Intrigues and mysteries of the elections in 223 precinct

Intrigues and mysteries of the elections in 223 precinct

The Central Election Commission of Ukraine has processed 100% of protocols of re-elections of people's deputies. Opposition is convinced that their victory was stolen, authorities claims that the elections were legitimate.

Team of Promedia Consulting decided to explore details of the situation around 223 electoral precinct applying statistical analysis.

 

Why the election results did not match with the data exit polls results?

The first thing that attracted attention of opposition politics is a serious difference between the voting results and the data obtained by exit polls.

According to exit poll results at 223 electoral precinct according to the exit poll the candidate of AUU "Svoboda" Yuri Levchenko had to win with 51.8% of the votes and the second place had to be taken by pro-government candidate Victor Pilipishin with 34.8% of votes. Election results shows 40.6% of support for the first candidate and only 44.9% of votes for the second one.

The first thing noteworthy fact that in all four precincts sociologists predict bigger support for opposition candidates and underestimated support for pro-government candidates.

If interviewed respondents of the exit poll were selected randomly, we can estimate a confidence interval. In other words, we can predict limits for election results with 95% accuracy. Below is a table for results of exit polls for main candidates. The table represents precinct numbers, names of candidates, exit poll results, 95% confidence interval and the significance of binomial tests (if the result is less than 0.05, we can assume that the election results and exit polls results differ significantly).

Precinct

Surname

Exit-poll results

Lower bound of the confidence interval

Higher bound of the confidence interval

Election results

Significance

132

Kornackyi A. A.

48.8%

46.6%

51.0%

45.31%

0.001368

132

Kruglov N. P.

43.6%

41.5%

45.8%

47.85%

0.0001097

194

Poplavskyi N. N.

47.3%

45.0%

50.0%

53.54%

1.16e-07

194

Bulateckyi N. I.

42.2%

40.0%

44.5%

35.87%

2.698e-08

197

Dacenko L. N.

72.0%

69.6%

74.3%

63.51%

3.974e-12

197

Chervonopiskii S. V.

15.6%

13.8%

17.5%

23.30%

1.906e-13

223

Levchenko Yu. V.

51.8%

49.5%

54.1%

44.89%

1.668e-09

223

Pylypyshyn V. P.

34.8%

32.6%

37.0%

40.58%

2.363e-07

As can be seen, there are no exit poll results, which predicted the election results with sufficient accuracy. However, we should not immediately blame sociologists. After all, if the result of the exit-poll were "corrected" intentionally, it would not make any sense to inflate the result of Datsenko, which in any case would win with a large margin, or overstate result of Bulatetskii, who is also clearly wouldn't win the election.

So we need to seek the truth in the process of polling voters. In particular, we should look at few points:

  1. The survey was not conducted on the special stages (for example, in Kiev prison Viktor Pilipishin got 1227 votes out of 1957).
  2. The survey was conducted in the villages until 17:00 to and in the cities until 18:30 . So there could be a major changes during the last 1.5 hour.
  3. Respondents of surveys were selected according to the principle of uniformity. But the final number of respondents is not proportional to turnout:
    • 1900 respondents for 73,549 turnout in 223 electoral precinct
    • 2100 respondents for 68,755 turnout in 132 electoral precinct
    • 1800 respondents for 67 132 turnout in 194 electoral precinct
    • 1500 respondents for 43,643 turnout in 197 electoral precinct

These results can be explained by few factors. Firstly, on some districts interviewers work was hindered by unknown people. Secondary, there survey showed a very high percentage of refusals to participate in the exit poll (about 30% of respondents refused to answer, there are only 10% of refusals were on the previous elections). Thirdly, there is an unusual spatial polarization of voters for the pro-government candidate. A more detailed picture can be revealed from exploratory analysis of election results.

 

Statistical analysis of elections for 223 precinct

To analyze the election results we have used the statistical programming language R with Hmisc and ggplot2 libraries. The districts voting results were obtained from the official website of the Central Election Commission. To account the location of district commissions we geocoded their addresses to latitude and longitude values. Special districts were excluded from the analysis.

A characteristic feature of any local re-elections is the reduced turnout. This happens for various reasons. Often it affected by a poor informing citizens about the election. To get a general idea of turnout change we plotted the density of 2012 and 2013 turnouts distributions.

Density plot of turnout on the 223 precinct on the Election of People's Deputies 2012-2013

Both distributions have left-skewed asymmetry, which indicate general high activity of citizens. Nevertheless, it is clear that in 2013 the asymmetry is clearly reduced. Besides you should pay attention to the right "tail" in the density plot of 2012 turnout. Such phenomena occurs when turnout is intentionally forced or results are "corrected" by electoral commissions. Such "tails" can be found in a number of Russian elections.

