An empirical analysis of milk addiction.

We show the estimable rational addiction model tends to yield spurious evidence in favor of the rational addiction hypothesis when aggregate data are used. Direct application of the canonical model yields results seemingly indicative that non-addictive commodities such as milk, eggs, and oranges are rationally addictive. Monte Carlo simulation demonstrates that such results are likely to obtain whenever the commodity under scrutiny exhibits high serial correlation, or when even a small amount of the variation in prices is endogenous, or when overidentified instrumental variables estimators are used, or when commonly imposed restrictions are employed. We conclude that time-series data will often be insufficient to differentiate rational addiction from serial correlation in the consumption series.