21 February 2009
I met one of my friends on basketball court. This is selection. I select him as my friend because he plays good basketball and is an avid player. We have been friends for almost three years. When either of us wants to play, most times we will call each other and meet on the court. I think without knowing him, I will still play basketball, but not that many times. So we influence each other. Sometimes we eat Vietnamese noodles together at Le's right after game. Contextual factors matter, but it is him who makes me eat more times of noodles than I would have by myself. Probably, our friendship has some impacts on both of our weights and may make them change more synchronously. Similarly, if you are a runner, you will surely like running with your friends and may run more because you get a runner as friend. So the empirical question is whether you indeed play more basketball when you get a friend who likes playing basketball and run more if you get a runner friend. It is also possible that because you play more or run more, you eat more, which offsets the weight loss due to those extra exercises.
Given only observational data, it is hard to disentangle the effects of selection, induction and contextual factors on weight changes. We have to assign you friends (roommates) randomly and check if you and your friends gain/lose weight together, possibly because you two play more basketball, run more, eat similar things, have similar living styles, share similar standards about what consists of a normal weight, etc.
It is interesting to see that the effects of friendship seem to be directional or asymmetric. Only people you think as friend can induce you to lose weight. You can not induce a person who does not think you are his friend to lose weight, although you think he is your friend. This is kind of opinion leader effect.
The directionality of friendship effects also counters the challenging of contextual factors hypothesis, because if contextual factors matter, you would expect friends' weight changes correlate without directionality. Also, if they matter, you would expect your neighbors' weight changes synchronize with yours and the weight of your friend who lives hundreds of miles away should not correlate with yours. But neither is corroborated by data.
Hence selection should be the largest concern in this case. Now the questions are whether using weight changes or obese status changes will remove the selection effect and how we could control it better.
One of my friends told me two weeks ago that, he did not buy the points in "The Spread of Obesity in a Large Social Network over 32 Years" until he read the real paper. I confessed, "Same here." Read the real paper, not the popular press. But you are absolutely not obligated to buy the points. Here are more.
P.s. My friend and I have successfully induced several of our friends who originally do not play basketball to play more. But hopefully they can gain some weight rather than losing weight so that we can play more strongly and better.