{smcl}
{* *! version 1.0.2 18Jan2012}{...}
{cmd:help isa}
{hline}
{title:Title}
{p2colset 9 19 21 2}{...}
{p2col: {cmd:isa} {hline 2}}Imbens' (2003) sensitivity analysis{p_end}
{p2colreset}{...}
{title:Syntax}
{p 8 15 2}
{cmd:isa}
{depvar}
{it:treatmentvar}
[{indepvars}]
{ifin}
{weight}
{cmd:,} {opt tau(real)} [{it:options}]
{synoptset 26 tabbed}{...}
{synopthdr}
{synoptline}
{syntab:Model}
{p2coldent:* {opth tau(real)}}the target size of the coefficient of the treatment variable.{p_end}
{synopt:{opth obs:ervation(int)}}the number of observations; default is {cmd: observation(20)}{p_end}
{synopt:{opth res:olution(int)}}the number of iterations for each fixed value of alpha; default is {cmd: resolution(20)}{p_end}
{synopt:{opth mina:lpha(real)}}the minimum value of alpha in the iterations; default is {cmd: minalpha(0)}{p_end}
{synopt:{opth maxa:lpha(real)}}the maximum value of alpha in the iterations; default is {cmd: maxalpha(10)}{p_end}
{synopt:{opth mind:elta(real)}}the minimum value of delta in the iterations; default is {cmd: mindelta(0)}{p_end}
{synopt:{opth maxd:elta(real)}}the maximum value of delta in the iterations; default is {cmd: maxdelta(5)}{p_end}
{synopt:{opt nodots}}suppresses display of iteration dots.{p_end}
{syntab:SE/Robust}
{synopt :{opth vce(vcetype)}}{it:vcetype} may be {opt oim},
{opt r:obust}, {opt cl:uster} {it:clustvar}, or {opt opg}{p_end}
{syntab:Graph}
{synopt:{opt nograph}}suppresses the figure{p_end}
{synopt:{opt noplot}}suppresses the plots of the partial R-sq of each control variable in the figure {p_end}
{synopt:{opth nplots(int)}}specifies the number of control variables plotted on the graph; default is {cmd: nplots(5)}. The first {it:n} variables in {it:indepvars} are selected.{p_end}
{syntab:Advanced}
{synopt:{opth prec:ision(real)}}the percentage error from {opt tau(real)} within which iteration ends; default is {cmd: precision(0.1)}{p_end}
{synopt:{opth smina:lpha(real)}}the minimum value of alpha in the 2nd run of the iterations; default is {cmd: aminalpha(0)}{p_end}
{synopt:{opth smaxa:lpha(real)}}the maximum value of alpha in the 2nd run of the iterations; default is {cmd: amaxalpha(10)}{p_end}
{synopt:{opth smind:elta(real)}}the minimum value of delta in the 2nd run of the iterations; default is {cmd: amindelta(0)}{p_end}
{synopt:{opth smaxd:elta(real)}}the maximum value of delta in the 2nd run of the iterations; default is {cmd: amaxdelta(5)}{p_end}
{synopt:{opth ml_iterate(int)}}specifies the number of iterations in each MLE; default is {cmd: ml_iterate(5)}{p_end}
{synopt:{opt skipr:angecheck}}skips the check of initial values of minalpha, minalpha, minalpha and minalpha.{p_end}
{synopt:{opt showr:angecheck}}shows the results of the rangecheck performed in Step 2.{p_end}
{synopt:{opt quick}}directs -{cmd: isa}- to switch the role of {opt alpha} and {opt delta} at an earlier point than default.{p_end}
{synopt:{opt tstat(real)}}the target size of the t-value of the treatment variable.{p_end}
{synoptline}
{p2colreset}{...}
{p 4 6 2}* {opt tau(real)} is required.{p_end}
{p 4 6 2}* The value of {opt tau(real)} must be positive and smaller than the original treatment effect.{p_end}
{p 4 6 2}At least one {indepvars} is required.{p_end}
{p 4 6 2}{it:treatmentvar} must be binary.{p_end}
{p 4 6 2}{cmd:aweight}s and {cmd:fweight}s are allowed; see {help weight}.{p_end}
{p 4 6 2}{cmd:aweight} in -{cmd:logit}- is no longer supported in the current version of Stata. Thus, -{cmd:isa}- uses -{cmd:logit}- in version 9.{p_end}
{title:Description}
{pstd}
-{cmd:isa}- produces a figure for the sensitivity analysis developed by Imbens (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.
