topic Re: Odds Ratio Proc Logistic Question in Statistical Procedures

Odds Ratio Proc Logistic Question

david_mener_gmail_com — Tue, 19 Feb 2013 15:30:37 GMT

I am running several proc logistic models, utilizing the odds ratio estimates. Although SAS will produce the point estimate, and 95% wald confidence limits, I can not figure out the code for SAS to give me the p value for each point estimate in the odds ratios. Any suggestions? Thanks again!

Re: Odds Ratio Proc Logistic Question

Reeza — Tue, 19 Feb 2013 16:20:25 GMT

The p-value is in the estimates table.

Re: Odds Ratio Proc Logistic Question

david_mener_gmail_com — Tue, 19 Feb 2013 16:23:27 GMT

I don't think I am being as clear as I should. In the odds ratio tables I have generated from the Proc Logistic, I have asked sas in the original logistic model to compare (example: men, vs women) or diabetic vs non-diabetic. The estimates table does not give these comparisons as displayed in the odds ratio table.

Is that a little clearer? Thanks again for your help. ~Dave

Re: Odds Ratio Proc Logistic Question

Reeza — Tue, 19 Feb 2013 16:28:27 GMT

Where do odds ratio come from? They are calculated from the parameter estimates, exp(estimate) equal the odds ratio. Therefore p-value's are the same.

If the estimates aren't the comparison you're trying to make you can change your reference levels.

Re: Odds Ratio Proc Logistic Question

david_mener_gmail_com — Tue, 19 Feb 2013 17:22:24 GMT

Perhaps then my question is more of an understanding of SAS output. What's the reason that the PR > chisq in the parameter estimates value would be non-significant (ex: p 0.77) but the 95% wald confidence limits for the odd ratios is greater than 1 (ex: point estimate 1.023, 95% CI 1.018-1.028) for both the lower and upper limit. THe later for me seems to imply that the comparison is significant. thanks again for your help.

Re: Odds Ratio Proc Logistic Question

Reeza — Tue, 19 Feb 2013 17:30:28 GMT

Show your code and output.

Re: Odds Ratio Proc Logistic Question

david_mener_gmail_com — Tue, 19 Feb 2013 17:46:20 GMT

Code:

Proc Logistic data=NH.DiAudioALQ4YrFFQ2YrDepLimited descending;

class bptacat (ref='normal') riagendr (ref='1') race (ref='Non-hispanic White') educ (ref='Less than HS grad') dhtn (ref='No') dstroke (ref='No') dm (ref='nondiabetic') nfcsmoke (ref='never') hause (ref='No');

Model MajorDepression=bptacat riagendr age dhtn dstroke race educ dm nfcsmoke hause;

Weight wtmec4yr;

run;

Analysis of Maximum LIkelihood Estimates

Parameter Df Estimate SE Wald Chi-Sq Pr> Chisq

Intercept		1	-7.5198	8.7698	0.7353	0.3912
bptacat	Mild	1	4.9201	8.4785	0.3368	0.5617
bptacat	Moderate	1	5.4493	8.4785	0.4131	0.5204
bptacat	Profound	1	-7.7375	31.6642	0.0597	0.8070
bptacat	Severe	1	-7.5110	14.8127	0.2571	0.6121
age		1	-0.0457	0.000588	6048.8190	<.0001
RIAGENDR	2	1	0.2018	0.00182	12274.5339	<.0001
dhtn	Yes	1	0.0439	0.00184	570.7600	<.0001
dstroke	Yes	1	0.7128	0.00208	117361.331	<.0001
race	Mexican American	1	3.6052	2.2411	2.5879	0.1077
race	Non-hispanic Black	1	2.8109	2.2411	1.5731	0.2098
race	Other - Including Multiracial	1	-11.1484	8.9643	1.5467	0.2136
race	Other Hispanic	1	2.2693	2.2411	1.0254	0.3113
educ	HS grad	1	0.3007	0.00225	17917.2131	<.0001
educ	Some college or more	1	-0.5879	0.00260	51011.4087	<.0001
dm	diabetic	1	-0.0123	0.00193	40.8251	<.0001
nfcsmoke	current	1	0.2691	0.00313	7379.2558	<.0001
nfcsmoke	former	1	-0.2601	0.00245	11235.5209	<.0001
hause	Yes	1	-0.4633	0.00373	15459.4100	<.0001

