28.12.2012, 22:33:48

Elections to the State Duma. Photo: Zamir Usmanov / ITAR-TASS

Corrected to believe

Konstantin Sonin spoke about the analysis of fraud in the 2011 elections

At the end of December 2012, an article by NES scientists Ruben Enikolopov, Vasily Korovkin, Maria Petrova and Konstantin Sonin and Alexey Zakharov from the Higher School of Economics appeared in the journal Proceedings of the National Academy of Sciences, in which they addressed the topic of fraud in the State Duma elections in December 2011 in Moscow.

The researchers managed to conduct a large-scale experiment ( pdf ), which revealed two remarkable effects. First: the election results in polling stations with and without observers differed markedly. Second: this difference was not in favor of United Russia - scientists were even able to estimate the approximate scale of fraud in Moscow, which amounted to at least 10.8 percent. Lenta.ru asked one of the authors of the work, NES Vice-Rector Konstantin Sonin, about the world and domestic experience in studying electoral fraud, what methods the researchers used, how their results relate to previous studies and why the simplicity of fools is deceptive.

"Lenta.ru": Please tell us a little about the history of the issue.

Konstantin Sonin: The history of analyzing Russian falsification for me personally began in 2007. Then I read in the blog of Alexander Kireev (this is an electoral geographer) that, of course, there were falsifications, but they did not exceed one or two percent. And then, the day after the December elections, I wrote a post on my LiveJournal about a comparison of two Moscow polling stations - in one an observer was expelled (from Yabloko, it seems), and in the other - not. I looked at how these polling stations voted in 2003. This was not so easy to do because the plots were renamed and work had to be done to study which houses belonged to which plot. And it became clear that the results at these polling stations were the same in 2003, and in 2007, United Russia in one of them received almost twice as many votes as in 2003.

I then wrote that I don’t understand why everyone thinks that the falsifications are small. Of course, one example - only two areas, two time points and two geographical ones - does not give the right to assess the size of falsification in the country. But this, it seemed to me, should shake faith in the thesis that, they say, everyone is already for “United Russia”, what difference does it make how the votes are counted.

So I wrote that post, and then it turned out that it had a completely unexpected consequence: Sergei Shpilkin read it, and from this his study of electoral mathematics began, so to speak. He then did much more than all of us ( the authors of the original work - hereinafter, Lenta.ru's note ) to analyze data on the Russian elections.

Konstantin Sonin. Photo: Mitya Aleshkovsky / ITAR-TASS

And anything about historical, world experience? For example, in your work you refer to an analysis of elections in Ghana.

World political science began with the study of those countries that we now call developed democratic countries. Therefore, if you look at political science as a science, for many years it was a science that studied exclusively democracy. Everything in it was intended for democracy, and, accordingly, electoral fraud was studied not as some kind of special phenomenon, but as a pathology. They were mainly discussed by practitioners - say, an American senator won an election with the help of some kind of theft of votes, his opponent sued, there were lawsuits - this is the story that was studied.

For example, Pulitzer Prize winner Roberto Caro has a wonderful book, a biography of American President Lyndon Johnson. So, the second volume of his book is entirely devoted to the 1948 elections (about 600 pages), which Lyndon Johnson won with the help of minor fraud. Small in the sense that they amounted to some tenths - hundredths of a percent relative to the number of voters. That is, he stole several thousand votes out of several million. But there is exclusively a “clinical history of the disease”: how people steal votes, rewrite protocols, and how the courts then deal with this.

But that was the case until recently. About 15-20 years ago, political scientists became more interested in what was happening in less developed countries. And, of course, works appeared on the meaning of electoral fraud and their assessment. In particular, we refer only to works on assessment, but we also mention those works that talk about meaning, because sometimes this is a mysterious thing.

At what stage of development is this science now? Say, in the field of mathematical methods?

Mathematical methods are used everywhere there in one way or another. What does it mean at what stage? Falsification is always a dynamic thing, because each subsequent falsifier, of course, does not read the articles, but tries to somehow use the mistakes of the previous ones. We clearly saw this in the March elections in Moscow (I don’t know about other cities) - the methods of falsification there were different from those in December. In March, for example, people were registered as continuous cycle production workers and all that. This did not happen in December - that is, previous experience was taken into account. But analysis allows us to find out what was really happening in the country.

If you are asking whether these scientific results have any real consequences, then I would say that there are no such direct consequences. But, of course, there are political consequences of falsifications.

I was asking about something else, of course. In your post in which you mention your work, you write that this is a rather marginal topic.

Yes.

