Presidential Election in Serbia - Part 2: Ipsos' Fast and Reliable Data

Ipsos was the only one to correctly predict the election, placing each of the 11 candidates in the correct order.

Presidential Election in Serbia - Part 2: Ipsos' Fast and Reliable Data

All the elections in Serbia are traditionally accompanied by the parallel counting of votes on the election day, and Ipsos Strategic Marketing started this 17 years ago. Twenty times in Serbia and seven times in Montenegro, the results based on this method were always reported very quickly after the end of voting, as well as turnout. Another two agencies have been doing the same, both originating from the NGO sector and both methodologically supported by Strategic Marketing during their professional growth.

Given that the final results became public four days after the election day, our data, which were stable one hour after closing of polling places were taken for final by all participants in the election. The results were presented through a software platform created for that purpose only – live on the most watched TV, which all the other media took information from. Mr Marko Uljarević and Mr Srdjan Bogosavljavić gave a series of statements on the election day – since two strongest TV channels had improvised studios set up in our premises.

The results deviate just scarcely – for less than 0.5 percentage points from the final, primarily because the votes obtained from the diplomatic missions of Serbia, from prisons and from the territory of Kosovo, which is under the special UN status, were not covered by the opinion poll.

The software platform used for the presentation of the results is available on the link It was created by the Ipsos team led by Mr Nebojša Sušić, the software platform creator, and Mr Aleksandar Zorić, the director of Ipsos Adria Cluster and Ipsos Strategic Marketing Operations, and supervised by Mr Dragiša Bjeloglav, retired COO, and Mr Marko Uljarević, Public Affairs director.

The sample was random stratified sample of polling places, size n= 850 out of 8500 polling places. Very detailed stratification was conducted on the basis of the results of the previous voting - 40 strata. This, de facto cluster sample of population was treated as simple random sample of polling places with unproportional allocation by strata. The data were entered through interviewers’/ observers’ mobile phone application, on polling places. For instance, for the 3pm breakdown, the results for 700 out of 850 units were reported in 6 minutes.

The Ipsos data were not only broadcasted live on the biggest TV channels in Serbia - the Public Service and the most commercial TV channel, but also the turnout appeared half an hour before all other data, and the results appeared even earlier.

Same as before, Ipsos had fast and reliable data. This time the speed was extreme, and it was analysed in detail. We are still to examine the observed extreme precision of scores on a lot smaller sample, even below half, but only in strata with at least 20% of realised units, as well as the fact that empirically the Mean Squared Error is systemically considerably smaller than theoretical.