Longitudinal versus cross sectional data in economic experiments: The controversial example of public smoking bans and weight gain

Many countries have recently introduced public smoking bans. These have been advocated by policy makers as a method for reducing population level smoking, and reducing the potential damage to health from second hand smoke. One problem with any intervention that affects lifestyles is that they may have unintended consequences for other outcomes, such as weight gain and consequently obesity. This paper investigates whether smoking bans have any impact on body mass index (BMI).

By exploiting smoking bans in Australia and Great Britain as natural experiments we are able to identify the causal factors running from the bans to BMI. We find results using cross sectional data suggest that the bans have increased body mass index (BMI) in both countries. However, if unobservable heterogeneity is controlled for using longitudinal data it is shown that BMI has reduced following the introduction of the bans.

These reductions have been largest for men, although there is evidence of significant effects for women who smoke. These are important new findings, with implications concerning the robustness of the use of cross sectional data for investigations of this nature, spurious results may be produced.  

Cross sectional data may imply that smoking bans have led to an increase in BMI.  Using more robust longitudinal data  point towards a general healthier lifestyle encouraged by smoking bans - drinking and eating behaviour may have changed as a consequence.  Given general increases in BMI, it is important to note that smoking bans have not contributed to this.  Analyses using longitudinal data in the way we propose highlights important differences in economic experiments of this nature, and results from use of cross sectional data should be treated with appropriate caution.