.-
help for ^opartchi^                                               [STB-51: sg118]
.-
.
Partitions of Pearson's Chi-square for two-way tables with ordered columns
--------------------------------------------------------------------------
.
^opartchi^ column_var [^if^ exp] [^in^ exp] [weight]^, by(^row_var^)^
   [^tab^le ^orows^ ^l^oglinear ^sc^ore^(^column_score_var^)^ ^rsc^ore^(^row_score_var^)^]
.
^fweight^s are allowed. See help @weights@.
.
.
Description
-----------
.
This command is suitable for exploratory analysis of data that can be 
summarised in a two-way contingency table with ordered columns given by
^column_var^ and rows given by ^row_var^. The option ^table^ can be used to @tabulate@
the data showing row percentages. The command provides test statistics that 
summarise the extent to which the distribution of responses on the ordered 
scale differs between the rows. The first way in which these distributions may
differ is in their location on the ordered scale; the second way is in their 
dispersion across the ordered scale.
.
Two types of test statistics are provided. By default, Pearson's Chi-square 
statistic is partitioned into components that describe the appropriate row 
differences. The ^loglinear^ option provides comparable deviance statistics from 
fitting log-linear models to the data. Both of these procedures require the 
attachment of increasing scores to the ordered columns. By default, increasing 
integer scores are used but the ^score()^ option can be used to provide 
user-defined scores.
.
If the rows also have a natural ordering then the option ^orows^ can be used to 
obtain a sensitive test of the extent to which location differences increase 
(or decrease) monotonically across the rows.
.
The data can be provided either at an individual level (one ordinal outcome 
value per observation) or at a contingency table level in which case there is 
a variable in the dataset that contains the observed counts for each cell of 
the table. The latter requires the use of the optional frequency weighting.
.
.
Options
-------
.
^by(^row_var^)^ is required. The variable row_var gives the rows of the 
    contingency table.
.
^table^ is optional and displays the two-way contingency table (with row 
    percentages) that is to be analyzed.
.
^score(^column_score_var^)^ is an option to supply user-defined scores for the 
    column values instead of using the default increasing integer scores.
.
^orows^ is optional and should only be used where row_var is an ordinal 
    variable. This option displays the nested test for trend within the
    location component of row differences.
.
^rscore(^row_score_var^)^ is an option to supply user-defined scores for the row
    values instead of using the default increasing integer scores.
.
^loglinear^ is optional. Specifying this option displays deviance statistics
    from log-linear models as well as the counterpart Pearson's chi-square
    partitions.
.
.
Examples
--------
.
*Individual-level data
^opartchi recovery, by(treatmnt)^
^opartchi recovery, by(treatmnt) loglin orows rscore(tr5)^
.
^egen midrank=rank(recovery)
^opartchi recovery, by(treatmnt) score(midrank)^

* Contingency table data
^opartchi pay_grp [fweight=counts], by(state)
.
.
Authors
-------
.
  Rory Wolfe 
  Royal Children's Hospital, Australia
  wolfer@@cryptic.rch.unimelb.edu.au
.
.
Also see
--------
.
Manual: ^glm, fam(poisson) link(log)^ for log-linear modelling
On-line: help see @tabulate@, @kwallis@, @nptrend@, @ologit@
FAQ: @http://www.stata.com/support/faqs/stat/trend.html@
STB: STB-44 sg88 (@gologit@ if installed)