3.2. Variable definitions
3.2.1. Risk perception variables
The main survey asked respondents to indicate the strength of
the relationship between smoking and the probability of getting
specific diseases. The response categories were: (1) definitely does
not increase; (2) probably does not increase; (3) slightly increases;
(4) probably increases; and (5) definitely increases. The main survey
inquired about the link between smoking and these smokingrelated
diseases: lung cancer; lung diseases (pneumonia, chronic
obstructive pulmonary disease (COPD)); heart disease; and stroke.
In our basic specification, we measure risk perceptions in three
different ways. First, we define a binary dependent variable, set
to 1 if the person answered “probably increases” or “definitely
increases” the risk of getting lung cancer and otherwise set to 0.
Second, in an alternative specification, the choice of a category
from the five-point scale for lung cancer (see Fig. 1) is the dependent
variable. Third, we define binary variables with “probably” and
“definitely increases” set to 1 and other responses set to 0 for each
of the four smoking-related diseases. The “four-count” variable has
values from 0 to 4, 4 if the respondent believed that smoking “defi-
nitely increases” the probability of getting each ofthe four diseases.
3.2.2. Smoking behavior variables
We define four dependent variables for smoking behavior:
binary variables for whether the man: (1) quit smoking in the two
years following the lung cancer diagnosis; (2) attempted to quit in
the past year conditional on not actually having quit; (3) planned
to quit in the next six months; and (4) planned to quit in the future
but not in the next six months given that he did not plan to quit in
the next six months. The follow-up period for the third and fourth
dependent variables starts at the date of the main survey.