If a survey was conducted using correct demographic quotas (usually gender, age groups and regions), the results can be extrapolated to the whole population of the corresponding target group. Nevertheless, the results received in each survey contain a margin of error because only a part of the population (called sample) is interviewed.
The more respondents completed the survey, the lower the margin of error is. But this relation is not linear. If we have a sample of 10 completes, the margin of error is 31%. With 100 completes it gets down to almost 10%, and with 1000 completes it is 3,1%. To reach 2% we need to interview 2400 completes, and to reach 1% we need at least 9600 respondents.
So, there is no big difference in data accuracy when we deal with thousands of completes, but a bigger sample size might be necessary to analyze smaller population groups (for example, to split the data by cities, or when combining several variables). At least 100 completes per group is recommended for data analysis.
The margins of errors presented below are calculated with 95% confidence interval. It means that if we sampled 1000 respondents, and 50% of them answered that they drive a car, there is 95% probability that in the whole population 50% +/-3,1% people drive a car, i.e. between 46,9% and 53,1%. And there is a 5% probability that the real number in the population is outside this interval.
| Sample size | Maximum margin of error |
| 5000 | 1,4% |
| 3000 | 1,8% |
| 2000 | 2,2% |
| 1000 | 3,1% |
| 500 | 4,4% |
| 300 | 5,7% |
| 200 | 6,9% |
| 100 | 9,8% |
| 50 | 13,9% |
| 30 | 17,9% |
| 20 | 21,9% |
| 10 | 31,0% |