Appendix F: Ecological fallacy
Social science researchers often study groups of cases, especially groups of people. To draw warranted conclusions from such research, we must be very clear about whether we are drawing conclusions about groups or individuals. A common error is to attribute group-level characteristics to individuals; this error is the ecological fallacy. A researcher could succumb to the ecological fallacy by erroneously assigning a group characteristic—most people like Star Wars—to an individual—Sally likes Star Wars. That kind of ecological fallacy is easy enough to spot. Trickier to spot is the ecological fallacy of assuming relationships observed at the group level also describe relationships at the individual level. We can fall into this trap when we forget that group summary statistics, like the group average, can hide a lot of variation within groups.
For example, imagine we conducted a survey of volunteers at the Downtown Food Bank, Midtown Food Bank, and Uptown Food Bank, asking them how many hours they volunteer per month and whether they consider themselves to be generally happy. We want to know if there is an association between the amount of time spent volunteering and volunteers’ happiness. We collect the following data:
Food Bank | Monthly volunteer hours | Generally very happy? |
Downtown | 21 | No |
Downtown | 41 | No |
Downtown | 56 | Yes |
Downtown | 60 | Yes |
Downtown | 48 | Yes |
Downtown | 18 | No |
Downtown | 36 | No |
Midtown | 14 | Yes |
Midtown | 12 | Yes |
Midtown | 8 | No |
Midtown | 10 | No |
Midtown | 38 | Yes |
Midtown | 46 | Yes |
Midtown | 12 | No |
Uptown | 6 | Yes |
Uptown | 7 | Yes |
Uptown | 12 | Yes |
Uptown | 15 | Yes |
Uptown | 1 | No |
Uptown | 5 | No |
Uptown | 24 | Yes |
We can summarize our data at the group level like this:
Food bank | Average monthly hours per volunteer | Generally very happy |
Downtown | 40 | 43% |
Midtown | 20 | 57% |
Uptown | 10 | 71% |
When our unit of analysis is the food bank, we see a negative association between the average monthly hours per volunteer and a food bank’s percentage of generally happy volunteers. We may be tempted to apply this finding at the individual level, concluding that people who volunteer more are less happy. What happens, though, when we conduct our analysis at the level of the individual? Let’s compare the average hours worked by generally happy volunteers to the average hours worked by their less happy peers—so, still looking for a relationship between the independent and dependent variables, but this time without first grouping our cases:
Generally very happy | Not generally very happy |
28 | 17 |
Now, we see a positive association between time spent volunteering and general happiness; volunteers who describe themselves as generally happy volunteer, on average, more per month than everyone else. That’s the opposite conclusion we had reached before! The difference? Before, we applied a finding about the relationship between two variables at the group level to individuals—we posited an ecological fallacy.
Social researchers are drawn to making group comparisons because they often reveal interesting patterns in our social world. We all base a lot of our own self-identities in our group memberships, so it can be easy to embrace results that confirm our biases about other groups or confirm our own positive self-perceptions, even if the results reflect an ecological fallacy.