Nevertheless, we observed a rather interesting phenomenon of turnout territorial polarization. In 2012 the turnout can be represented as a random variable, but in 2013 we clearly observed turnout increase in the southern part of the precinct. The figure below shows scatter plots of territorial location of district election commissions. Dots sizes indicate the size of the district (the total number of voters), and the color represents turnout.

Scatter plot of locations of district election commissions with the district size and turnout in the 2012 electionsScatter plot of locations of district election commissions with the district size and turnout in the 2013 elections

More precise characterization can be given with correlation coefficients (measure of connection) between the coordinates of the electoral commissions and the percentage of voters. Correlation coefficients for elections 2012  were r = -0.097 between turnout and latitude (p = 0.39) and r = 0.200 (p = 0.07) between turnout and longitude. Coefficients for the 2013 are r = 0.353 (p = 0.0013) and r = -0.409 (p = 0.0001), respectively. Thus, we can state the growth of turnout from northwest to south-eastern parts of the precinct.

It becomes more evident if we build a similar plot of voter turnout change.

Turnout change of elections 2012-2013 on the 223 precinct

Plot shows the regular decrease of voters in the northern part of the precinct, while the southern part shows almost no drop in the number of voters. And often this law deals specifically with large districts.

And now, in the same coordinates it will be interesting to consider the increase of votes for each candidate in 2013 compared to a number of votes in 2012.

Gain of votes number for Viktor Pilipishin on the election 2012-2013Gain of votes number for Yurii Levchenko on the election 2012-2013

As you can see Victor Pilipishin headquarter did not pay attention to small districts and clearly engaged to bring the candidate voters in large districts in the electorate commissions during the election day. We can quantify results of such work using histograms of increase of the number of votes for both candidates.

Histogram of gain number of votes for Viktor Pilipishin on the elections 2012-2013Histogram of gain number of votes for Yurii Levchenko on the elections 2012-2013

According to the plot, you can see that Yuri Levchenko observed almost exclusively gain votes compared to the elections in 2012. Growth can be explained by participation only of a single opposition candidate. On the other hand, some votes of Yuri Levchenko were taken by his counterparts. The correlation between the percentage of votes that candidate and his opponent Andrey Levchenko got is 0.33 (p = 0.0028). In other words, in those areas where Yuri Levchenko received great support, the great support was also  received by Andrei Levchenko, indicating that the technology of counterparts were used efficiently.

Regression analysis also reveals that results of both main candidates shows a relationship between the percentage of received votes and turnout. But direction of this relationship is positive for Victor Pilipishin and negative for Yuri Levchenko. More clearly it is presented on the plot below.

Scatter plot of the distribution of votes for Victor Pilipishin depending on the turnout on the elections 2013Scatter plot of the distribution of votes for Yurii Levchenko depending on the turnout on the elections 2013

Some experts believe that the support for candidates on the whole precinct should be uniform. In other words, the closer the regression line on the plot to the horizontal line, the more honest elections were held. The line stretches up should indicate “attributing” votes and forcing turnout, and a line going down shows that the votes were stolen.

On the other hand, in order to accept the hypothesis of fraud it's necessary to ensure fidelity of at least three assumptions:

  1. Candidate Support should be uniform for the whole precinct.
  2. Agitation must operate with same efficiency in all precinct area (to perform Step 1).
  3. Voters who hadn't make a decision about their choice at the time of the last poll should decide on their support in accordance with the existing support distribution.

Positive relation between turnout and the percentage of candidate votes may be provided by the mobilization of voters and reinforced by agitations, while rivals show small activity. For example, there is a strong positive relationship between turnout and the percentage of votes for Irina Farion in 2012. Although she won election with margin 50% of voters (about 55,000 votes), so there was no any sense to make a fraud.

 

Instead of afterward

Ultimately, according to the research results it could be hypothesized that the Victor Pilipishin headquarters used administrative resources (involving state employees to vote) and also there was a statement of AUU "Svoboda" about bribe voters. It is understandable that people who were attracted to the vote that way could often refuse to participate in the exit poll. But in order to make the final conclusions it would be nice to have sufficient documentary base (acts of observers, court decisions, etc.). Otherwise we can only agree with the experts that the opposition is not thoroughly worked in the elections or south-eastern part of the precinct was very upset with destruction of Lenin memorial by people in Maidan.

In justification of the exit poll results we may also mention the effect of "spiral of silence" by E. Noelle-Neumann. When most of the Kiev actively supports Evromaidan, there is only a little number of people who are prepared to speak openly about the fact that they voted for the pro-government candidate.

Nevertheless, these results are likely pose new questions than answers old ones. And in this case it is clearly that political technologists should pay special attention to the study of technology of winning the elections on 223 precinct.