help nonparmde                                                    Version 1.0.0
-------------------------------------------------------------------------------

Title

    nonparmde -- Calculates the minimum detectable effect (MDE) using a
    Horvitz-Thompson estimator and a size (and covariate) adjusted Raj
    estimator for cluster randomized controlled experiments.

Syntax

    nonparmde varlist , mtreatment(clusternum) mcontrol(clusternum)
    [n(clustersizevar) kx(numlist) averages power(num) ci(num) crossfold(num)
    avgclustersize(num)]

Description

    nonparmde is a method for calculating the minimum detectable effect (MDE)
    using the nonparametric estimators proposed in Middleton & Aronow (2011).
    This program is for use on cluster-level data to calculate the minimum
    effect that a cluster randomized experiment would be able to reliably
    detect at a given level of statistical power.

Options

    varlist the name of the prior DV variable, optionally followed by a list
    of covariates that are expected to explain variance in the DV variable
    (all variables must be either cluster averages or cluster totals)

    mtreatment(clusternum) number of clusters assigned to treatment group
 
    mcontrol(clusternum) number of clusters assigned to control group
 
    n(clustersizevar) name of the variable that denotes cluster size (if left
    unspecified avgclustersize(num) must be specified)
 
    kx(numlist) regression coefficients of covariates regressed on prior DV
    variable, controlling for cluster size (calculated automatically if left
    unspecified)
 
    averages forces the program to treat all variables as cluster averages
    (n(clustersizevar) must be specified if this option is used)
 
    power(num) statistical power (if left unspecified assumes 80% power)
 
    ci(num) confidence interval (if left unspecified assumes 95% CI)
 
    crossfold(num) crossfolds used for the calculation of kx; can be either
    an integer or a variable specifying cutoffs (if left unspecified no
    crossfolding is done). The command xvalols must be installed.
 
    avgclustersize(num) average cluster size (use only if n(clustersizevar)
    is left unspecified)
 


Examples

    . nonparmde Y_total, mtreatment(50) mcontrol(50) avgclustersize(20.61)

    . nonparmde y_avg x_avg n, n(n) averages mtreatment(50) mcontrol(50)

    . nonparmde Y_total X_total Z_total, mtreatment(50) mcontrol(50) kx(.2
    .1) averageclustersize(20.61)

    . nonparmde Y_total X_total Z_total, mtreatment(50) mcontrol(50)
    averageclustersize(20.61)

    . nonparmde Y_total X_total Z_total, mtreatment(50) mcontrol(50) n(n)

    . nonparmde Y_total X_total Z_total n, mtreatment(50) mcontrol(50) n(n)

    . nonparmde Y_total X_total Z_total n, mtreatment(50) mcontrol(50) n(n)
    crossfold(cut)

    . nonparmde Y_total X_total Z_total n, mtreatment(50) mcontrol(50) n(n)
    crossfold(2)


Saved Results

    nonparmde saves the following in e():

    Scalars             
      e(Vht)                   Horvitz-Thompson variance
      e(Vraj)                  Raj variance
      e(Vrajcovars)            covariate adjusted Raj variance

References
    Middleton, Joel A. and Aronow, Peter M., Unbiased Estimation of the
    Average Treatment Effect in Cluster-Randomized Experiments (April 5,
    2011). Available at SSRN: http://ssrn.com/abstract=1803849

Authors

    Joel Middleton
    New York University
    joel.middleton@gmail.com

    John Ternovski
    Analyst Institute
    johnt1@gmail.com