Abstract
BACKGROUND: It has been known that smoking is negatively related to weight-related outcomes. However, it has been difficult to determine whether the relationship is causal, and if so, how strong it is. We attempted to estimate the approximately causal effects of smoking on weight, body mass index (BMI), and obesity. METHODS: The Indonesian Family Life Survey provided a sample of over 9000 men aged 15-55 years-each of them was observed in 1993, 1997, 2000 and 2007. The preferred method was a fixed effects model; that is, we related changes in smoking status or smoking intensity to changes in weight-related outcomes, while controlling for time-varying covariates. We also compared these results to those estimated by ordinary least squares and assessed the importance of controlling for time invariant individual heterogeneity. RESULTS: Although the effects of smoking were precisely estimated in a statistical sense, their size was minuscule: a quitter would gain weight by at most 1 kg, or a smoker would lose weight by the same amount. The results were similar for BMI and obesity. When we did not control for time invariant individual heterogeneity, the size of the relationship was overestimated at least three times. CONCLUSIONS: Smoking exerted little influence on weight, and it was important to control for bias stemming from time invariant individual heterogeneity.