Making fun of assumptions economists make is an enjoyable past-time. But we shouldn’t forget that it is the immense difficulty in understanding the complex economic system that leads to these often outrageous (sometimes useful) simplifying assumptions. This same difficulty has lead serious economists to be extremely cautious about the appropriate use of statistical methods and their interpretations.
This caution, and deep understanding, is what makes the empirical approaches of economics so valuable to the social sciences, especially in the interpretation of statistical results as potential policy advice.
Consider recent research published about the impact of TV watching on life expectancy. The widely publicised conclusion was “Every single hour of television watched after the age of 25 reduces the viewer’s life expectancy by 21.8 minutes”
That is a basic interpretation assuming that the authors correctly control for other variables. It is not clear from the actual study which variables are controlled for, and why, and it is fairly clear that no other socio-economic factors besides hours of TV watching are in their model at all.
In addition, the authors explicitly allow TV watching to capture all variation (after controlling for known health factors) in mortality compared to a population that watches no TV.
What sort of policy advice arises from this result? Would we expect a public campaign to reduce TV watching (using TV!), to result in people living longer? This study provides no answers to these questions.
Indeed, one of the fundamental problems with studies of health effects of different activities is that they ignore opportunity costs. If the TV watching cohort in the study were not watching TV, would the other activity that fills their time be better or worse for their health? Might they instead be down the local burger joint before heading to the pub with a fag in their mouths?
This very problem prompted some economists to try and better analyse data on television watching and developmental outcomes in children (amongst other activities, such as computer games). Their study uses data the Longitudinal Study of Australian Children (LSAC) biannual survey which has complete time use data collected from parents (see the forms used in the appendix).
While the authors have done their best to read something into their results and get their paper published, the really important result is from their use of a fixed effects model to control for individual child variation. There is not much use in a result that says smarter kids like to do activities they are better at, such as writing, reading and so forth. It is worth knowing that if an individual child goes from low to high TV watching that their developmental outcomes suffer (or vice-versa).
And the fixed effect model shows that there is almost no significant correlation between time spent on any activity, from computer games, TV, school and activities with parents, and their developmental outcomes.
The only statistically significant correlation appears to be that more schooling provides children with a better vocabulary.
An irrational attachment to TV watching as a health issue, while ignoring substitute sedentary activities such as reading books (or sitting at the bar), doesn’t provide good policy advice, or even health advice generally. Better advice would focus on how to ensure sufficient physical activity is undertaken even with the spread of sedentary working conditions.
Having a deep understanding of the appropriate use of statistical models helps economists play a role in translating health research, and research from other fields in general, into actionable policy advice. For example, in cost-benefit analysis of urban transport investments, factoring the health benefits of substitution of time between sedentary forms of commuting would provide better policy analysis.
Tips, suggestions, comments and requests to [email protected] + follow me on Twitter @rumplestatskin