In theory

Blogger doesn't really do a good job of allowing me to write mathematical notation. If it did, I would share some of the interesting things I've been learning the past two days. I'm in San Antonio at the Maternal and Child Health Epidemiology Training on Multilevel Modeling. And basically what we're learning is, as an example - does a certain intervention have an effect on behavior when looking at the data as being "nested" within some cluster. What?

Take for example a smoking prevention curriculum given to students in Hawaii. Some classes are randomly assigned to be given the curriculum, and some are not. We give them all a pre-test on their tobacco knowledge and attitudes, then we give the curriculum to those assigned to receive it, then we test everyone again. Then, typically, we look at the scores and ask: is there a difference in post-test scores between students who got the intervention and those who did not?

Multilevel modeling asks, however, is there an effect of the school or the county that the students are in? This is called nesting. Students are nested in classrooms; classrooms are nested in schools; schools are nested in counties. Does this have an effect on the scores of the students? Sure this isn’t anything earth shattering. We all know that kids in certain school districts don’t do well at certain things because their programs have less funding from which to draw resources.

And though we often talk about this in the public health world – calling it the “social determinants of disease” (where disease could have a lot of meanings – in this case, low knowledge of ill effects of tobacco use) – we haven’t have the mathematical techniques to deal with these hypotheses until the past 10-15 years. So, these past two days have been learning some of these advanced techniques.

I certainly learned some funky statistics, but really what I’ve come to after these two days of training is: Is any of my research valid? And furthermore, how much is out in the literature that is invalid? Sometimes I really wonder about my field, and how we ever come to any conclusions about anything. A lot of what we do is try to account for variability in data using fancy math. We try to explain human behavior with statistical models. But human behavior is so bizarre and hard to account for. It’s true, we’ve made some good approximations that have helped reduce some disease burden, but when it comes to social disease (poverty, violence, obesity), I feel like we’re grasping for straws. Of course, this is not a reason to stop doing what we do. It’s just that I feel frustrated with the limitations on my own ability to make concrete conclusions and affect any kind of real change.

I suppose it would be worse if I was a theoretical statistician.