Odds Ratio Estimates

bptacat Mild vs Normal	1.042	1.034	1.049
bptacat Moderate vs Normal	1.769	1.751	1.786
bptacat Profound vs Normal	<0.001	<0.001	>999.999
bptacat Severe vs Normal	<0.001	<0.001	>999.999
age	0.955	0.954	0.956
RIAGENDR 2 vs 1	1.497	1.486	1.508
dhtn Yes vs No	1.092	1.084	1.100
dstroke Yes vs No	4.160	4.127	4.195
race Mexican American vs Non-hispanic White	3.134	3.101	3.167
race Non-hispanic Black vs Non-hispanic White	1.416	1.402	1.430
race Other - Including Multiracial vs Non-hispanic White	<0.001	<0.001	>999.999
race Other Hispanic vs Non-hispanic White	0.824	0.799	0.849
educ HS grad vs Less than HS grad	1.014	1.006	1.021
educ Some college or more vs Less than HS grad	0.417	0.413	0.421
dm diabetic vs nondiabetic	0.976	0.968	0.983
nfcsmoke current vs never	1.321	1.308	1.333
nfcsmoke former vs never	0.778	0.772	0.783
hause Yes vs No	0.396	0.390	0.402

Re: Odds Ratio Proc Logistic Question

david_mener_gmail_com — Tue, 19 Feb 2013 17:48:21 GMT

an example would be bptacat for mild: p value in estimate is 0.5617, but the 95% CI for mild vs normal is greater than 1, implying significance.

Re: Odds Ratio Proc Logistic Question

Reeza — Tue, 19 Feb 2013 17:55:21 GMT

You need to add param=ref to the class statement to get reference coding, otherwise you have effect coding.

I don't know if that's the issue though...can you post the output from adding that.

Proc Logistic data=NH.DiAudioALQ4YrFFQ2YrDepLimited descending;

Model MajorDepression=bptacat riagendr age dhtn dstroke race educ dm nfcsmoke hause;

Weight wtmec4yr;

run;

Re: Odds Ratio Proc Logistic Question

Reeza — Tue, 19 Feb 2013 17:57:02 GMT

There's also this little annoying note in the docs:

Caution:PROC LOGISTIC does not compute the proper variance estimators if you are analyzing survey data and specifying the sampling weights through the WEIGHT statement. The SURVEYLOGISTIC procedure is designed to perform the necessary, and correct, computations.

Re: Odds Ratio Proc Logistic Question

david_mener_gmail_com — Tue, 19 Feb 2013 18:20:31 GMT

Here's the rest of the output

Any thoughts? I don't think a param=ref is necessary since I have asked sas to compute the model descending and have specified the reference for each variable in the class statement, however, i am still very new at this, so please correct me if i am at all mistaken or making assumptions. thanks again for your help.

The LOGISTIC Procedure

Data Set	NH.DIAUDIOALQ4YRFFQ2YRDEPLIMITED
Response Variable	MajorDepression
Number of Response Levels	2
Weight Variable	wtmec4yr	wtmec4yr
Model	binary logit
Optimization Technique	Fisher's scoring

Number of Observations Read	1175
Number of Observations Used	955
Sum of Weights Read	16270021
Sum of Weights Used	14184719

1	1	38	424187
2	0	917	13760531

Probability modeled is MajorDepression=1.

Note:

220 observations were deleted due to missing values for the response or explanatory variables.

bptacat	Mild	1	0	0	0
	Moderate	0	1	0	0
	Normal	-1	-1	-1	-1
	Profound	0	0	1	0
	Severe	0	0	0	1
RIAGENDR	1	-1
	2	1
race	Mexican American	1	0	0	0
	Non-hispanic Black	0	1	0	0
	Non-hispanic White	-1	-1	-1	-1
	Other - Including Multiracial	0	0	1	0
	Other Hispanic	0	0	0	1
educ	HS grad	1	0
	Less than HS grad	-1	-1
	Some college or more	0	1
dhtn	No	-1
	Yes	1
dstroke	No	-1
	Yes	1
dm	diabetic	1
	nondiabetic	-1
nfcsmoke	current	1	0
	former	0	1
	never	-1	-1
hause	No	-1
	Yes	1

Convergence criterion (GCONV=1E-8) satisfied.

