.-
help for ^stmh^                                           (STB-40: ssa10)
.-

Calculation of rate ratios using ^st^ data
------------------------------------------

^stmh^ varname [varlist] [^if^ exp] [^in^ range]  [, ^by(^varlist^)^ 
       ^c^ompare^(^rule1^,^rule2^)^ ^fc^odes^(^rule^)^ ^nom^iss ^l^evel^(^#^)^]


Description
-----------

In its simplest use, ^stmh^ estimates the ratio of the rates of failure for 
two categories of the explanatory variable (the first argument). Categories
to be compared may be defined, as ^recode^ rules, in the ^compare^ option. 
Alternatively the command may be used to carry out trend tests for a 
metric explanatory variable. In this latter case a one-step Newton 
approximation to the log-linear Poisson regression coefficient is also 
computed.

The remaining variables before the comma are categorical variables 
which are to be controlled for using stratification. Strata are defined
by cross-classification by all of these variables and the rate ratio 
estimate is combined over strata using the Mantel-Haenszel method. 
Using the ^by^ option, the variation of the rate ratio with further 
categorical variables may be explored.

Upon completion, the macro ^S_1^ contains the overall Mantel-Haenszel 
estimate of the rate ratio, thus facilitating bootstrap and jackknife 
evaluation of confidence intervals.


Options
-------

^by(^varlist^)^ specifies categorical variables by which the rate ratio 
is to be tabulated. A separate rate ratio produced for each category
or combination of categories, and a test for unequal separate rate ratios 
is given. The same applies when the rate ratio for a unit increase in
xvar is given in place of the rate ratio.

^compare(^rule1^,^rule2^)^ defines, the categories of the exposure
variable to be compared. Categories are defined by the left-hand side
of ^recode^ rules. The first rule defines the numerator categories and
the second the denominator. When ^compare^ is absent and there are
only two categories, the larger is compared to the smaller; when there
are more than two categories an analysis for (log-linear) trend is
carried out.

^fcodes()^ specifies, as the left-hand side of a recode rule, the
codes for the failure indicator in the ^st^ data which will be
considered as failure. All other codes are treated as censoring.

^nomiss^ specifies that only cases that have no missing values for any 
stratifying variables should be included. Otherwise missing values will 
define new strata in the analysis.

^level(^#^)^ gives the level for the confidence intervals (default 95).


Examples
--------

^stmh smoke age, c(2,1) by(sex)^  - RR for smoking, controlled for age, by sex

^stmh smoke age sex, c(2,1)^      - RR for smoking, controlled for age and sex

^stmh ncigs age sex^              - Trend analysis for cigarettes/day  

^stmh ncigs age sex, c(21/max,1/20)^ - Heavy smoking vs light smoking


Authors
-------

      David Clayton
      MRC Biostatistical Research Unit
      Cambridge
      email: david.clayton@@mrc-bsu.cam.ac.uk

      Michael Hills
      London School of Hygiene and Tropical Medicine (retired)
      email: mhills@@regress.demon.co.uk


Also see
--------

    STB:  STB-40 ssa10
On-line:  help for @stmc@, @dymh@, @stlexis@