Title
avg_effect -- Calculate mean (standardized) effect size across multiple outcomes
Syntax
avg_effect yvar1 yvar2 [ yvar3 ... ] [if] , x(varlist) effectvar( varlist) controltest(string) [ options ]
options Description ------------------------------------------------------------------------- Main x(varlist) list of all right-hand-side (independent) variables effectvar(varlist) one or more x variables that constitute outcomes to test controltest(string) if-style test for membership in the control group robust use heteroskedasticity-robust standard errors cluster(varname) use clustered standard errors keepmissing do not drop down to a common sample of observations -------------------------------------------------------------------------
Description
avg_effect follows Kling et al. (2004) and Clingingsmith et al. (2009) in calculating average (standardized) effect size using the seemingly-unrelated regression framework to account for covariance across estimates.
Call avg_effect with two or more y (outcome) variables and a list of x variables in x (exactly as you would pass them to the regress command).
Use effectvar to specify the variable whose coefficient represents the individual effect estimates (e.g., effectvar(treated)). You can specify multiple effect variables, separated by spaces.
Use controltest to specify the if-style test for membership in the control group (e.g., controltest(treated==0)). The control group is used to calculate the standard deviation of each outcome variable, which is then used to standardize the effect sizes.
Works cited:
Clingingsmith, David, Khwaja, Asim Ijaz and Kremer, Michael (2009) “Estimating the Impact of the Hajj: Religion and Tolerance in Islam's Global Gathering,” Quarterly Journal of Economics, 124(3), pp. 1133-1170.
Kling, Jeffrey R., Liebman, Jeffrey B., Katz, Lawrence F. and Sanbonmatsu, Lisa (2004) “Moving to Opportunity and Tranquility: Neighborhood Effects on Adult Economic Self-Sufficiency and Health from a Randomized Housing Voucher Experiment,” KSG Working Paper No. RWP04-035, Available from: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=588942 (Accessed 25 May 2010).
Options
+------+ ----+ Main +-------------------------------------------------------------
x(varlist) specifies the full list of right-hand-side (independent) variables, as when passed to regress. Required.
effectvar(varlist) indicates which of the x variables corresponds to the effect of interest (i.e., which coefficient constitutes the effect of interest). To simultaneously test multiple effects, list multiple variables separated by spaces. Required.
controltest(string) specifies the if-style test used to identify the control group (e.g., controltest(treated==0)). The control group is used to calculate the standard deviations by which effect sizes are standardized. Required.
robust uses heteroskedasticity-robust standard errors, using suest's vce(robust) option.
cluster(varlist) uses clustered standard errors, using suest's vce(cluster varlist) option.
keepmissing uses as many observations as possible when calculating each standard deviation and running each individual regression. By default, avg_effect considers only observations with non-missing values for the full set of y (outcome) variables; this ensures that the results apply to a common sub-sample. For example, observations with missing yvar1 will not be used when considering the standard deviation and regression for yvar2. With keepmissing, this is no longer the case, and thus the sub-sample can change with each individual regression.
Examples
. sysuse auto
. avg_effect price mpg, x(foreign weight) effect(foreign) control(foreign==0)
Author
Christopher Robert, Harvard University, chris_robert@hksphd.harvard.edu
Also see
Online: help for suest