Turkey Insights Lab publishes survey-led research with enough context for a reader to identify what was asked, who answered, which responses were included in a calculation and which limitations matter. A percentage without its base or question context is incomplete evidence.
1. Research question and questionnaire design
Each study starts with a defined research objective. Questions are written to measure a specific behaviour, preference, expectation or attitude. We aim to use concrete timeframes such as “in the past 30 days” when recency matters, and answer options that can be understood without specialist terminology.
When a question allows more than one response, that is disclosed because the resulting percentages may legitimately sum to more than 100%. When only a subgroup is eligible to answer a question, the subgroup base is reported instead of the total study sample.
2. Online-panel fieldwork
Our published original studies can use online-panel respondents living in Turkey. Participation in an online panel is not the same as selection through a probability sample of every person in the country. For that reason, a study page identifies the mode and describes results as findings from the participating sample unless a different design is explicitly documented.
Sample size is determined study by study. A larger sample can make descriptive percentages more stable, but sample size alone does not make a non-probability online panel nationally representative. Coverage, recruitment, eligibility, routing and the composition of the achieved sample remain important.
3. Eligibility, routing and analytical bases
Survey routing can change the denominator used for a result. For example, a question about payment methods may be asked only to respondents who recently shopped online. In that case, the percentage is calculated among recent shoppers rather than everyone who entered the study. We call this the analytical base.
Research pages state important bases close to the result and explain routing in the methodology section. This is especially important when comparing percentages across questions, because two percentages from the same study may use different denominators.
4. Data-quality checks
Before publication, study data can be checked for duplicate participation, invalid routing, internally inconsistent records, unusually fast completion or other study-specific quality signals. The appropriate checks depend on the questionnaire and collection system; there is no single universal rule that fits every project.
When a data-quality rule materially changes the analytical base, the published study should explain the rule. For example, if repeated participant tokens are found, the publication can specify which completed record was retained and report the resulting unique analytical sample.
5. Weighting and unweighted reporting
Weighting changes the contribution that different observations make to an estimate. It can be useful when a study has a defined benchmark and a defensible weighting plan, but weighting can also increase variance and does not repair every form of sampling bias. We therefore disclose whether a study is weighted or unweighted rather than implying that one approach is automatically appropriate for every project.
When a publication is labelled “unweighted,” the percentages are calculated directly from the included responses in the stated analytical base.
6. Percentages, counts and rounding
Where useful, we publish both percentages and counts. Counts help readers understand the scale behind a percentage. Percentages may not total exactly 100% because of rounding, excluded responses or multiple-response questions. A multi-select result should be interpreted as the share of the stated base selecting each option, not as mutually exclusive market share unless the question was designed that way.
7. Comparisons and subgroup findings
Differences between groups can be descriptively interesting, but small subgroups can produce unstable percentages. A large percentage difference based on a very small base should not be treated the same way as a difference supported by hundreds or thousands of observations. When subgroup results are published, the relevant base should be visible.
We also distinguish descriptive comparisons from causal claims. A survey showing that two characteristics occur together does not by itself establish that one caused the other.
8. Publication and traceability
Each original study receives a research ID in the format TIL-YYYY-###. The research page is the primary source for the study and includes the fieldwork period, analytical base and interpretation notes. When an aggregate topline download is provided, it contains summarized counts and percentages rather than respondent-level identifying records.
9. Corrections
If a material reporting error is found after publication, the affected content should be corrected. Material changes should be documented when they alter the interpretation of a published result. Editorial clarifications that do not change the underlying data may be made for readability.