..-
help for ^ocratio^ Statalist, 18Feb1998
..-
Continuation-ratio models for ordinal responses
- -----------------------------------------------
^ocratio^ depvar varlist [weight] [^if^ exp] [^in^ range] [, ^t^est
^li^nk^(^linkname^)^ ^e^form ^cum^ulative ^l^evel^(^#^)^ ]
^ocrpred^ [type] newvarlist [, ^xb^ ^pr^ob ]
where linkname is one of
^l^ogit | ^p^robit | ^c^loglog
^fweights^ are allowed; see help @weights@ and examples below.
^ocratio^ shares most features of estimation commands; see ^[U] 26 Estimation^
^and post-estimation commands^. However, to obtain predictions from the model
^ocrpred^ should be used in place of @predict@.
To reset problem-size limits, see help @matsize@.
Description
- -----------
^ocratio^ fits models based on continuation-ratio probabilities to ordinal
responses (using maximum-likelihood). Three models are available using the
logit, probit and complementary-log-log (f(x)=log[-log(1-x)]) functions.
These models are an alternative to the models based on cumulative
probabilities fitted by @ologit@ and @oprobit@. ^ocratio^ also enables the
fitting of a "ocloglog" model by using the ^cumulative^ option described
below. This is possible because the "ocloglog" model is equivalent to the
continuation-ratio complementary-log-log.
^ocrpred^, for use after ^ocratio^, creates newvar containing the index (i.e.
xb, see @predict@) or newvarlist containing predicted probabilities for each
response category (similarly to @ologitp@).
Options for use with ^ocratio^
- ----------------------------
^test^ requests the calculation of the likelihood-ratio test of whether the
model parameters are constant across the categories of the response
variable.
^link(^linkname^)^ specifies the link function. The default is ^link(logit)^
unless ^cumulative^ is requested in which case it is ^link(cloglog)^.
^eform^ displays the exponentiated coefficients with corresponding standard
errors and confidence intervals as described in ^[R] maximize^. For the
logit link, exponentiation results in odds ratios of continuation-ratio
probabilities; for the cloglog link, exponentiation results in ratios
of discrete hazard functions.
^cumulative^ requests the fitting of the continuatio-ratio cloglog model and
the display of the cut-points of the equivalent ordered cloglog model.
^level(^#^)^ specifies the confidence level in percent, for confidence intervals.
The default is ^level(95)^ or as set by @set level@.
Options for use with ^ocrpred^
- ----------------------------
^xb^ requests the linear predictor xB and one name should be supplied in
newvarlist. This is the default if no options are specified.
^prob^ requests the predicted probabilties for each response category. There
should be K names supplied in newvarlist where K is the value of $S_3
(i.e. the number of response categories) from the previous ^ocratio^
command. An easy way of doing this is by using ranges, e.g. if K=8,
ocrpred p1-p8, prob.
Examples
- --------
. use auto.dta, clear
. ^xi: ocratio rep78 i.foreign weight, link(probit) level(95)^
. ^ocratio, level(90)^ /* Redisplay results with different CIs */
. ^ocrpred pr1 pr2 pr3 pr4 pr5, prob^ /* Obtain fitted probabilities */
. ologit rep78 weight /* Look at ordered logit model */
/* Now fit ordered complementary-log-log model for comparison*/
. ^ocratio rep78 weight, cumul link(clog)^
. tab foreign rep78
. ^xi: ocratio rep78 i.foreign, test^ /* Test coefficients. */
/* Now put data into 2-way */
. gen obs=1 /* table of counts to */
. collapse (count) obs, by(rep78 foreign) /* demonstrate use of */
. ^xi: ocratio rep78 i.foreign [fweight=obs], test^ /* fweights. */
/* and look at exponentiated coefficients, ie. odds ratios. */
. ^ocratio, eform^
Saved results
- -------------
In addition to standard post-estimation results, ^ocratio^ saves in
the global ^S_^# macros:
^S_1^ likelihood-ratio value for test of effect constancy assumption
^S_2^ dof for likelihood-ratio test
^S_3^ number of distinct response categories (= no. of cut-points +1).
Note that S_1 and S_2 are only defined if the ^test^ option is provided.
Author
- ------
Rory Wolfe , Royal Children's Hospital,
Melbourne, Australia.
Also see
- --------
On-line: help for @ologit@, @oprobit@, @ologitp@