4.3. Data: Find Variance

It is pointless to ask a social scientist why a certain historical event happened or why a specific individual acted the way she did at a certain point in time at a given place. Social science is not rocket science. It is way more complicated. We do not examine perfectly predictable outcomes determined by only a few variables. There are so many candidate conditions and circumstances, forces and factors, genes and genomes, motives and musings, and other things that may have given rise to the event or the behavior that it is impossible to know which one is the culprit. Therefore, you need a simple system to organize the multitude of factors.

Suggestion number 1, specify units of analysis and look for variance, is the result of an often overlooked regularity in the social sciences. The regularity is that the degree of variance in the variables limits the validity of the conclusions. A research question can only be answered in a meaningful way if there is variance to be explained in the first place. This is the reasoning behind the recommendations #6 and #8 for research questions: make meaningful comparisons.

For many phenomena it can be useful to think about three types of variance, or levels of analysis:

  1. variance between individual units of analysis;
  2. variance within the units of analysis over time; and
  3. variance at higher order units.

In our example of figure 8, the variance between individual units would be the differences between individual citizens in their religiosity and volunteering behavior. Some individuals volunteer, and others do not. The research question about the cross-sectional variance between units would then be: do religious individuals volunteer more, and if so, how can this difference be explained?

The variance over time is represented by individuals moving into and out of the volunteer work force, and increasing or decreasing their level of engagement, i.e., spending more or fewer hours. The research question to be answered here would then be: do religious individuals start to volunteer more often, quit less often, and are they more likely to increase their engagement than non-religious individuals, and if so, why?

The variance in higher order units refers to a higher level in which the individual units of analysis are nested. Individual units can be located within higher order units at various levels of aggregation, such as households, corporations, parishes, municipalities, counties, or larger regions such as countries. An example of a research question about variance at higher levels is to what extent individual volunteering behavior is correlated with the volunteering behavior of other individuals in the same household, or how the number of blood donors in a municipality affects the likelihood that individuals start to give blood.

To structure your thinking, the three level ABC model is a useful tool (Bekkers, 2013; see Table 3).

Table 3. The 3 level ABC model

A good way to get to a list of variables of interest for your research is to start with the actor model we discussed earlier. Using the actor model you can identify the actors at the different levels of analysis that you need to take into account. In many cases, you will get to actors at three levels: the micro-level of the individual person, the meso-level of the organization or region in which the individual is embedded, and the macro-level of the country. At each level, it is useful to think about three types of variables: Antecedents, Behaviors and Consequences (ABC). Behaviors are the things that actors do that you want to explain. Antecedents are potential causes; they precede the behavioral outcomes. Consequences are the results of the behaviors you seek to explain.