If your research question requires you to assess whether the variation in some individual-level factor depends on some macro-level variable, then you will need to linked macro and micro data to answer that question. The following are examples of such questions.

There are some questions that require both micro and macro data, but the macro data can be generated from the micro data, and so it isn't clear that you are required to link two different sources of micro and macro data, when you might instead be able to generate the macro data using the micro data that you have already. For example, suppose you are interested in whether the level of trust in politicians varies between countries according to the level of interpersonal trust. It could be that you have two separate data sources, one at the individual level showing the level of trust in politicians, and one at the macro level providing the level of interpersonal trust in each country. But since the level of interpersonal trust is simply the aggregation of an individual level survey question to the country level, it maybe possible to generate the country level of interpersonal trust from the individual level survey data if the relevant question is included in the same survey as the 'trust in politicians' question.

Sometimes cross-national aggregate statistics from micro data are already available, e.g. for the Eurobarometer, Latinobarometer and some World Bank datasets. You might find it easier to link the pre-prepared aggregates than to write code to generate them yourself.

In cases such as these, you need to consider whether there is better quality macro data available from sources separate from the micro data source to make it worthwhile linking the macro data to the micro data, rather than generating macro data from the micro data. For instance, suppose you are interested in whether there are neighbourhood (or contextual) effects in operation so that people are more likely to vote Labour in more working class constituencies. Here there is a choice between generating a measure of the class composition of the constituency using micro level information on the class membership of the respondents to the survey (e.g. the British Election Study) and aggregating it to the macro (constituency) level, or linking the census data on the class composition of constituencies to the micro (survey) data. In practice it is clear that the census data is of much higher quality than macro data generated from any survey that could be used to analyse voting behaviour, since in any suitable survey there are only a small number of respondents in each constituency.

In addition to assessing whether the level of some particular micro level outcome vary between macro units according to characteristics of the macro units, it may be that patterns of association between two or more micro level variables might differ between macro units because of some macro level factors. Such effects are sometimes described as cross-level interactions, because the effect of one micro variable on another depends on the level of a macro variable. Research questions which posit these kinds of effects also require linkage between micro and macro data. The following are examples.