{pstd}
Specifically, Imbens (2003) set up the situation in which researchers can manipulate the magnitude of the bias by changing the coefficient of an unobservable on a treatment variable ({opt alpha}) and that on an outcome variable ({opt delta}).
Thus, there exist a number of pairs of {opt alpha} and {opt delta} that change the treatment effect by a given size.
In principle, Imbens' (2003) sensitivity analysis presents the contour of an equal bias having {opt alpha} in the horizontal axis and {opt delta} in the vertical axis.
{pstd}
-{cmd:isa}- automatically maximizes the likelihood function developed by Imbens (2003), find such pairs with an efficient algorithm, and draw a contour by connecting these pairs.
How to interpret the resulting figure is explained below in {opt Example:.}
{title:Options}
{dlgtab:Model}
{phang}
{opt tau(real)} specifies the target size of the coefficient of the treatment variable.
For example, if a researcher finds the treatment effect of 1.7 and wants to know the strength of the confounding by an unobservable that halves the coefficient, s/he should set {opt tau(0.85)}.
The contour in the figure represents the set of partial effects of an unobservable that changes the coefficient to 0.85.
{phang}
{opt observation(int)} specifies the number of points at which the value of {opt alpha} in the horizontal axis is fixed.
The values of alpha are automatically selected based on the number of {opt observation} and the range of {opt alpha}.
In the 2nd run, the roles of {opt alpha} and {opt delta} are switched.
In general, a larger value of {opt observation(int)} results in a better figure at the cost of computation time.
{phang}
{opt resolution(int)} specifies the number of iterations in finding the value of {opt delta} for each fized value of {opt alpha}. The algorithm searches {opt delta} by splitting the range of {opt delta} into half in each iteration.
Thus, the resolution specifies how fine estimate about {opt delta} a researcher wants to obtain.
For example, when the range of {opt delta} is [0,8], {opt resolution(1)} only tells us whether the value of {opt delta} lies in [0,4) or [4,8] while {opt resolution(3)} provides the value of {opt delta} at one-unit precision, say [3,4) or [4,5].
Default is {cmd: resolution(20)}.
In the 2nd run, the roles of {opt alpha} and {opt delta} are switched.
In general, a larger value of {opt resolution(int)} results in a more accurate figure at the cost of computation time.
{phang}
{opt minalpha(real)} and {opt maxalpha(real)} specify the range of {opt alpha} within which the values of {opt alpha} are searched.
In Imbens (2003), {opt alpha} denotes the coefficient of unobservable in the treatment assignment equation and is represented by the horizontal axis.
In the first run, -{cmd:isa}- fixes the value of {opt alpha} at {opt observation(int)} different points and, for each value of alpha,
finds {opt delta} that changes the treatment effect to a given amount by changing its value.
In the second run, {opt alpha} plays the same role as {opt delta} in the first run.
{phang}
{opt mindelta(real)} and {opt maxdelta(real)} specify the range of {opt delta} within which the values of {opt delta} are searched.
In Imbens (2003), {opt delta} denotes the coefficient of unobservable in the outcome equation and is represented by the vertical axis.
In the first run, for each value of alpha, -{cmd:isa}- finds {opt delta} that changes the treatment effect to a given amount by changing its value.
The value of {opt delta} is searched until the range of {opt delta} is halved by {opt resolution(int)} times,
or the updated treatment effect is close enough to the target value.
In the second run, {opt alpha} plays the same role as {opt delta} in the first run.
{dlgtab:SE/Robust}
{phang}
{opt vce(vcetype)} "specifies the type of standard error reported, which includes types that are derived from asymptotic theory, that are robust to some kinds of misspecification that allow for intragroup correlation;
see {helpb vce_option:[R] vce_option}."
(*cited from Stata manual.)
{dlgtab:Advanced}
{phang}
{opt precision(real)} specifies the percentage error from {opt tau(real)} at which iteration ends.
