Survey results are easy to compress into a headline: “64% use X” or “54% expect Y.” The difficult part is deciding exactly who the percentage describes, what question produced it and how far the result can reasonably be generalised. Good survey reading starts with the denominator, not the headline.
1. Start with the exact population measured
Ask who was eligible to participate and how respondents were recruited. A result from an online panel is a result from that participating online-panel sample unless the study documents a design that supports a broader population estimate. The phrase used in the article should match the population actually measured.
This is why “89% of respondents” and “89% of people in Turkey” are not interchangeable statements. The first describes the observed sample. The second makes a population-level claim that requires a stronger sampling and estimation basis.
2. Find the analytical base
The total number of completed interviews is not always the denominator for every question. Survey routing often means that a later question is shown only to respondents who meet a condition. A study might have 5,077 included respondents overall but ask payment-method questions only to the 4,532 people who recently shopped online.
If 2,903 of those 4,532 recent shoppers choose credit card, the relevant percentage is calculated on 4,532, not on the full sample. A careful publication therefore states the subgroup base close to the result.
3. Read the question wording
Small wording changes can alter what is being measured. “Which services have you ever used?” is different from “Which services did you use in the past 30 days?” A question about expected future behaviour is different from a forecast of what the market will do.
Timeframes, definitions and answer options matter because respondents answer the question that was presented, not the shorter label used later in a chart.
4. Check whether the question was single-choice or multi-select
In a single-choice question, each respondent usually contributes to one answer category, so percentages commonly add to about 100% after rounding. In a multi-select question, one respondent can select several options. The percentages can therefore add to well above 100% without any error.
Multi-select percentages are best read as reach: “the share of the eligible base selecting each option.” They should not automatically be called market share because the options are not mutually exclusive.
5. Look at counts as well as percentages
A percentage can hide a small base. Fifty percent of 20 people and fifty percent of 2,000 people look identical in a headline, but they do not provide the same amount of descriptive information. When subgroup results are important, the base should be visible.
Counts are also useful for checking a calculation. If a page reports 90.8% and shows 4,113 of 4,532, the count and base make the percentage auditable.
6. Understand weighting
Weighting changes how much different observations contribute to a reported estimate. A weighted result should identify the weighting approach or at least disclose that weighting was applied. An unweighted result is calculated directly from the included observations.
Weighting is not a universal quality switch. It can help align selected characteristics with external benchmarks, but it can also increase variance and it cannot automatically correct every source of selection bias. The important editorial requirement is transparency about what was done.
7. Distinguish descriptive differences from meaningful conclusions
When two percentages differ, consider the base sizes and the study design before treating the gap as a durable difference. Very small subgroups can move sharply with only a few responses. Repeated comparisons also increase the chance of noticing patterns that may not reproduce.
Descriptive survey reporting can say that one group reported a higher percentage than another. Stronger inferential or causal language requires additional statistical and design justification.
8. Treat future-intent questions carefully
A question such as “Do you expect your online shopping to increase in the next 12 months?” measures respondents’ expectations for their own behaviour at the time of the survey. It does not directly forecast national e-commerce revenue, transaction volume or market growth.
Intent can be useful because it captures direction and sentiment, but the publication should describe it as intent or expectation rather than converting it into a market forecast.
9. Check the fieldwork dates
Technology use, prices, platform preferences and consumer intentions can change quickly. Fieldwork dates tell you when the responses were collected. A result can remain historically useful while becoming less suitable as a description of the current moment.
10. Read limitations before generalising
Every survey design has boundaries. For online-panel research, relevant limitations can include panel membership, internet access, non-probability recruitment and the achieved composition of the sample. A transparent limitation does not make a study useless; it defines how the evidence should be used.
A compact checklist
- Who was eligible and how were respondents recruited?
- What is the total analytical sample?
- Does this result use the total sample or a subgroup?
- What exactly was the question and timeframe?
- Was the question single-choice or multi-select?
- Is the result weighted or unweighted?
- Are the counts or subgroup bases large enough for the intended comparison?
- What dates was fieldwork conducted?
- Does the headline match the population actually measured?
- What limitations are stated on the source study?
How Turkey Insights Lab applies this
Our source study pages identify the research ID, fieldwork period, analytical base and reporting approach. When routing creates a subgroup base, the study explains it. When a question allows several responses, the page notes that percentages can exceed 100%. Aggregate toplines are provided for selected studies so readers can inspect counts and percentages without exposing respondent-level records.