mhbounds
outmatj containing the strata-specific test statistics, j = 1, ...,#strata.
4.1 Typical Examples
1. Running mhbounds after psamtch2
• psmatch2 college, outcome(wage) pscore(pscore) caliper(.25) common nore-
placement
• mhbounds wage, gamma(1 (0.05) 2) [performs sensitivity analysis at Gamma
= 1,1.05,1.10,...,2.]
2. Running mhbounds with user-defined treatment-, weight- and support-indicators
• mhbounds outcome, gamma(1 (0.05) 2) treated(mytreat) weight(myweight)
support(mysupport)
3. Running mhbounds with user-defined treatment-, weight- and support-indicators
with different strata in the population
• mhbounds outcome, gamma(1 (0.05) 2) treated(mytreat) weight(myweight)
support(mysupport) stratum(mystratum) stratamat
Please note that mhbounds is suited for k-nearest neighbor matching without re-
placement and for stratification matching.
5 Examples
To illustrate mhbounds we give two examples, where the first one is taken from the book
of Rosenbaum (2002) and the second one relates to the well known and much discussed
studies by Lalonde (1986), Dehejia and Wahba (1999) and Smith and Todd (2005).
5.1 Rosenbaum Example
The first example is given in Rosenbaum (2002, Table 4.11, p. 130) and comes from a
medical study of the possible effects of the drug allopurinol as a cause of rash (Boston
Collaborative Drug Surveillance Program, 1972). The treatment in this case is the use
of the drug (D ∈ {0, 1}) and the binary outcome variable is to have a rash or not
(Y ∈ {0, 1}). Table 1 summarises the available data from a case-referent study, where
treated and control group are already comparable and we distinguish two strata of the
population (S = 1 for males and S = 2 for females).
A first look at the distribution of outcomes between treated and control units would
suggest that the treatment in fact has a positive effect on the outcome variable, since,
e.g. 33 ≈ 15% of the treated males have an outcome of 1 whereas this is true for only
6465 ≈ 6% of the control individuals. In order to replicate the example we generate a
sample of individuals according to the distribution of D and Y in Table 1.