The default is {cmd: precision(0.1)}, which means that when a researcher sets {opt tau(.314)}, for each fixed value of alpha(1st run) and delta(2nd run),
-{cmd:isa}- stop iteration and move to the next value if the t-statistics of the treatment effect take any value between .313686(99.9% of .314) and .314314(100.1% of .314).
There is a trade off between computational time and accuracy, but the default value should be sufficiently small.
{phang}
{opt sminalpha(real)} and {opt smaxalpha(real)} specify the range of {opt alpha} within which the values of {opt alpha} are searched in the second run.
In the second run, for each value of delta, -{cmd:isa}- finds {opt alpha} that changes the treatment effect to a given amount by changing its value.
The value of {opt alpha} is searched until the range of {opt alpha} is halved by {opt resolution(int)} times, or the updated treatment effect is close enough to the target value.
Researchers may set these options if they want to narrow down the range of {opt alpha} in the second run.
{phang}
{opt smindelta(real)} and {opt smaxdelta(real)} specify the range of {opt delta} within which the values of {opt delta} are searched in the second run.
In the second run, -{cmd:isa}- fixes the value of {opt delta} at {opt observation(int)} different points and, for each value of delta,
finds {opt alpha} that changes the treatment effect to a given amount by changing its value.
Researchers may set these options if they want to increase the range of {opt delta} in the second run because needed to set the narrow range for {opt delta} in the 1st run.
{phang}
{opt skiprangecheck} skips the range check of initial values of minalpha, minalpha, minalpha and minalpha.
This option should be specified if researchers are certain about their specification of the initial values.
(e.g., when researchers ran -{cmd:isa_search}- before running -{cmd:isa}-.)
{phang}
{opt showrangecheck} shows the results of the range check performed in Step 2.
This option helps users figure out how much change in {opt alpha} or {opt delta} is necessary for the program to pass the range check.
If the dataset is small or highly unbalanced, -{cmd:isa_search}- is better option to specify the values of {opt alpha} and {opt delta}.
{phang}
{opt quick} directs -{cmd: isa}- to move to the next step when the contour plots are more likely to be found if it switches the role of {opt alpha} and {opt delta}.
Specifically, -{cmd:isa}- moves to the next step if it cannot find {opt delta} smaller than 75% of the target value for a given fixed {opt alpha} or {opt delta}.
For small, unbalanced data, it is possible, however, that this option mistakenly skips the rest of iterations due to the finding of irrelevant contour.
{phang}
{opt tstat(real)} specifies the target size of the t-statistics of the treatment variable.
Because the standard errors produced by -isa- are not accurate, the users should use -gsa- if they want to draw the bound in terms of test statistics.
See Harada (2012) for the detail.
{title:Example}
{pstd}
The example below evaluates the effect of union membership on hourly wage assuming that union membership is a treatment variable.
{phang}. {stata sysuse nlsw88:sysuse nlsw88}{p_end}
{phang}. {stata xi i.race:xi i.race}{p_end}
{phang}. {stata rename _Irace_2 black:rename _Irace_2 black}{p_end}
{phang}. {stata rename _Irace_3 other:rename _Irace_3 other}{p_end}
{phang}. {stata xi i.industry:xi i.industry}{p_end}
{phang}. {stata rename _Iindustry_5 mnfctr:rename _Iindustry_5 mnfctr}{p_end}
{phang}. {stata isa wage union grade mnfctr south black age other married c_city, tau(.314):isa wage union grade mnfctr south black age other married c_city, tau(.314)}{p_end}
{pstd}
If covariates in the model capture all factors that affect the wage and the likelihood of becoming a union member, the treatment effect is unbiased.
However, researchers might be worry about the possibility that some unobserved individual predispositions affect the decision of becoming a union membership and the wage.
In the original model, union membership increases the hourly wage by .628 , so I set {opt tau(.314)}, which reduces the treatment effect into half.
{pstd}
In 10 minutes or so, -{cmd:isa}- produces the figure with a convex contour and several plots below the contour.
This contour indicates the statistical power that an unobservable needs to have for the treatment effect to become half.
For example, an unobservable needs to have the partial R-squares of 0.025 in the outcome equation and 0.1 in the treatment assignment equation to halve the treatment effect.
Several plots provide further information about the strength of this assumption.
