Start with the question
The unit of analysis is the thing a study compares or describes. In civic research, it might be a person, an electoral district, or a country. Choosing one is not a technical detail to settle after collecting data. It determines what counts as an observation, which comparisons are possible, and what the findings can reasonably say.
Suppose the question is, “Who feels confident explaining how a bill becomes law?” People are a natural unit because confidence belongs to individuals. If the question is, “Do districts with more public meeting places tend to have higher participation?” districts may be more useful. A question about how different electoral systems organize representation might call for countries. Before choosing data, complete this sentence: “I want to explain differences among ____.” If the blank is unclear, the research question may need sharpening.
People: useful for individual experiences and behavior
Person-level data can describe attitudes, knowledge, reported experiences, or actions. A survey, for example, could compare how respondents learn about a local policy proposal. Its observations are people, even if every respondent lives within a district. This level is useful when the conclusion concerns individuals: whether people who attend meetings report feeling more informed, for instance.
The limits matter. Survey answers may be shaped by question wording, memory, or who chooses to respond. Two people in the same district also share some surroundings, so their answers may not be wholly independent. A study should describe how participants were selected and avoid assuming its respondents represent a wider population without a sound basis. Most importantly, a person-level association does not by itself show that one experience caused another. People who attend meetings may already be more interested in local affairs.
Districts: useful for places and shared conditions
Districts can be appropriate when the question concerns local institutions, service coverage, or geographic patterns. A researcher might compare the availability of polling places across districts or examine whether areas with different transit access show different turnout patterns. In each case, a district is one observation, even if its measures summarize many residents.
District averages conceal variation within their borders. A district with a high average income includes residents with different incomes; a district with high turnout includes people who did not vote. Inferring an individual’s behavior from a district-level pattern is an ecological fallacy. District boundaries can also affect the result: combining neighborhoods one way may produce a different pattern from combining them another way. When comparing districts, check whether they cover comparable populations, whether boundaries changed, and whether a count should be expressed as a rate with a clearly stated denominator.
Countries: useful for institutions, but hard to compare
Country-level analysis helps with questions about national institutions and broad policy designs. Researchers might compare how legislatures are structured or how responsibility for a service is divided between national and local governments. Here, each country contributes an observation, regardless of its population size. A country comparison can highlight possibilities and generate questions, but it should not flatten complex political systems into a single label.
Terms such as “participation” or “local authority” may be defined or recorded differently across countries. Countries also differ in many ways at once, making it difficult to isolate the role of one institution. A pattern across countries does not establish how individual residents think or act. Clear definitions, attention to historical and institutional context, and caution about causal claims make these comparisons more informative.
Keep the unit consistent with the conclusion
A practical check is to write down three items before analyzing anything: the unit, the measure, and the intended claim. For example: “The units are districts; the measure is the share of eligible voters who voted; the claim concerns differences between districts.” If the proposed conclusion suddenly says that residents with a particular characteristic are more likely to vote, it has moved beyond what those district-level observations can show.
Some projects need more than one level. People live in districts, and districts operate within countries. A study can collect information at several levels to explore how individual experiences relate to local or national conditions. Doing so requires keeping those levels distinct rather than treating every observation as interchangeable. When a complex design is not feasible, a narrower question with a well-matched unit is often more useful than a sweeping conclusion. The best choice is the one that lets the evidence answer the question actually asked.
