help emh                                                 Version 1.0 2010-10-12
-------------------------------------------------------------------------------

Title

    emh -- Extended Mantel-Haenszel Statistics


Syntax

        emh var1 var2 [if] [in] [weight] [, options]

    options                     Description
    -------------------------------------------------------------------------
    Main
      strata(varlist)           stratify on varlist
      correlation               Correlation Statistic; the default
      anova                     ANOVA (Mean Scores) Statistic
      general                   General Association Statistic
      transformation(scoretype) use transformation scores, e.g., ranks
    -------------------------------------------------------------------------
    by is allowed; see [D] by.
    fweights are allowed; see weight.
    Only one of correlation, anova and general is permitted at a time.


Description

    emh calculates extended Mantel-Haenszel chi-square test statistics (also
    known as Cochran-Mantel-Haenszel Statistics) for stratified tables.
    Tables are defined by var1 and var2, both which must be numeric.
    Stratification is optional.


Options

        +------+
    ----+ Main +-------------------------------------------------------------

    strata(varlist) specifies variables on which to stratify the analysis.
        Stratification variables may be numeric, string or a mix of the two.

    correlation Correlation Statistic--tests for a linear association between
        var1 and var2.  Suitable when both are at least ordinal.

    anova ANOVA (Row Mean-score Differences) Statistic--tests for a
        difference in means of var1 between groups that are defined by var2.
        Suitable when var1, the response variable, is at least ordinal.
        var2, the predictor variable, may be nominal (unordered).

    general General Association Statistic--most general form of association;
        of interest especially when both var1 and var2 are nominal.

    transformation(scoretype) When correlation is specified, transforms both
        var1 and var2; when anova is specified, transforms the response
        variable (var1).  This option is ignored when general is specified.
        Available scoretypes are table (untransformed--the default), integer,
        rank, ridit, modridit (standardized midranks), savage (Savage
        scores), mood (Mood scores), median (above or at-or-below median),
        vdw (van der Waerden scores) and klotz (Klotz scores).


Remarks

    emh computes chi-square test statistics for association of row and column
    variables in unstratified or stratified tables.  A good introduction to
    the use of these statistics is Chapters 2 through 7 of M. E. Stokes, C.
    S. Davis and G. G. Koch, Categorical Data Analysis Using the SAS(R)
    System Second Edition.  Cary, North Carolina:  SAS Institute, 2000.

    emh uses tabulate internally.  The number of levels of var1 and var2 are
    limited to what tabulate can accept.  In order to accommodate an analysis
    with many levels of either or both of these variables, e.g., with
    continuous variables, matsize and memory might need to be set larger.

    Computing extended Mantel-Haenszel statistics involves inverting a pooled
    covariance matrix.  For some datasets, the matrix might be singular.
    When this occurs, a warning is displayed, and the chi-square statistic
    and associated p-value are set to missing.  Such occurrences are more
    likely when requesting the General Association statistic for a dataset
    with many levels of row and column variables and with a substantial
    amount of missing data.

    Some transformation types available in emh have an established use in
    unstratified two-group mean score-difference analysis, e.g., Mood test,
    van der Waerden test, Klotz test.  Their use does not appear so well
    known in stratified analysis of multi-group (>2) datasets.  In this
    context, such transformations might not display the desired sensitivity
    to location or scale difference.  In general, users should exercise care
    in choosing whether and how to employ various scoring schemes.


Examples

    . sysuse auto

    . emh rep78 price, s(foreign)


Saved results

    emh saves the following in r():

    Scalars   
      r(chi2)        chi-square test statistic
      r(df)          degrees of freedom for chi-square statistic
      r(p)           p-value for reported chi-square statistic and degrees of
                       freedom

    Macros    
      r(scoretype)   short-name for score type, as specified in
                       transformation(scoretype)
      r(ScoreType)   formal name for score type, as displayed in output


Author

    J. Coveney E-mail jcoveney@bigplanet.com if you observe any problems.


Also see

      Help:  [ST] epitab; vanelteren (if installed), somersd (if installed), 
             [R] tabulate twoway