A two stage cluster sampling design was employed to select the initial sample. Within each TV region a random sample of enumera- tion districts19 were selected (1 in 40) and within each district a random sample of households were selected (1 in 20) using com- puter generated lists. Interviewers then visited the selected households and one resident member was selected for the sample using a pseudo-random Kish-grid method.20 Only adults, aged 16 or older, were sampled. Inter- viewers and respondents were unaware of the forthcoming TV campaign or of the intention to conduct follow up interviews. The decision not to tell respondents that they were taking part in a study risked higher non-contact and refusal rates at follow up than might be expected under a formal clinical trial protocol, but minimised the likelihood of high rates of deception about smoking status and follow up attrition related to the intensity of the intervention. The follow up interviews were conducted by a diVerent group of field workers who were unaware of the pre-campaign responses or the relationship between the interviews and the interventions. They were instructed to make five attempts to re-interview the original respondent, failing