So I ask, are there already some postulates and theories in this science? Is it developing? And if so, how active? Or is everyone dealing with this topic in passing?

The statistical methods that are now used did not exist 15-20 years ago, that is, statistical analysis has become more powerful. We did not use these methods, but those people who assume some kind of distribution and show what the differences from it are are, in a sense, using international experience. That is, the more elections around the world we study, the more we know that whenever there is a non-normal distribution, it almost certainly indicates fraud. You see, economics (everything intersects here - economics and political science) is best compared to medicine. We have a greater understanding of how the disease occurs and we can treat it better. But this is not at all like, for example, appendicitis, which can be cut out in any district clinic; it is more complicated here.

It’s good that you remembered these methods, I wanted to ask about them too. After all, it is precisely this kind of work that has come, so to speak, into the focus of public attention. They were criticized quite strongly, they said that the distribution can be not only normal, but also lognormal?

On the one hand, there was no serious criticism. There was a lot of chatter, there were some words from opponents, but Churov and those people who are with him are completely illiterate, there are many empty talkers among them. It happens that a person formally has some kind of mathematical achievements or education, so he proves that there are two-humped distributions. On the one hand, as a mathematically precise statement, this is, of course, true: indeed, there are two-humped distributions in nature. On the other hand, the distribution in Moscow was normal ( A normal distribution has one hump - note from Lenta.ru ) in 2003, in 2007, and on December 4, 2011 it suddenly became abnormal, and by March 4, 2012 it became again normal. Any event is likely to happen. For me, the fact that this two-hump fell on December 4, and in all other years everything was as it should be, indicates falsification with such obviousness that there is, in fact, nothing to talk about here. I didn’t even try to argue with such windbags, although there were decent and honest people among them.

Distribution of votes for the United Russia party in polling stations without observers, with observers and in polling stations where observers did not record any violations. Illustration from the article Enikolopov et al., PNAS, 2012

Fine. Then let's talk about the work you've done. You've already told us a little about the story. Tell us about the methodology?

The methodology is extremely simple. There was a group of people from Citizen Observer. They distributed their observers on December 4 at 3164 sites in a random or, more precisely, quasi-random manner - observers were at the first, twenty-fifth, fiftieth, seventy-fifth sites. Since there is nothing to say that the plot numbers were allocated on purpose, this effectively results in a random distribution. There were a total of 156 sites in the sample.

This made it possible to compare two types of sites: sites where observers were present and sites where they were not. So we compare them according to the average value of those who voted for one or another party. Thanks to this, we were able to assess the extent of the falsifications for United Russia.

It must be emphasized that for the comparison to be meaningful, the distribution of observers must be random. My co-author Ruben Enikolopov ( NES professor, now working at the Institute for Advanced Study in Princeton ) for several years here in Moscow tried to convince everyone to conduct these random experiments, that is, scatter observers randomly. I don't know why Citizen Observer decided to do this - perhaps influenced by these attempts to push modern methods of analysis.

And, in fact, why did they decide not to distribute by lot? Because this random method is much more transparent to the surrounding public. It is still not very transparent, because a normal person, not with a higher education, usually does not understand probability theory and probabilistic concepts well. Because of this, people in general don't really understand why a sample of 156 precincts gives a very accurate estimate of what the scale of fraud was.

A huge number of things show that randomness and performance are quite complex things. Here is the simplest example. Ask your friends, just friends or from Facebook, which of them went to Bolotnaya Square. I have 20 or 30 percent of them. This means that the sample of my friends is completely unrepresentative, because otherwise, if it were representative, there would be 3 million Muscovites on Bolotnaya. This is a very common thing: people are bad at assessing the representativeness of their samples.

Popular argument: “None of my friends voted for United Russia.”

Yes, but the opposition, frankly speaking, does not have a monopoly on this mistake. In the same way, Putin continues to talk about nostalgia for the Soviet Union, although there is no reason to think that anyone other than his entourage and himself experiences this nostalgia. But they constantly talk about it to each other, and it seems to them that everyone thinks so.

Truth in numbers

How mathematicians look for traces of election fraud

Let's return, however, to the plots.

So, we had material for comparison. Of course, the question arose: what to publish here, because these results had already been announced by Citizen Observer on December 6 or 7, and at the same time there was a column by Alexei Zakharov in Elephant. But the scientific contribution of our paper is that we carefully wrote everything down, carefully explained what was done, plus we also performed some robustness checks. Robustness testing is, roughly speaking, when scientists suggest other (different from their own) possible explanations for a phenomenon and come up with a test that can be used to disprove these alternative explanations.

