Guest post: Eszter Bartha, Attila Gulyás, Gray zone – modelling the opinion distribution before the general elections using simulation methods

I. Introduction
As the election date gets closer (06.05.2010) the major public opinion research institutes publish daily results about people’s sympathy for the different parties. There are numerous parties to select from but usually the three major parties are dealt with in the published results: the Conservative, the Labour and the Liberal Democrat party.
The results in the news and on the web describe the headline voting intentions only. But what about those yet undecided, or not willing to answer the survey questions? The ratio of these voters is substantial, ranging from 11% to 31% in the fresh data of the major public opinion research institutes. Of course the participation of these people is very important and it may influence the outcome of the elections seriously.
Our goal in this short paper is to introduce the results of a simulation method for the 2010 general election party sympathies. We do not aim to guess the number of seats obtained by the parties, but we try to estimate the ‘real’ opinion of people and – given the participation – the overall distribution of votes.
II. The simulation model
The most spectacular feature of the English polls found in the media is that the presence of people hiding their opinion is not strongly emphasized (compared to Hungarian practice). These people hide their opinions as because of some reason. This phenomenon is described by the ‘threshold models’. These models assume that people have a given threshold – a ratio of people sharing the same opinion as the given person – below which a person wouldn’t express his/her opinion.
We assume that this ‘threshold’ is also influenced by the party, for which a person feels sympathy. Also the importance of this ‘threshold’ when expressing someone’s opinion is influenced by the structure of the person’s network – namely the ratio of people with similar and different opinions.
Based on these assumptions we created a model (partially following the first steps of Krassa (1999), then extending it with our assumptions) which simulates the assumed opinion distribution based on existing public opinion data. We used the data available on the net – to be more exact we used the latest raw data from YouGov, Ipsos Mori, ICM, Populus and Opinium.
III. Simulation results
We also had to make some exact initial assumptions. First we assumed that the threshold values according to the parties have certain differences with the Labour party sympathizers having the highest threshold – meaning that they hide their opinion mostly. Then the Conservative party follows, followed by the Liberal Democrat part and the smaller parties. Also we assume that the smaller party sympathizers are the most open among the parties and that they have ‘more dense’ relationships towards each other. These assumptions given as an input parameter set to our model resulted in the following distribution:

Party Conservative Labour LibDem Other Don't know/refuse

Voter rato 0.27 0.23 0.22 0.09 0.19

Sim result 0.2494 0.1885 0.2059 0.830 0.2732

Simulation results using this data are partially in line with major results. Notice the difference between people who don’t have an opinion at this time and the sim results. The sim results include those as well who wouldn’t have expressed their opinion.
So it shows that per the simulation the supporters of the Labour party are as many as those of Liberal democrats. Also conservative voters are over-represented in some samples.
Of course our work also aims to give an overview about the general distribution of votes (not the seats!). To achieve this, we had to make assumptions about the mobilization as well. We assumed that the labour party does not have such high mobilization resources as others do. We assumed that small parties and the Libdem party are at a smaller advantage when compared to bigger parties. In our simulation we assumed three possible scenarios for mobilisation (M1-3). As per the trend of the last elections we expect a participation rate of ~67-70%. Supposing this the following general distribution was found:
Party Conservative Labour LibDem Other Participation

M1 0.7 0.5 0.75 0.85
Votes 33.59% 22.17% 30.43% 14% 67.29%

M2 0.7 0.6 0.7 0.8
Votes 32.96% 26.21% 28% 12.92% 68.41%

M3 0.7 0.6 0.75 0.8
Votes 32.29% 25.67% 29.17% 12.86% 69.80%

This table shows that the Conservative party takes lead having the LibDem behind them. The labour party is far behind these two, but above the votes given for the smaller parties. Although the Liberal Democrat Party has a smaller advantage in terms of mobilization, their support – as shown by the assumed distribution – is not enough to catch up with the conservative party. The support of other parties – assumed that their mobilization is strong – is relatively high (we did not analyze the support individual small parties).
IV. Summary
In this short paper the results of the application of a simulation model was introduced. Its purpose is to estimate the real opinion distribution in the society using data from public opinion research institutes, and also to give an estimation to the overall distribution of votes. We do not aim to give seat estimates. As the tables show, the conservatives are in a leading position and their victory is secured. If the effect of the labour party campaign really is not big enough encourage voters for the participation then the labour party will lose by a substantial margin with the LibDem party finishing second. The other parties also consume a fair amount of votes, and the participation rate may get even closer to 70%.
The configuration of the House of Commons depends mostly on the local results, so the new boundaries may also play an important role, but these external effects were not in the scope of our model.
If you have any comments or questions, please write a mail to Attila.gulyas@uni-corvinus.hu