topic Re: Logistic, Wald chi-sq is sig but CI includes 1.0 in Statistical Procedures

Logistic, Wald chi-sq is sig but CI includes 1.0

patriciav — Thu, 21 Mar 2013 20:07:11 GMT

Hello,

I'm trying to predict preterm birth (yes/no) from weight gain during pregnancy, controlling for race/ethnicity, bmi at entry to prenatal care, and parity (number of previous births). I've run a logistic regression with the following model statement:

preterm=wtgaincat ethnicity bmicat parity;

wtgaincat has 4 levels (ethnicity has 3, bmicat has 4)

In my output, the p-value for the Wald Chi-Square in Type 3 Analysis of Effects table is .0292 - so less than .05, so significant.

In the Maximum Likelihood parameter estimates table, the p-value for one of the levels of wtgaincat is .0130, so significant.

The OR is for that level of wtgaincat 0.332.

BUT the CI is 0.082 -1.345 - so not significant?

Are there some situations where the p-value for the ML estimate is better, and other situations where you'd favor the 95% CI around the OR????

I'm not sure what to think or to report. Any help would be much appreciated!

Thanks,

Patricia

Re: Logistic, Wald chi-sq is sig but CI includes 1.0

ballardw — Thu, 21 Mar 2013 20:21:08 GMT

It would be helpful if you showed your code so we can avoid making suggestions already present in your model and the output as something else may be needed to answer the question.

Re: Logistic, Wald chi-sq is sig but CI includes 1.0

Reeza — Thu, 21 Mar 2013 20:21:15 GMT

What does the rest of your code look like within proc logistic or genmod?

Re: Logistic, Wald chi-sq is sig but CI includes 1.0

patriciav — Thu, 21 Mar 2013 20:39:11 GMT

Sorry, here is the proc, is that what you need? If more, I can paste that as well.

Thank you!

Patricia

proc logistic descending data=jain_ob7 ;

title1 'Assoc between wtgaincat and preterm';

title2 'Controlling for bmicat, ethnicity, parity';

class wtgaincat bmicat ethnicity;

model preterm=wtgaincat ethnicity bmicat parity;

format wtgaincat wtgaincatrevf.;

where bmicat>=3 and gaentry28=1;

run;

Re: Logistic, Wald chi-sq is sig but CI includes 1.0

Reeza — Thu, 21 Mar 2013 20:50:11 GMT

Does that give you the odds ratio and CI by default?

Re: Logistic, Wald chi-sq is sig but CI includes 1.0

patriciav — Thu, 21 Mar 2013 20:58:34 GMT

Yep.

Don't know if this will come through, but there is the relevant output.

"wtgaincat 0 above recommended" is the level that I'm talking about, that has conflicting p-value and OR CI.

Patricia

Type 3 Analysis of Effects

Wald

Effect DF Chi-Square Pr > ChiSq

wtgaincat 3 9.0083 0.0292

Ethnicity 2 12.8604 0.0016

bmicat 3 2.3003 0.5125

Parity 1 1.7427 0.1868

Analysis of Maximum Likelihood Estimates

Standard Wald

Parameter DF Estimate Error Chi-Square Pr > ChiSq

Intercept 1 -1.4628 0.3557 16.9141 <.0001

wtgaincat 0 above recommended 1 -0.9070 0.3652 6.1689 0.0130

wtgaincat 1 2 recommended 1 0.6497 0.3272 3.9432 0.0471

wtgaincat 2 1 below recommended 1 0.0617 0.3884 0.0253 0.8737

Ethnicity AA 1 0.9401 0.3552 7.0041 0.0081

Ethnicity H 1 -0.9584 0.2920 10.7740 0.0010

bmicat 30-35 1 -0.0366 0.3246 0.0127 0.9102

bmicat 35-40 1 -0.4800 0.3837 1.5646 0.2110

bmicat 40-45 1 0.4624 0.3772 1.5025 0.2203

Parity 1 -0.2109 0.1598 1.7427 0.1868

Odds Ratio Estimates

Point 95% Wald

Effect Estimate Confidence Limits

wtgaincat 0 above recommended vs 3 0 wt loss 0.332 0.082 1.345

wtgaincat 1 2 recommended vs 3 0 wt loss 1.575 0.417 5.942

wtgaincat 2 1 below recommended vs 3 0 wt loss 0.875 0.208 3.685

Ethnicity AA vs White 2.514 0.703 8.988

Ethnicity H vs White 0.377 0.129 1.101

bmicat 30-35 vs 45+ 0.913 0.266 3.140

bmicat 35-40 vs 45+ 0.586 0.153 2.251

bmicat 40-45 vs 45+ 1.504 0.396 5.717

Parity 0.810 0.592 1.108

Re: Logistic, Wald chi-sq is sig but CI includes 1.0

Reeza — Fri, 22 Mar 2013 14:20:51 GMT

patriciav wrote:

Sorry, here is the proc, is that what you need? If more, I can paste that as well.
Thank you!
Patricia

proc logistic descending data=jain_ob7 ;
title1 'Assoc between wtgaincat and preterm';
title2 'Controlling for bmicat, ethnicity, parity';

class wtgaincat bmicat ethnicity/param=ref;
model preterm=wtgaincat ethnicity bmicat parity; 

format wtgaincat wtgaincatrevf.;
where bmicat>=3 and gaentry28=1; 
run;

try adding the param=ref option (see above) as specified in this note:

36478 - The odds ratio for a confidence interval disagrees with the parameter p-value or confidence interval

Re: Logistic, Wald chi-sq is sig but CI includes 1.0

patriciav — Fri, 22 Mar 2013 16:08:06 GMT

That is perfect, thank you so much!!!!

Re: Logistic, Wald chi-sq is sig but CI includes 1.0

patriciav — Fri, 22 Mar 2013 16:30:26 GMT

Reeza, I can't figure out how to mark your response as correct, if this is important pls let me know and I'll do it.

Patricia