19.05.2014 Views

Testing Gateway Theory: do cigarette prices affect illicit drug use?

Testing Gateway Theory: do cigarette prices affect illicit drug use?

Testing Gateway Theory: do cigarette prices affect illicit drug use?

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

682 M. Beenstock, G. Rahav / Journal of Health Economics 21 (2002) 679–698<br />

The solution to this problem is to instrument C t−1 in Eq. (1). No <strong>do</strong>ubt Z and X largely<br />

overlap. An element of Z is the price of <strong>cigarette</strong>s (and other variables discussed in Section 4)<br />

that prevailed when individual n was growing up (P n(t−1) ), which are hypothesized to<br />

determine the demand for <strong>cigarette</strong>s but not the demand for cannabis. We expect C t−1 to<br />

vary inversely with P t−1 . There is no reason to suspect that P t−1 and v t−1 are correlated<br />

(nor P t−1 and u t ). If S t <strong>do</strong>es not depend upon P t−1 , then P t−1 identifies the effect of C t−1<br />

upon S t in Eq. (1).<br />

The next step in the gateway chain concerns the relationship between cannabis and<br />

hard <strong>drug</strong>s. The hypothesis of interest here is <strong>do</strong>es exposure to cannabis induce a higher<br />

probability of subsequently using hard <strong>drug</strong>s? We denote H n = 1 if individual n <strong>use</strong>d hard<br />

<strong>drug</strong>s after using cannabis and 0 otherwise. We express this as follows:<br />

H n(t+1) = λY n(t+1) + δS nt + ρ y D y + w n(t+1) (3)<br />

where Y is a vector of controls for hard <strong>drug</strong> <strong>use</strong>. <strong>Gateway</strong> <strong>Theory</strong> suggests δ>0. Since<br />

we suspect that E(wu) >0 unobserved heterogeneity is likely to bias upwards estimates<br />

of δ. Ideally, we seek exclusion restrictions for Y and X such as <strong>drug</strong> price data, which are<br />

not available in Israel. Instead, we suggest using as an instrument for S t in Eq. (3) its fitted<br />

value from IV estimation of Eq. (1). We refer to this as “<strong>do</strong>mino” identification beca<strong>use</strong><br />

the natural experiment is indirect. Clearly, such indirect identification weakens the power<br />

of tests on the value of δ. Ran<strong>do</strong>m exposure to <strong>cigarette</strong>s at time t − 1 induces cannabis<br />

consumption at time t, which in turn, induces hard <strong>drug</strong> <strong>use</strong> at time t + 1. If β and δ are<br />

properly identified, and there is a causal gateway effect, we can expect a chain reaction<br />

where raising the price of <strong>cigarette</strong>s will reduce <strong>cigarette</strong> smoking, which will reduce the<br />

subsequent <strong>use</strong> of cannabis, which, in turn, will reduce the subsequent <strong>use</strong> of hard <strong>drug</strong>s.<br />

Note that Eqs. (1)–(3) control for cohort effects. This is possible beca<strong>use</strong> we <strong>use</strong> several<br />

surveys in Section 4 (y = year of survey − age). Had there been only one survey it would<br />

not have been possible to identify the separate effects of P t−1 and birth cohort upon <strong>drug</strong><br />

consumption.<br />

2.2. Recursive bivariate probit and two-stage procedures<br />

If u and v in Eqs. (1) and (2) happen to be bivariate normal with cov(u, v) = ρ then<br />

the model may be estimated by maximum likelihood as a recursive bivariate probit (RBP)<br />

model (Maddala, 1983, pp. 122–123, Greene, 2000, pp. 852–825). Note that in this case C in<br />

Eq. (1) <strong>do</strong>es not have to be instrumented beca<strong>use</strong>, in contrast to the linear probability model,<br />

it is not estimated by least squares. However, Maddala points out that identification requires<br />

that X omit covariates in Z.IfX = Z the model is not identified, even parametrically.<br />

Maddala suggests that an alternative procedure is to specify prob (C ∗ > 0) in Eq. (1)<br />

instead of C, where C ∗ denotes the underlying latent variable that measures the propensity<br />

to smoke <strong>cigarette</strong>s. In this case, it may be shown that a two-stage procedure provides<br />

consistent estimates of the parameters in Eqs. (1) and (2). In the first stage, Eq. (2) is<br />

estimated by probit or logit, and in the second stage the predicted probability of C obtained<br />

from the first stage is <strong>use</strong>d to replace C in Eq. (1).<br />

There are advantages and disadvantages to both procedures. The main disadvantage of<br />

RBP is that it may be sensitive to parametric assumptions about the unobserved heterogeneity.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!