Estimating Distributions of Counterfactuals with an Application ... - UCL
Estimating Distributions of Counterfactuals with an Application ... - UCL
Estimating Distributions of Counterfactuals with an Application ... - UCL
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
408 CARNEIRO, HANSEN, AND HECKMAN<br />
Let ̂m c = t 1 − µ c m (̂x) to obtain<br />
Pr (M c ≤ ̂m c , M d = (0,...,0) | X = ̂x) = F U c m ,Ũ d m (t 1, t 2 )<br />
We know the left-h<strong>an</strong>d side <strong>an</strong>d thus identify F U c m ,Ũ at the evaluation point t 1, t<br />
m d 2 .<br />
Since (t 1 , t 2 ) is <strong>an</strong>y arbitrary evaluation point in the support <strong>of</strong> Um c , Ũd m we c<strong>an</strong> thus<br />
identify the full joint distribution.<br />
<br />
PROOF OF THEOREM 2.<br />
(<br />
c1 (Q 1 ) − ϕ (Z)<br />
Pr (D 1 = 1 | Z, Q 1 ) = Pr<br />
> ε )<br />
W<br />
σ W σ W<br />
Under (A-1), (A-2), (A-6), (A-7), <strong>an</strong>d (A-9), it follows that c 1(Q 1 ) − ϕ(Z)<br />
σ W<br />
<strong>an</strong>d F˜εW<br />
(where ˜ε W = ε W<br />
σW<br />
) are identified (see M<strong>an</strong>ski, 1988; Matzkin, 1992, 1993). Under<br />
r<strong>an</strong>k condition (A-7), identification <strong>of</strong> c 1(Q 1 ) − ϕ(Z )<br />
σ W<br />
implies identification <strong>of</strong> c 1(Q 1 )<br />
σ W<br />
<strong>an</strong>d ϕ(Z )<br />
σ W<br />
separately. Write<br />
( ) ( )<br />
c2 (Q 2 ) − ϕ (Z) c1 (Q 1 ) − ϕ (Z)<br />
Pr (D 2 = 1 | Z, Q 1, Q 2 ) = F˜εW − F˜εW<br />
σ W σ W<br />
From the absolute continuity <strong>of</strong>˜ε W <strong>an</strong>d the assumption that the distribution function<br />
<strong>of</strong>˜ε W is strictly increasing, we c<strong>an</strong> write<br />
c 2 (Q 2 )<br />
σ W<br />
[<br />
( )]<br />
= F −1<br />
c1 (Q 1 ) − ϕ (Z)<br />
˜ε W<br />
Pr (D 2 = 1 | Z, Q 1 , Q 2 ) + F˜εW + ϕ (Z)<br />
σ W σ W<br />
Thus, we c<strong>an</strong> identify c 2(Q 2)<br />
σ W<br />
over its support <strong>an</strong>d, proceeding sequentially, we c<strong>an</strong><br />
identify c s(Q s)<br />
σ W<br />
, s = 3,...,S. Under (A-8) we c<strong>an</strong> identify η s , s = 2,...,S. <br />
Observe that we could use the final choice (Pr(s = S)) rather th<strong>an</strong> the initial<br />
choice to start <strong>of</strong>f the pro<strong>of</strong> <strong>of</strong> identification using <strong>an</strong> obvious ch<strong>an</strong>ge in the assumptions.<br />
PROOF OF THEOREM 3. From (A-2), the unobservables are jointly independent <strong>of</strong><br />
(X, Z, Q). For fixed values <strong>of</strong> (Z, Q s , Q s−1 ), we may vary the points <strong>of</strong> evaluation<br />
for the continuous coordinates (ys c ) <strong>an</strong>d pick alternative values <strong>of</strong> X = ̂x to trace<br />
out the vector µ c (X) up to intercept terms. Thus, we c<strong>an</strong> identify µ c s,l<br />
(X) upto<br />
a const<strong>an</strong>t for all l = 1,...,N c,s (Heckm<strong>an</strong> <strong>an</strong>d Honoré, 1990). Under (A-2), we<br />
recover the same functions for whatever values <strong>of</strong> Z, Q s , Q s−1 are prespecified<br />
as long as c s (Q s ) > c s−1 (Q s−1 ), so that there is <strong>an</strong> interval <strong>of</strong> ε W bounded above<br />
<strong>an</strong>d below <strong>with</strong> positive probability. This identification result does not require <strong>an</strong>y<br />
passage to a limit argument.