These plots represent the partial R-squares when each of them is omitted from the models.
In other words, these plots show the statistical power of each covariate.
{pstd}
For instance, the plots of {opt grade} (years of education) and {opt mnfctr} (manufacturing sector job) are located near the contour.
This indicates that if there is an unobservable of which effects on the outcome variable and the treatment variable are as strong as the effects of {opt grade} and {opt mnfctr}, the treatment would become half.
Whether this assumption is strong or weak depends on researchers substantive knowledge.
For example, if a researcher believes that an unobservable that is comparable to education is unlikely to exist, the treatment effect of union membership would be fairly robust to against unobserved confounders.
{title:Tips}
{pstd}
1. Do not set {opt maxalpha} and {opt maxdelta} too large.
This increases the probability that -{cmd:isa}- finds the irrelevant contour, which messes up the algorithm.
Default values often specify the reasonable range.
{pstd}
2. If -{cmd:isa}- produces oddly-shaped, too-short, or no contour, try -{cmd:isa_search}-.
If a contour is almost OK but you want a better contour,
you may only need to adjust some options such as {opt observation}, {opt resolution}, {opt minalpha}, {opt maxalpha}, {opt mindelta}, and {opt maxdelta}.
{pstd}
3. After -{cmd:isa_search}-, if setting {opt minalpha}, {opt maxalpha}, {opt mindelta}, and {opt maxdelta} according to the result of -{cmd:isa_search}- still does not produce a good contour,
set {opt sminalpha}, {opt smaxalpha}, {opt smindelta}, and {opt smaxdelta}.
However, this does not necessarily mean that setting these options always increases the probability of producing a good contour.
Therefore, try both with and without these options.
{pstd}
4. You can manually drop problematic observations that produce the irrelevant contour(s) and reproduce the contour using -{cmd:isa_graph}-.
{pstd}
5. Without setting {opt sminalpha}, {opt smaxalpha}, {opt smindelta}, and {opt smaxdelta},
the options "{opt skip} {opt obs(200)}" may help produce a contour.
In this case, you probably need to drop some problematic results manually.
{title:Saved results}
{pstd}
-{cmd:isa}- saves the following variables.
The graph can be reproduced using these saved variables.
The results of {it: _n} th iteration are recorded in the {it: _n} th row:
{synoptset 20 tabbed}{...}
{synopt:{cmd:isa_tau}}the updated treatment effect when the coefficients of unobservable are {opt alpha} and {opt delta}{p_end}
{synopt:{cmd:isa_se}}the standard error of the updated treatment effect{p_end}
{synopt:{cmd:isa_t}}the t-statistics of the updated treatment effect{p_end}
{synopt:{cmd:isa_alpha}}the value of alpha{p_end}
{synopt:{cmd:isa_delta}}the value of delta{p_end}
{synopt:{cmd:isa_partial_rsq_y}}the partial r-square of an unobservable in the outcome equation{p_end}
{synopt:{cmd:isa_partial_rsq_t}}the partial r-square of an unobservable in the treatment assignment equation{p_end}
{synopt:{cmd:isa_converge}}identifier of whether the estimation achieved convergence or not{p_end}
{synopt:{cmd:isa_partial_rsq_yx}}the partial r-square of the {it: k} th covariate in the outcome equation{p_end}
{synopt:{cmd:isa_partial_rsq_tx}}the partial r-square of the {it: k} th covariate in the treatment assignment equation{p_end}
{synopt:{cmd:isa_plotvar}}the name of the {it: k} th covariate{p_end}
{title:Also see}
{psee}
{space 2}Help: {help isa_search} and {help isa_graph}
{title:Reference}
{pstd}
{browse "https://files.nyu.edu/mh166/public/docs/quick_guide_isa.pdf":Here} is the link for the quick guide by the author of the program.
{pstd}
If you use this program, please cite:
{pstd}
Imbens, Guido W. 2003. "Sensitivity to Exogeneity Assumptions in Program Evaluation." {it:The American Economic Review} 93(2):126-132.
{pstd}
Harada, Masataka "Generalized Sensitivity Analysis." {it:Working paper}.
{title:Contact}
{pstd}
Please feel free to contact Masataka Harada(masatakaharada@nyu.edu) for any feedback or comments.