Logistic regression and testing for trend

[Chris Hearne]

I have a logistic model with an ordered categorical independent variable (4 levels) and two other covariates I want to control for. The ORs increase in magnitude with higher levels of the independent variable (though none is significant on its own). I want to run a test of trend to show the trend formally, but I'm having trouble doing this in Stata.

I tried the technique mentioned here: http://www.stata.com/statalist/archi.../msg00813.html but I'm getting an error at the third step (where you rerun the code without the indicator variable prefix, I'm told "outcome does not vary.") Can anyone shed light on this, or perhaps suggest the most straightforward way of doing this? Thank you in advance.

[Joseph Coveney]

The Forum's FAQ encourages you to post both the code that gave rise to the error message as well as the result. You haven't shown any code and so it's impossible to see what you've done wrong, but, based upon the error message, it seems as if you accidentally omitted the dependent variable in your third step and Stata saw one of your (never-zero) independent variables as the dependent variable.

Regardless, you can test for linear trend using the contrast postestimation command. An example of its use is illustrated below (begin at the "Begin here" comment).

.ÿversionÿ14.2

.ÿ
.ÿclearÿ*

.ÿsetÿmoreÿoff

.ÿ
.ÿquietlyÿsysuseÿauto

.ÿ
.ÿforeachÿvarÿofÿvarlistÿgear_ratioÿmpgÿ{
ÿÿ2.ÿÿÿÿÿÿÿÿÿegenÿdoubleÿstd_`var'ÿ=ÿstd(`var')
ÿÿ3.ÿ}

.ÿ
.ÿegenÿbyteÿcat_priceÿ=ÿcut(price),ÿgroup(4)

.ÿ
.ÿ/*ÿcat_priceÿisÿyourÿordered-categoricalÿpredictor;
>ÿÿÿÿstd_gear_ratioÿandÿstd_mpgÿareÿyourÿtwoÿ"control"ÿpredictorsÿ*/
.ÿ
.ÿ*
.ÿ*ÿBeginÿhere
.ÿ*
.ÿlogisticÿforeignÿi.cat_priceÿc.(std_*),ÿnolog

LogisticÿregressionÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNumberÿofÿobsÿÿÿÿÿ=ÿÿÿÿÿÿÿÿÿ74
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿLRÿchi2(5)ÿÿÿÿÿÿÿÿ=ÿÿÿÿÿÿ55.81
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿProbÿ>ÿchi2ÿÿÿÿÿÿÿ=ÿÿÿÿÿ0.0000
Logÿlikelihoodÿ=ÿ-17.126305ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿPseudoÿR2ÿÿÿÿÿÿÿÿÿ=ÿÿÿÿÿ0.6197

--------------------------------------------------------------------------------
ÿÿÿÿÿÿÿforeignÿ|ÿOddsÿRatioÿÿÿStd.ÿErr.ÿÿÿÿÿÿzÿÿÿÿP>|z|ÿÿÿÿÿ[95%ÿConf.ÿInterval]
---------------+----------------------------------------------------------------
ÿÿÿÿÿcat_priceÿ|
ÿÿÿÿÿÿÿÿÿÿÿÿ1ÿÿ|ÿÿÿ3.816533ÿÿÿ5.271325ÿÿÿÿÿ0.97ÿÿÿ0.332ÿÿÿÿÿÿ.254686ÿÿÿÿ57.19169
ÿÿÿÿÿÿÿÿÿÿÿÿ2ÿÿ|ÿÿÿ60.48421ÿÿÿ106.0823ÿÿÿÿÿ2.34ÿÿÿ0.019ÿÿÿÿÿ1.944163ÿÿÿÿ1881.704
ÿÿÿÿÿÿÿÿÿÿÿÿ3ÿÿ|ÿÿÿ29.31048ÿÿÿ54.43213ÿÿÿÿÿ1.82ÿÿÿ0.069ÿÿÿÿÿ.7696037ÿÿÿÿ1116.294
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|
std_gear_ratioÿ|ÿÿÿÿ30.3164ÿÿÿ30.49308ÿÿÿÿÿ3.39ÿÿÿ0.001ÿÿÿÿÿ4.221972ÿÿÿÿ217.6907
ÿÿÿÿÿÿÿstd_mpgÿ|ÿÿÿ1.427715ÿÿÿÿ1.01586ÿÿÿÿÿ0.50ÿÿÿ0.617ÿÿÿÿÿ.3539873ÿÿÿÿ5.758316
ÿÿÿÿÿÿÿÿÿ_consÿ|ÿÿÿ.0157372ÿÿÿ.0224363ÿÿÿÿ-2.91ÿÿÿ0.004ÿÿÿÿÿ.0009625ÿÿÿÿ.2573195
--------------------------------------------------------------------------------

.ÿ
.ÿcontrastÿp.cat_price,ÿnoeffects

Contrastsÿofÿmarginalÿlinearÿpredictions

Marginsÿÿÿÿÿÿ:ÿasbalanced

------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿÿÿÿ|ÿÿÿÿÿÿÿÿÿdfÿÿÿÿÿÿÿÿchi2ÿÿÿÿÿP>chi2
-------------+----------------------------------
ÿÿÿcat_priceÿ|
ÿÿÿ(linear)ÿÿ|ÿÿÿÿÿÿÿÿÿÿ1ÿÿÿÿÿÿÿÿ4.31ÿÿÿÿÿ0.0379
(quadratic)ÿÿ|ÿÿÿÿÿÿÿÿÿÿ1ÿÿÿÿÿÿÿÿ0.99ÿÿÿÿÿ0.3205
ÿÿÿÿ(cubic)ÿÿ|ÿÿÿÿÿÿÿÿÿÿ1ÿÿÿÿÿÿÿÿ1.62ÿÿÿÿÿ0.2027
ÿÿÿÿÿÿJointÿÿ|ÿÿÿÿÿÿÿÿÿÿ3ÿÿÿÿÿÿÿÿ6.05ÿÿÿÿÿ0.1091
------------------------------------------------

.ÿ
.ÿexit

endÿofÿdo-file


.

[Chris Hearne]

Joseph, thank you very much. That provided exactly what I was looking for.