Media, polls & interpretation
A poll cannot interview everyone in the Czech Republic in a single week. It uses a selected sample, weights and modeling rules to estimate support for parties within a particular time window. That is not a reason to discard the results; it is a reason to state the methodology and uncertainty whenever we turn them into a story.
1. What the evidence shows
Kantar CZ publishes a description of the research and its results for the Trendy Česka (Trends in the Czech Republic) project. The model uses a sample survey and statistical processing; it does not pretend to be a census of the entire population.[1]
Czech Television publishes data and methodological information for Trendy Česka as time series. A single wave should therefore be read alongside previous measurements and the conditions under which the data were collected.[2]
CVVM notes that interpretation requires knowing the wording of the question, the population, the sample, the fieldwork period and the statistical uncertainty. A difference between two point estimates need not be statistically significant.[3]
2. How to read the claim in context
Sampling error is not the only source of uncertainty. Results may be affected by some people’s reluctance to respond, weighting, decisions about who counts as a likely voter, and the allocation of undecided respondents. The methodology mitigates these problems; it does not eliminate them.
The media often turn ordinary noise into a narrative: a party “rises” or “falls” based on a difference in a single wave. It is safer to speak of a trend when a movement recurs over time and, ideally, across independent studies.
An election model and an election result refer to different moments. The model captures stated preferences under specified assumptions; campaigning, turnout and last-minute decisions can change the actual election.
- The model is based on a sample and statistical processing.
- The methodology and fieldwork period are part of what the result means.
- A small year-on-year or month-on-month difference may be indistinguishable from noise.
- The exact future election result.
- The preferences of people who systematically do not respond, without additional assumptions.
- A single universal adjustment for every form of non-participation.
3. Five questions to ask
- When and how were the data collected?
- Who does the model include in the denominator?
- Which weights and likely-turnout filters does it use?
- Is the change larger than the published uncertainty?
- Is it confirmed by several waves or another study?
4. Conclusion
An election model is neither the voice of the entire country nor a worthless poll. It is a conditional estimate. Its value as evidence increases when the number is accompanied by the method, uncertainty and a time series.
— Jiný Kontext