Questions and answers

If the sample is built properly — every person in the group being studied had a chance of being included — a small sample reflects the whole well. For a simple random sample of 1,000 people and a proportion of around 50%, the margin of error is roughly ±3.1 percentage points.

No. The margin of error shrinks in proportion to the square root of the sample size, so to halve it you need a sample roughly four times larger.

No. The statistical sampling error does not account for poor question wording, non-response or an unrepresentative sample. These errors can matter more than the margin of error itself.

Who commissioned and who conducted it, the dates, the survey method, the size and composition of the sample, the exact question wording and the margin of error. Professional associations of researchers, such as AAPOR and ESOMAR, have standards for disclosing methodology.

With few reviews, the average score depends heavily on each individual opinion. Five glowing reviews give a perfect score but say far less about the product than a thousand reviews with a slightly lower average.

Not quite. Reviews are voluntary and are more often left by people who are very satisfied or very dissatisfied. Such a sample is not random, so reviews are useful for spotting specific problems but poor at showing how most people rate a product.

Only that, under the criteria and weights chosen by the compiler, that entry scored the most points. With other criteria the order could change, and places separated by fractions of a point may differ by chance.

Look for a disclosed methodology: who compiled it, using what criteria, on what data and over what period. If the methodology is not published, the ranking is hard to verify and should be relied on with caution.

The mean is the sum of the values divided by their number; it is sensitive to extreme values. The median is the middle value of an ordered series; extreme values barely move it. That is why incomes and house prices are often reported as medians.

Not necessarily. Correlation does not imply causation: both quantities may depend on a third factor or coincide by chance. To talk about cause, you need additional data and a well-thought-out study design.

It is a situation in which a trend within each group disappears or reverses when the groups are combined. The classic example is the 1973 graduate admissions data from the University of California, Berkeley: the overall gap between men and women was explained by which departments they applied to.

Russia's official statistics are published by Rosstat, the Federal State Statistics Service. One example is the results of the 2020 All-Russian Population Census, which, because of the pandemic, was held from 15 October to 14 November 2021.