ISA: Stata module to perform Imbens' (2003) sensitivity analysis

To read the full-text of this research, you can request a copy directly from the author.


isa produces a figure for the sensitivity analysis developed by Imbens (American Economic Review, 2003). Observational studies cannot control for the bias due to the omission of unobservables. The sensitivity analysis provides a benchmark about how strong assumption about unobservables researchers need to make to maintain the causal interpretation.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the author.

... Imbens initially designed his sensitivity analysis for cases with continuous outcomes and binary treatments, but Harada (2012) has developed a Stata module that generalizes Imbens' approach to allow for binary outcomes and continuous treatments. This generalized sensitivity analysis specifies two equations, one of which refers to how the unobserved factor affects the treatment and the other to how it affects the outcome. ...
... We use these predictors as the benchmarks because: (1) theory and prior research has indicated that selfcontrol confounds the effects on delinquency for both peer influence and school commitment (Gottfredson and Hirschi 1990); (2) as demonstrated below, self-control and prior delinquency are consistently the predictors that are closest to the estimated contour lines and; (3) Imbens (2003) has explicitly stated that contour line surpassing the prior delinquency term provides evidence of a robust effect. We estimate these models using Stata's generalized sensitivity analysis ("gsa") command (Harada 2012). 10 We estimate robust where y is the outcome, w is the treatment, x is the vector of controls, tstat(1.96) ...
Full-text available
Objective Criminologists have long questioned how fragile our statistical inferences are to unobserved bias when testing criminological theories. This study demonstrates that sensitivity analyses offer a statistical approach to help assess such concerns with two empirical examples—delinquent peer influence and school commitment. Methods Data from the Gang Resistance Education and Training are used with models that: (1) account for theoretically-relevant controls; (2) incorporate lagged dependent variables and; (3) account for fixed-effects. We use generalized sensitivity analysis (Harada in ISA: Stata module to perform Imbens’ (2003) sensitivity analysis, 2012; Imbens in Am Econ Rev 93(2):126–132, 2003) to estimate the size of unobserved heterogeneity necessary to render delinquent peer influence and school commitment statistically non-significant and substantively weak and compare these estimates to covariates in order to gauge the likely existence of such bias. ResultsUnobserved bias would need to be unreasonably large to render the peer effect statistically non-significant for violence and substance use, though less so to reduce it to a weak effect. The observed effect of school commitment on delinquency is much more fragile to unobserved heterogeneity. Conclusion Questions over the sensitivity of inferences plague criminology. This paper demonstrates the utility of sensitivity analysis for criminological theory testing in determining the robustness of estimated effects.
... Similar to the other two approaches, we have built software that will be available to researchers; ours is implemented in the freely available R software package. 7 isa (Harada, 2011) performs the sensitivity analysis of Imbens (2003) in Stata. 8 gsa (Harada, 2012) performs the sensitivity analysis of Harada (2013) in Stata. ...
A major obstacle to developing evidenced-based policy is the difficulty of implementing randomized experiments to answer all causal questions of interest. When using a nonexperimental study, it is critical to assess how much the results could be affected by unmeasured confounding. We present a set of graphical and numeric tools to explore the sensitivity of causal estimates to the presence of an unmeasured confounder. We characterize the confounder through two parameters that describe the relationships between (a) the confounder and the treatment assignment and (b) the confounder and the outcome variable. Our approach has two primary advantages over similar approaches that are currently implemented in standard software. First, it can be applied to both continuous and binary treatment variables. Second, our method for binary treatment variables allows the researcher to specify three possible estimands (average treatment effect, effect of the treatment on the treated, effect of the treatment on the controls). These options are all implemented in an R package called treatSens. We demonstrate the efficacy of the method through simulations. We illustrate its potential usefulness in practice in the context of two policy applications.
ResearchGate has not been able to resolve any references for this publication.