Up until the end of 2021, the STEM election model was calculated as a so-called model with an input condition. Participants were included in the model if they met three conditions described below; the preferences of participants who didn’t meet either of the three conditions were excluded from the model. On the contrary, the current STEM model is computed as a probabilistic model: for each participant, we compute a weight of their vote based on their voting motivation and previous behaviour; the weight ranges from 0 to 1.
The legacy STEM election model expressed a percentage (an electoral grouping’s expected vote share at the time of the survey) based only on the responses of participants we assumed would actually turn out for the election. To define the expected voters, we took into account other circumstances. First, we excluded the participants who chose the answer “no” in response to the question whether they would go to the elections (on a “Yes – Not sure – No” scale). This question was always asked in the beginning of the questionnaire, even before the question for electoral preferences. Note that other research agencies could have provided different estimated shares of non-voters due to differing scales (like “yes-no” or “definitely yes – rather yes – rather no – definitely no”).
We asked about electoral preferences using an open-ended question. We did not provide the participants with any lists or parties or any other aid – we recorded verbatim what the participant stated. Only afterwards did we process this answer. For the election model, we did not take in account the proportion of those who had not indicate a party, indicated “none” or “don’t know”.
Immediately after the question on electoral preference, we asked a question on the strength of the participant’s attachment to the beforementioned party. This question had four response options, and we excluded the responses of people who, although they had named a preferred party, stated that they had a “very weak attachment” to that party.
Finally, we took into account the participant’s overall interest in politics. This question is regularly asked at the beginning of the questionnaire – it has four response options, and we exclude those who answered they were “not at all” interested in politics.
All this marked out a subset of roughly 55–60% of likely voters, whose preferences were reflected in the legacy election model.
Terminology overview
Each institution uses different indicators and terminology in their publications, which can make it difficult for the casual reader to navigate what is meant by specific terms in electoral surveys. On top of that, Czech terminology may not necessarily correspond to terms commonly used in the English-speaking world. In the Czech Republic, the terminology of research organisations has converged over time due to co-operation within the SIMAR association. Here are the definitions of some basic terms used in our election models:
- “Party preferences” = the proportion of participants in the entire sample indicating support for an electoral grouping (including people who state their preference as “don’t know” or “won’t vote”).
- “Election model” = an estimate of a hypothetical election outcome based on a mathematical model. The way in which this model is constructed may differ from company to company. An election model is not an election forecast, that is, it does not say what the results of the election are going to be, but rather what the current preferences are.
- “Voter preferences” = the proportion of participants who have not ruled out participating in the election, and who have stated preference for a given electoral grouping.
- “Party sympathies” = the proportion of participants who have not ruled out participating in the election, and who have either preference for a given electoral grouping, or, if unclear about their preference, expressed their interest in the electoral grouping in another question.
- “Electoral potential” = the proportion of participants who are considering voting for an electoral grouping, whereby each participant may name more than one electoral grouping.
Party preferences give the most complete picture of the current political sentiments in society. It shows what percentage of the total population prefers each party, what is the proportion of undecided, and what proportion of people is unable or unwilling to indicate a preference for any party. The sum of all these proportions gives one hundred percent, and STEM presents this value as the response to a single question which is not conditional on having answered any other questions in the survey, nor as the calculated result of two or more different indicators. The party preference indicator may be affected by a procedure called data weighting. STEM asks how the participant voted in the previous election and weights the surveyed proportions to better reflect known voter’s behaviour in those elections.
The election model uses current survey data as the basis for estimating the result of a hypothetical election outcome at the time of data collection but also supplements it with additional data and sometimes with assumptions based on previous experience that are not necessarily derived from surveying. This may include, for example, how disciplined each party’s voters are to turn out for the election, observed last-minute shifts in preferences before previous election and so on. There is no single universally accepted methodology for election models; each research agency may calculate their models differently, giving different weight to different aspects. As already stated, the basis for the model is stated party preferences. These show what the research institution actually surveyed (after weighting, if applicable). If a research agency does not disclose these preferences and only publishes a predictive model, both the expert and the layperson may have no idea how the research agency arrived at such a result.
Why political party preferences differ across polling agencies
The differences lie primarily in the different methodologies used by the organisations. STEM asks all the surveyed people about their electoral preferences using an open-ended question, meaning that the participant themselves names the party (or movement or coalition) they would vote for if the elections were held the following week. Some research organisations present a list of political parties to the participants and the participant chooses from this list. In this way, a higher proportion of people often state preference for smaller, less well-known political parties than when people spontaneously name their preferred party. The difference between the results of different agencies may also be due to the different proportion of surveyed “don’t know” responses.
