(no subject)

[kireev]

В "Аргументах и Фактах Online" опубликованы мой график корреляции между голосованием за "Единую Россию" и явкой в разрезе участков, плюс сравнение результатов выборов на участке в главном здании МГУ и Кащенко. В целом ряде блогов были размещены ссылки на эти материалы, так что количество моих френдов за последние пару дней значительно выросло. Это вообще уже стало характерно для моего блога: количество френдов растет рывками прямо перед важными выборами (скажем, президенсткие выборы в США год назад), во время выборов и сразу же после выборов.  Я, разумеется, не могу сразу со всеми познакомиться. Но если вы убеждены, что мне будет интересно читать ваш блог, что у нас похожие интересы, напишите, не стесняйтесь.

Comments

[cia27012009.livejournal.com]

These are all interesting questions and my short answer is: I don't know and I don't care.

My point is that these data do not prove falsifications. They may be explained by falsifications but you need to have some other argument to validate the falsifications explanations. Falsifications are usually claimed because of the discrepancies with exit polls.

Kireev assumed that distributions with small number of voters are fair representation. So any deviations from this at large numbers should be due to some factor. Because increasing the number of voters should supposedly give you smoother distribution but with the same shape, the large numbers are then obtained unfairly.

I say: what if the large numbers of voters represent the real distribution? That means that the deviation at small numbers are caused by some factor which makes small sets invalid. How can it be? Well, if you are sampling from inhomogeneossly distributed representation, that's what you get. "Activity" here is just some factor which determine the probability for the voter to participate in the election.

By the way, I believe it is accepted theory that non-conformist voters are more likely to vote then conformist voters. Because it is very common to ask voters if they are going to vote and how, this could be relatively easy to correlate with the data. Ignoring this makes the conclusion more intuitive and less rigorous.

It is like you want to find the average weight of the women in your company and kindly asked them to come. I bet you number will be well below the true value unless you haul them all forcibly for the measurements.