AIC	3813142.2	3495939.9
SC	3813147.0	3496032.3
-2 Log L	3813140.2	3495901.9

Likelihood Ratio	317238.268	18	<.0001
Score	366764.970	18	<.0001
Wald	304524.605	18	<.0001

bptacat	4	14514.4119	<.0001
age	1	6048.8190	<.0001
RIAGENDR	1	12274.5339	<.0001
dhtn	1	570.7600	<.0001
dstroke	1	117361.331	<.0001
race	4	46042.7111	<.0001
educ	2	51550.2043	<.0001
dm	1	40.8251	<.0001
nfcsmoke	2	11605.2709	<.0001
hause	1	15459.4100	<.0001

Likelihood estimates

Intercept		1	-7.5198	8.7698	0.7353	0.3912
bptacat	Mild	1	4.9201	8.4785	0.3368	0.5617
bptacat	Moderate	1	5.4493	8.4785	0.4131	0.5204
bptacat	Profound	1	-7.7375	31.6642	0.0597	0.8070
bptacat	Severe	1	-7.5110	14.8127	0.2571	0.6121
age		1	-0.0457	0.000588	6048.8190	<.0001
RIAGENDR	2	1	0.2018	0.00182	12274.5339	<.0001
dhtn	Yes	1	0.0439	0.00184	570.7600	<.0001
dstroke	Yes	1	0.7128	0.00208	117361.331	<.0001
race	Mexican American	1	3.6052	2.2411	2.5879	0.1077
race	Non-hispanic Black	1	2.8109	2.2411	1.5731	0.2098
race	Other - Including Multiracial	1	-11.1484	8.9643	1.5467	0.2136
race	Other Hispanic	1	2.2693	2.2411	1.0254	0.3113
educ	HS grad	1	0.3007	0.00225	17917.2131	<.0001
educ	Some college or more	1	-0.5879	0.00260	51011.4087	<.0001
dm	diabetic	1	-0.0123	0.00193	40.8251	<.0001
nfcsmoke	current	1	0.2691	0.00313	7379.2558	<.0001
nfcsmoke	former	1	-0.2601	0.00245	11235.5209	<.0001
hause	Yes	1	-0.4633	0.00373	15459.4100	<.0001

Odds Ratio Estimates with Confidence Intervals

bptacat Mild vs Normal	1.042	1.034	1.049
bptacat Moderate vs Normal	1.769	1.751	1.786
bptacat Profound vs Normal	<0.001	<0.001	>999.999
bptacat Severe vs Normal	<0.001	<0.001	>999.999
age	0.955	0.954	0.956
RIAGENDR 2 vs 1	1.497	1.486	1.508
dhtn Yes vs No	1.092	1.084	1.100
dstroke Yes vs No	4.160	4.127	4.195
race Mexican American vs Non-hispanic White	3.134	3.101	3.167
race Non-hispanic Black vs Non-hispanic White	1.416	1.402	1.430
race Other - Including Multiracial vs Non-hispanic White	<0.001	<0.001	>999.999
race Other Hispanic vs Non-hispanic White	0.824	0.799	0.849
educ HS grad vs Less than HS grad	1.014	1.006	1.021
educ Some college or more vs Less than HS grad	0.417	0.413	0.421
dm diabetic vs nondiabetic	0.976	0.968	0.983
nfcsmoke current vs never	1.321	1.308	1.333
nfcsmoke former vs never	0.778	0.772	0.783
hause Yes vs No	0.396	0.390	0.402

Percent Concordant	67.4	Somers' D	0.371
Percent Discordant	30.4	Gamma	0.379
Percent Tied	2.2	Tau-a	0.028
Pairs	34846	c	0.685

Re: Odds Ratio Proc Logistic Question

Reeza — Tue, 19 Feb 2013 18:26:25 GMT

It is necessary. Look at your design matrix, after

Note:220 observations were deleted due to missing values for the response or explanatory variables.

Note the 1 vs -1 for the parametrization, rather than what you'd expect. Anyways, I think the issue is with the weights statement instead (see last message).Try surveylogistic instead of proc logistic