A reliable political poll must carefully balance the sample composition and his number.
1. Are sociodemographic variables necessary?
Yes, they are essential. If a survey were limited to interviewing 1,000 people at random (for example, only those passing by downtown at 10 a.m. on a Tuesday), the results would be skewed: they would mostly interview older people or students, excluding full-time workers.
To ensure the representativeness, research institutes use a stratified sample by quotas, ensuring that the percentage of respondents accurately reflects the real composition of the population:
- Age and Gender: Indispensable. Political orientation varies dramatically across generations and genders.
- Educational Qualification and Profession: Crucial for analyzing social class, presumed income, and membership in specific constituencies.
- Geographical area/Neighborhood: In municipal or local elections, residence within the city is another decisive variable (e.g. suburbs vs. historic center).
If the sample size does not reflect the actual population, the survey loses scientific validity.
2. How many people should be interviewed in a city with 1 million voters?
Contrary to intuition, the size of the total population has little impact on the number of people to interview. What really matters is the sample size depending on the desired level of precision.
For a city of 1,000,000 voters, a sample of between 800 and 1,000 people is sufficient to obtain a scientifically reliable result.
| Sample Size | Margin of Statistical Error (95% Confidence Level) | Assessment |
| 400 interviewees | ± 4.9% | Not very precise, useful only for detecting macro-trends |
| 800 interviewees | ± 3.5% | Acceptable standard for local polls |
| 1,000 interviewees | ± 3.1% | The industry standard for reliability/cost |
| 2,500 interviewees | ± 2.0% | Very high precision (used for large national surveys or exit polls) |
Why are 1,000 people enough for 1 million voters?
In statistics, the metaphor of minestrone soup applies: to taste whether a soup is salty enough, you don’t need to drink half of it. Simply stir it well (i.e., layer the sample thoroughly) and drink just one spoonful (the sample of 1,000 people).
The practical margin of error:
If the poll gives a candidate the 40% of the votes with 1,000 respondents (margin ±3.1%), its real percentage in the final population will be between 36.9% and 43.1%.
Increasing the sample from 1,000 to 10,000 people would reduce the error to only about ±1%, but would triple or quadruple the cost and time to complete the study without adding proportional value.