|
Abstract:
|
Self-reported daily cigarette count data typically show large "heaps" at multiples of 20 cigarettes, the number of cigarettes in a pack in the United States. Evidence suggests smokers actually are consuming a variable number of cigarettes, which some then report rounded off to the nearest number of packs. Moreover, the propensity to report data in this heaped fashion appears to increase with the number of cigarettes smoked. Such heaping in the data generation process can cause profound biases if the data are taken at face value. We describe models for estimating the underlying cigarette count distributions from heaped data, which we fit using Bayesian methods to data from a smoking-cessation drug trial.
|
