Statistical analysis

Akin to Chalak et al,⁹ we used a ZIP regression model to calculate the own-rice and cross-price elasticities of demand for tobacco products for each country. The ZIP model is suitable for count data with excess zeros (non-smokers in our case). The ZIP model was estimated first for each gender group for each of the seven tobacco products. Then, the model was estimated for each gender for each tobacco product variety. Tobacco products were classified into three groups such that the within-group unit of measurement was homogeneous. These groups were (1) cigarettes, which included premium and discount cigarettes—measured in packs; (2) waterpipe tobacco (250 g), which included premium and discount waterpipe tobacco; and (3) waterpipe sessions, which included delivered waterpipe and discount and premium waterpipe café smoking sessions. For each group, the quantity was calculated as the sum of all quantities reported by each respondent at each set of prices. The price of each product variety was calculated using the Stone Index.²⁹ For each product variety, j, the Stone Index, PjS, is defined as

where M is the total quantity within each product variety, pm is the price of tobacco product m=1,…,M, and sm is the expenditure weight of product m such that ∑m=1Msm=1. The Stone Index is a weighted geometric mean of the price of tobacco products. Some respondents reported zero quantities at certain price levels. So calculating sm for each respondent at each scenario of prices yielded many null weighted prices. We therefore calculated sm as the average weight of all respondents for all price scenarios for each tobacco product variety m.

Two binary variables were constructed for the regression analysis used to estimate the price elasticities of demand. The first equaled one if the individual smoked any type of cigarettes, while the second equaled one if the individual smoked any type of waterpipe tobacco. The former binary variable was used as an inflation variable in the ZIP models of cigarette products, while the latter binary variable was used in the ZIP models of waterpipe products. For all models, the set of independent variables of the first part model included the logarithmic forms of the prices of all tobacco product varieties and a categorical variable of four income groups. The inclusion of the prices of all tobacco products allowed for calculating the cross-price elasticities between all tobacco products. In general, the set of independent variables of the inflation part of the ZIP model included the relevant binary variable of tobacco smoking, age (categorical variable), marital status (binary variable which takes one if the individual is single), a region variable (country-specific), employment status (unemployed, full-time and part-time employee), education level (less than university and with university education), and a binary variable that equaled one if the individual was responsible for taking decisions in the household. For some tobacco product varieties across genders, only a subset of these independent variables was used due to the small sample size of smokers. All analyses were conducted using Stata V.14.2.