Results of Poisson and binomial logistic regression analyses to detect an association between positive and negative emotive GRPs and tobacco use
| Outcome | Final models* | Linear term† OR (95% CI) | Smooth term EDF‡ | p Value§ |
|---|---|---|---|---|
| Average consumption | Negative emotive GRPs 1 month earlier | 1.65 | 0.016 | |
| Smoking prevalence | Positive emotive GRPs 1 month earlier | 0.93 (0.87 to 0.98) | 0.01 | |
| Negative emotive GRPs 2 months earlier | 0.96 (0.92 to 0.999) | 0.04 |
*All regression models were adjusted for cigarette costliness, tobacco control score, number of adults in the household, gender, age, income, social class, education, employment status and government office region of residence. GRPs for each campaign type at different lags were initially considered as smooth terms and we used backwards selection of the GRPs to find the best model. Any GRP term in the final model found to be linear (EDF=1) was replaced with a linear term.
†ORs and 95% CIs reported for smoking prevalence associated with a 400-point increase in GRPs.
‡The effective degrees of freedom (EDF) is a measure of how ‘wiggly’ the term is (ie, EDF=1 corresponds to a straight line, ie, a linear effect).
§p Value from a t test on the parametric regression coefficients and F test on smooth terms.
EDF, effective degrees of freedom.