What kind of testing did we do here? I will return once again to the randomness of the appointment of observers. Imagine that, as in previous elections or as it was in other cities, observers would go to the polling stations they wanted, and we would get similar results. That is, it would turn out that the average for areas with observers differs from the average for areas without observers. What would that mean? This could mean that observers wanted to go to very specific polling stations because, for example, these are their home polling stations, where United Russia has an abnormally small percentage of votes. We got a completely different picture. Since we obtained a difference in the results between sites with observers and sites without observers, and observers were randomly distributed among sites (first, 25th, 50th, 75th in each of the large territorial districts), it turns out that the difference appears in the results for a reason unknown to us.

To reject accusations that the sample is unrepresentative, we took data from 2007. This is how I say “they took the data” - it’s intellectually very simple to take the data. In fact, this data in some sense does not exist in nature. And all because of the Russian Central Election Commission. You must understand that the main method of falsification of the Churovsky Central Executive Committee is not at all meaningful actions (I know nothing about them), but completely blatant mediocrity and incompetence. Therefore, for 2007, for example, there is no table that would allow comparing sections of 2011 with 2007. Not out of malice, but simply because they are completely worthless professionals. And so, with the help of a lot of work, it covered about a third of all the plots that were in 2007, the houses were brought into line with the plots of 2011. Accordingly, we received a fairly large sample and could compare whether the sites we selected were different from the rest. And it turned out that in 2007 they were not different at all, that is, the average for them was the same as the average for all sections. Consequently, the fact that now, in 2011, this difference exists, indicates the presence of the “observer presence effect” we declared, which reduces the percentage of falsifications.

So I explained to you how we verified that these areas were representative - using data from 2007. If someone feels distrustful, then he can conduct the same check and compare the December elections and the March ones - it’s easier, the data is there - and make sure that the sample is representative. We did this too and did not find any differences between these areas and others.

Fine. So, they chose it - and what results did they get?

The main result is that there is very strong statistical evidence of large-scale (at least 10.8 percent) fraud ( In the elections in Moscow on December 4, 2011, United Russia received, according to official data, 46.5 percent of the votes - approx. " Tapes.ru" )

Look. There are different ways to quantify the size of fraud. For example, what Sergei Shpilkin and his co-authors Kobak and Pshenichnikov are doing. They try to guess what the correct pattern looks like and calculate vote theft as a deviation from that pattern. Their method gives more accurate estimates of the size of the falsification, but it requires stronger assumptions (we need to know the shape of the “correct” curve).

In the case of our estimates, it can be said that almost nothing is required other than minimal assumptions. By the way, in Moscow our results very closely coincide with their results. This means little for our work, but for them it confirms their results in an independent way.

Your result is significantly at odds with the argument popular on the Internet that it is impossible to rig elections by more than 5 percent .

It is difficult to imagine that it was some person of a level lower than Surkov or Volodin who would have known the scope of the entire operation. I’m not even sure that they, or Putin or Medvedev, know. It’s unlikely that anyone assigned specific figures or outlined percentages for the plots. Most likely, at night they simply banged their fists on the table: redo the protocols, throw in as much as possible...

Fine. So it's 10 percent...

Results of the State Duma elections on December 4, 2012 in Moscow

"United Russia" - 46.5 percent
Communist Party of the Russian Federation - 19.4 percent
"Fair Russia" - 12.2 percent
LDPR - 9.5 percent
"Apple" - 8.6 percent

10 percent is the minimum estimate. If we look at those areas where observers reported that there are no violations, then the estimate is higher - up to twenty percent. Our main result has nothing to do with what observers reported. Our assessment is based only on whether an observer was or was not present at the site. You can separately calculate what happened in those areas where observers were removed. Where it was removed, it gets sharply worse. But the main result is not related to this. We took into account only CEC data and data on the presence of observers.

Do you yourself see any alternative explanations for your own results?

All statistical evidence is circumstantial in nature. But, as we know from the results of the presidential elections, even direct evidence, that is, video camera recordings, is not accepted for consideration in court. And yet it seems to me that, if we talk about the elections in Moscow in December 2011, then this circumstantial evidence proved everything according to the strictest judicial criteria.

As we see, no.

What does it mean? It happens that a judge releases a criminal, but then this undermines the authority of the court in our eyes. I cannot imagine that a reasonable person who read our article would agree that the Central Election Commission competently conducted the elections and counted the votes. But, of course, the article does not say anything that they did this on purpose.

Then the question is: have you made a similar comparison for the presidential elections?

