger dots illustrating the CIPKR estimates, and three larger red dots illustrating the FBK (Navalny's group) estimates. The figure shows that the CIPKR tended to put Putin lower than the other pollsters and that FOM estimates that support for Putin is decreasing (from 83% in December to 75% on March 11). On the contrary, WCIOM estimates that Putin's support is stable at around 82%. In the end, there is a difference of 6-7 points between the estimates.


It will be interesting to see whether the Russian pollsters, particularly WCIOM and FOM managed to predict the support for the different candidates well. It is problematic that there is a ban on the publication of polls during the last week since we will not be able to estimate the reliability of polls and we cannot know whether the trend estimated by FOM went on during the last days or else, if support for Putin is stable, as estimated by WCIOM.

Unfortunately, I do not read Russian well enough (I read the numbers, at least 😊 and I get help from google translate) to examine the possible differences in the polls' methodology. I am open to receiving more information, in French or in English, on this question.

The main question on Sunday will be the turnout. In the WCIOM polls, about 70% say that they will go vote for sure. These figures are close to what we get in many established democracies. However, research shows that turnout is higher when elections are close races, which is not the case here. It will be a challenge to mobilize voters.