One approach to the problem with the adjusted trend is to start with the
command -tabodds-. -tabodds- provides a score test for trend that may
be adjusted for categorical variables. Your logistic model will provide
very nearly the same result as -tabodds-, as shown below:
This is an univariate analysis of age on low birthweight (low), using
data from Hosmer and Lemeshow's, 2000. Age has been categorized into 4
levels (1-4).
.tabodds low agecat
------------------------------------------------------------------------
--
agecat | cases controls odds [95% Conf.
Interval]
------------+-----------------------------------------------------------
--
1 | 15 36 0.41667 0.22814
0.76099
2 | 20 36 0.55556 0.32162
0.95966
3 | 15 21 0.71429 0.36823
1.38558
4 | 9 37 0.24324 0.11740
0.50397
------------------------------------------------------------------------
--
Test of homogeneity (equal odds): chi2(3) = 5.32
Pr>chi2 = 0.1501
Score test for trend of odds: chi2(1) = 0.70
Pr>chi2 = 0.4012
And here is the logistic command:
. logit low agecat
output:
Logistic regression Number of obs = 189
LR chi2(1) = 0.71
Prob > chi2 = 0.3990
Log likelihood = -116.98033 Pseudo R2 = 0.0030
------------------------------------------------------------------------
-
low | Odds Ratio Std. Err. z P>|z| [95% Conf. Interval]
-------+----------------------------------------------------------------
agecat | .8885542 .1249163 -0.84 0.401 .6745573 1.17044
------------------------------------------------------------------------
-
