There is a model that helps predict Seimas election results more accurately than surveys

As M. Jastramskis concludes in a monography endorsed by VU TSPMI scientists Can you predict elections? The problem of three bodies in Lithuanian politics, survey-based forecasting is problematic in Lithuania.

The article highlights sociologist Vladas Gaidys‘ statement in 1998 that the 1992 and 1996 surveys forecast the Seimas election results quite well, however the Lithuanian Democratic Labour Party’s (LDDP) results were unexpected even to itself. Such a conclusion was made based on the words of the party’s chairman Aloyzas Sakalas. Once again, as M. Jastramskis’ article notes, the accuracy of forecasting was not improving.

The article also quotes the founder of Rinkos Tyrimų Centras [Market Research Centre [ Mindaugas Degutis’ data from 2010 which showed how in 2008, predictions for the Seimas elections were inaccurate: surveys did not predict that the largest number of votes would go to the Homeland Union – Lithuanian Christian Democrats (TS-LKD) and National Resurrection Party (TPP). In 2014 VU communications sciences dr. Laima Nevinskaitė concluded that the survey performed closest to the 2012 elections rather accurately predicted the number of votes for the two winners of the elections (The Lithuanian Social Democratic Party (LSDP) and Labour Party), however overrated the performance of Order and Justice, while downplaying the performance of the TS-LKD.

The monography concludes that the victors of the 2016 elections (even considering the data of the last surveys up to the elections) would not have been any clearer. The key reason is that the LSDP obtained 9% less votes in the elections than was forecast for it based on the joint, adjusted average of the surveys. The joint rating’s predictions were rather accurate for smaller parties, bar the somewhat overestimated Labour Party.

Quarter of voters decide only several days to elections

M. Jastramskis’ article notes that even under the premise that surveys are representative, in any country they will face major challenges if the electorate tends toward shifting behaviour: “How can you speak of prognosis when a large number of survey takers could still change their opinions?” the book questions.

The article points out a conclusion made by L. Nevinskaitė in 2014 that at least 25% of voters decide with several weeks remaining to the Seimas elections. This is significant for forecasting because this many voters can make two parties the victors of the elections (the LVŽS and TS-LKD), while pushing a former favourite (the LSDP) into third, it notes.

The article presents econometric statistical models as the main alternative to supplement surveys. In them, from factors portrayed by historical data, a formula is devised and based on it, the number of votes a party or parties could receive is forecast. The formula usually involves the majority party ratings and the country’s economic performance metrics.

The article states that with a month left to elections, statistical models make more accurate forecasts than surveys on their own. This happens because campaigning activates voters’ thinking about the situation in the country.

In his article, M. Jastramskis notes that in countries such as Lithuania, there are two problems in applying this model: first, it is not completely clear, whose votes to apply, second, there is a major lack of historical data with the country not even having held 10 parliamentary elections and both parties and economic context having fluctuated greatly.

The political scientist has created a model in Lithuania based on academic articles, which propose forecasting according to past secondary elections and which suggest that in multi-party systems, it is typically the votes for the ruling coalition that are forecast. Based on this, the model forecasts majority parties’ performance in the Seimas elections based on past municipal elections.

“Everything is simple – this works for only the majority parties because the premise is that the number of votes they will receive in the municipal elections, later on that percentage of votes will diminish somewhat over the final year and a half. […] This works out for some parties and less for others,” M. Jastramskis told Delfi.

While the political scientist notes that a part of factors that decide results in the municipal and Seimas elections differ, such as the stronger locality of municipal elections, results during them allow parties to grasp how well they are doing.

“If a party’s popularity has declined, it will already be visible in the municipal elections. […] It will not show real popularity, but will certainly allow parties to find their bearings, whether they are doing poorly, badly or average,” M. Jastramskis said.

Model has flaws

The monograph points to a flaw of the model being that it can only forecast the performance of majority parties. It is based on one fundamental trait of the Lithuanian electorate: it, akin to other post-communist states’, is hyper-accountable and always punishes the government in elections.

“Nevertheless, there are exceptions such as the Liberal Movement in 2012 – so far it is the only majority party that managed to gain more votes after being in the coalition government than in the prior Seimas elections. In 2012 the model greatly underestimated right wing cabinet parties. In 2016, the model’s predictions fared better, but this time ruling coalition parties were somewhat overestimated, especially the Labour Party,” the article states.

It highlights that the key premise of fundamental models is that supply will not change significantly – powers with a similar image will remain in power, their work or their flaws are already clear, as is the country’s economic direction.

“This premise does not work in Lithuania. There have been a number of corruption scandals since the 2015 municipal elections. They clearly influenced ratings. […] A large flaw of both fundamental models and surveys in Lithuania is the inability to predict, what could emerge from small or new parties. Adding a leader as an accidental factor, which can go with a variety of parties and it becomes problematic to forecast in principle,” the article concludes.

A third problem is that the electoral system can strongly influence results and at the same time lead to inaccuracy in predictions. “Seimas elections in Lithuania are hard to predict,” the scientist concluded.