Go back

2021 State Duma elections: first statistical estimates

27.09.2021

Conducting mathematical analysis of the official election results published by the Russian Central Election Commission (CEC) has proven to be one of the critical mechanisms allowing the Russian civil society to uncover and demonstrate possible manipulations with the election outcome to the public.
Alongside publicly accessible live streams from the polling stations, publishing election results per polling station online in real-time has been a cornerstone of electoral transparency in Russia. Unfortunately, it seems that the 2021 election administration has attempted to put an end to both of these practices.
On September 19, Sergei Shpilkin – a mathematician and a physicist that has traditionally analyzed official election results – posted on Facebook that he was experiencing difficulties extracting the data from the CEC's website. According to him, apparently, the election administrators have put a scrambler on the displayed election results. Now, when one was trying to copy them off of the website, the numerical data would appear as 'strange' characters.
While these technical barriers have interfered with the swiftness of his analysis, Shpilkin and others have eventually been able to overcome them. Below, we offer a translation of Shpilkin's report.

The State Duma election: primary statistical estimations

Author: Sergei Shpilkin, researcher of electoral trends, physicist

Here is the data from 96,840 polling stations (107.9 million registered voters out of 109.2 million eligible to vote).

On the left is the chart of votes given for parties by 1% turnout interval. The thin line charts the votes for all parties except the United Russia (UR), scaled so that it coincides with the UR chart at an early stage. With an assumption (which is not fully accurate) that all falsifications for UR originate from ballot stuffing, the shaded area represents the volume of stuffing.

On the right are shares received by parties depending on turnout by polling stations, where each polling station is a set of dots of different colors, one color per party. The dense core represents the presumably unfalsified polling stations. The center of the UR core is established via Minimum Covariance Determinant method, where polling stations were attributed weights by the number of voters, the cut-off threshold 50% (the ellipsis shows the cut-off dots). Correspondingly, the UR core center coordinates help to estimate the real turnout and the actual UR outcome.

At polling stations where the results appear genuine, the turnout is on average 38% (akin to 2016), and UR gets between 31% and 33% by two estimates (akin to 2011). Notably, spikes of votes at whole-number percentages have decreased; yet, they are still well above the significance threshold, as calculated by Kobak.

The spike of 95% turnout is the e-voting in Moscow.

Legend:
Chart on the left:
Vertical axis – number of votes 
in polling stations by 1% turnout intervals;
Horizontal axis – turnout.
Chart on the right:
Vertical axis – candidate result;
Horizontal axis – turnout.

A detailed analysis of the official results per region may be found here.

An original report may be found here (RU).

Related analytics

See all