We are not. Because none of us work on this topic purposefully. There are people who do this. I find it especially interesting that Kirill Kalinin from the University of Michigan does this. He deals with an issue that worries me, as a citizen with a mathematical education: Russian sociological firms, both those with a reputation as Kremlin court sociologists, and those with a reputation as independent sociologists (VTsIOM, FOM, Levada) - they all predicted election results with the amount of fraud now estimated.

I would like, since sociologists themselves do not say anything about this, for someone to analyze such a phenomenon. What is this? Are they really completely dishonest people or are they somehow able to, without feeling shameless, build models that automatically adjust the data to what they want? This can always be achieved - just rebuild the model manually: leave the suitable data, throw away the unsuitable. In the end, perhaps even unnoticed by yourself, with the help of some procedures, you arrive at exactly what you needed to achieve. I would be interested to know about this. Although I don’t understand: all over the world sociological firms live on orders from businesses, because for large companies this is a huge deal - all sorts of sociological surveys. For example, Apple releases some product, and one squiggle in the word Apple in an advertising slogan can cost, I don’t know, tens of millions of sales, hundreds of millions of dollars in profit. But who places orders with FOM, VTsIOM and Levada, if it is known that they give the result that they themselves want to get? How does what they do differ from their expert opinion?

Any other research in this direction? Maybe someone else is working?

I'm sure people are writing books about this election. That is, they sit and collect the same data. And ours too. The same Sergei Shpilkin, although he is a person with a mathematical education, is not a scientist at all, but he was simply forced to become a political scientist.

He is a passionate person.

Rather, I would say, offended, like many citizens here. Because there are people, and I count myself among them, who do not have a deep internal impulse to rebuild society. But there is a certain limit of abomination, from which you find yourself forced to become a citizen. And it seems to me that Sergei became a political scientist in the same way and began to analyze elections. He would have worked very well as a programmer if there had not been such lies and falsifications.

Any other research that can be done in this direction?

An interesting scientific question that may seem strange to a normal person: why is all this necessary? If in our country there was no question by March 2012, who ruled the country in February and who will rule in April. However, extensive measures were taken so that 67 percent could be officially declared. Why did they need this falsification? Well, if there had been no falsification, he would have scored 52 percent...

This is still a question from the field of political science, not mathematics.

Yes, but it's an interesting question. For example, when I was a kid, every election had one candidate, and it was announced that that candidate had won 99.99 percent of the votes with a 99.99 percent turnout. And, apparently, these indicators were not even calculated or aggregated. At least since the 1970s and 1980s, no data has been preserved; no one counted them. Why was this necessary then? What, there were people who thought that this was an election? This is also, I would say, an interesting question for political historians. What symbolic meaning is there in such elections?

Here I want to slip into Pelevinism and say that this was just such symbolism, something sacred, not at all scientific, not logical and, most likely, supernatural.

I would say that such an answer would not suit me very much, because most political regimes, and our current one, are not particularly inclined to sacralize anything. Apart from sacralization itself, of course.

It just seems like it's easy to fall into the trap of trying to explain things rationally when answering this question. The problem often lies in the fact that there are simply no rational positions.

Not that often.

As one of my math teachers said, fools are worse than smart people not because they are fools, but because they are unpredictable. A real fool, he is like that - random, he can come up with something that would never occur to a normal person. The teacher, however, spoke about the remainder terms in the Taylor formula, but this is not the point - it is quite applicable to today.

This is true. The fool lives in a simpler world; he uses cruder models like “Russia is surrounded by enemies.” If you think about it, we have a huge number of problems in all areas simply because we are governed by fools. We don’t have competent people in power, so they use some terribly simple schemes. Take, for example, our policy in the near abroad: friend or foe. American presidents eighty years ago, looking at Latin America, also considered “friend or foe,” but experience taught them that the point is not whether a politician is one’s own or a stranger, but that one needs to look deeper. Maybe we will also learn in 30-40 years. In this sense, your idea is correct: fools are more unpredictable, their world is simpler, rougher.

It just seems to me that the source of their unpredictability is not even in the simplicity of the picture of the world, but in the fact that their different parts of this world are not connected with each other. They are locally simple, but globally, on the contrary, they are more complex. They have chaos, no global order.

That's right. We are talking about two parts of the same thing. They have such a local idea of the world, consistent, but since they cannot compare it with anything, they cannot take a step to the side, they are forced to jump very much from one to another. By the way, we have built a good model of our society and thinking, but because we have begun to understand more, it is not so clear how we should act better.

Moreover, it didn't make things any better at all.

No, well, always understanding is better than not understanding.

Interviewed by Andrey Konyaev

Corrected to believe

Previous materials on the topic

Related links

Lenta.ru: Progress: Believe the corrected