We used instrumental variable methods to
assess possible endogeneity of smoking in regression
models predicting new-onset mental
disorders.29---31 Instrumental variables are variables
that are associated with the exposure of
interest (in this case smoking), but are only
associated with the outcome of interest (newonset
mental disorders) through their association
with the exposure variable. The instrumental
variables we chose for our study were
state cigarette taxes and public attitudes toward
smoking. Strong evidence exists for an association
between these variables and initiation of
smoking,38,39 and it is implausible that state
cigarette taxes or public attitudes toward
smoking would directly affect incidence of
mood or anxiety disorders (Appendix A, available
as a supplement to the online version of
this article at http://www.ajph.org, provides
a more detailed description of the instrumental
variables and results of the analyses for testing
the validity of the instruments). We obtained
data on state cigarette taxes for 2001and 2002
collected and published by the Tax Policy
Center.40
Our data on attitudes toward smoking were
collected as part of the National Survey on
Drug Use and Health and aggregated for each
state.41 We derived our data on attitudes from
the following question that this survey asked of
all participants: “How much do people risk
harming themselves physically and in other
ways when they smoke one or more packs
of cigarettes per day?” Participants who
responded “great risk” were rated as having
a negative attitude toward smoking.
State taxes averaged over 2001 and 2002
ranged from 2.8 cents to $1.31 across states.
Prevalence of negative attitudes toward
smoking ranged from 64.4% to 72.2%. We
linked data on instrumental variables to the
NESARC data according to the state of
residence of the NESARC participants at
baseline.
Analyses
We conducted analyses in 3 stages. First, we
explored the possible endogeneity of regular
smoking in regression models predicting new
onset of mood and anxiety disorders through
a series of instrumental variable probit analyses.
42 Instrumental variable methods allowed
us to assess endogeneity (i.e., an association