turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- Odds Ratio and CL all Equal 1

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

04-26-2011 04:17 PM

What does it mean if the Odds Ratio and 95% Wald Confidence Limits all Equal 1?

I'm using Proc Logistic with a binary response and 11 predictors.

Thanks in advance for any help!

P.S. the predictor in question is quantitative, as are about half of the other predictors. Message was edited by: BTAinVA

I'm using Proc Logistic with a binary response and 11 predictors.

Thanks in advance for any help!

P.S. the predictor in question is quantitative, as are about half of the other predictors. Message was edited by: BTAinVA

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to BTAinVA

04-26-2011 09:35 PM

It is very strange.

If 95% Wald Confidence Limits all contain 1 ,then it only illustrate that your model is not very good, Maybe you miss some very important independent variable to influence the response variable.

Ksharp

If 95% Wald Confidence Limits all contain 1 ,then it only illustrate that your model is not very good, Maybe you miss some very important independent variable to influence the response variable.

Ksharp

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to Ksharp

04-27-2011 08:24 AM

Ksharp,

Thanks for the response. It is odd. There are no warnings or messages in the log. I'm trying to create a predictive model for customer behavior (0 = good, 1 = bad) based on a bunch of demographic data. Var1 in the model is Current Home Value and has a very right-skewed distribution. The dataset has 5429 observations, 2208 observations were deleted due to missing values for the response or explanatory variables.

Below is the log output and the Calculated Odds Ratios:

8 proc logistic data=Testdata5 desc outest=betas covout;

9 class var5 var6 var8 var13 var14;

10 model Cust_Type = var1 var5 var6 var8 var10 var13 var14 var15 var17-var20 var23

11 / selection=backward fast ctable

12 /* slentry=0.05*/

13 slstay=0.1

14 /* details*/

15 lackfit;

16 output out=pred p=phat lower=lcl upper=ucl

17 predprob=(individual crossvalidate);

18 run;

NOTE: PROC LOGISTIC is modeling the probability that Cust_Type=1.

NOTE: Convergence criterion (GCONV=1E-8) satisfied in Step 0.

NOTE: Convergence criterion (GCONV=1E-8) satisfied in Step 1.

NOTE: There were 5429 observations read from the data set TESTDATA5.

NOTE: The data set WORK.BETAS has 25 observations and 29 variables.

NOTE: The data set WORK.PRED has 5429 observations and 41 variables.

NOTE: PROCEDURE LOGISTIC used (Total process time):

real time 5.34 seconds

cpu time 0.54 seconds

Odds Ratio Estimates

Point 95% Wald

Effect Estimate Confidence Limits

var1 1.000 1.000 1.000

var6 M vs U 0.772 0.639 0.932

var6 S vs U 0.724 0.533 0.983

var8 H vs U 0.669 0.410 1.093

var8 R vs U 1.454 0.844 2.506

var10 0.989 0.979 0.999

var13 01 vs 06 1.228 0.883 1.707

var13 02 vs 06 1.579 1.131 2.205

var13 04 vs 06 1.593 1.119 2.270

var13 05 vs 06 1.949 1.258 3.019

var14 0 vs 4 1.778 1.269 2.489

var14 1 vs 4 1.273 0.960 1.689

var14 2 vs 4 1.344 1.027 1.758

var14 3 vs 4 0.902 0.692 1.174

var17 0.249 0.092 0.670

var19 4.860 1.312 18.013

var23 0.984 0.976 0.991

Thanks again for any insight!

Thanks for the response. It is odd. There are no warnings or messages in the log. I'm trying to create a predictive model for customer behavior (0 = good, 1 = bad) based on a bunch of demographic data. Var1 in the model is Current Home Value and has a very right-skewed distribution. The dataset has 5429 observations, 2208 observations were deleted due to missing values for the response or explanatory variables.

Below is the log output and the Calculated Odds Ratios:

8 proc logistic data=Testdata5 desc outest=betas covout;

9 class var5 var6 var8 var13 var14;

10 model Cust_Type = var1 var5 var6 var8 var10 var13 var14 var15 var17-var20 var23

11 / selection=backward fast ctable

12 /* slentry=0.05*/

13 slstay=0.1

14 /* details*/

15 lackfit;

16 output out=pred p=phat lower=lcl upper=ucl

17 predprob=(individual crossvalidate);

18 run;

NOTE: PROC LOGISTIC is modeling the probability that Cust_Type=1.

NOTE: Convergence criterion (GCONV=1E-8) satisfied in Step 0.

NOTE: Convergence criterion (GCONV=1E-8) satisfied in Step 1.

NOTE: There were 5429 observations read from the data set TESTDATA5.

NOTE: The data set WORK.BETAS has 25 observations and 29 variables.

NOTE: The data set WORK.PRED has 5429 observations and 41 variables.

NOTE: PROCEDURE LOGISTIC used (Total process time):

real time 5.34 seconds

cpu time 0.54 seconds

Odds Ratio Estimates

Point 95% Wald

Effect Estimate Confidence Limits

var1 1.000 1.000 1.000

var6 M vs U 0.772 0.639 0.932

var6 S vs U 0.724 0.533 0.983

var8 H vs U 0.669 0.410 1.093

var8 R vs U 1.454 0.844 2.506

var10 0.989 0.979 0.999

var13 01 vs 06 1.228 0.883 1.707

var13 02 vs 06 1.579 1.131 2.205

var13 04 vs 06 1.593 1.119 2.270

var13 05 vs 06 1.949 1.258 3.019

var14 0 vs 4 1.778 1.269 2.489

var14 1 vs 4 1.273 0.960 1.689

var14 2 vs 4 1.344 1.027 1.758

var14 3 vs 4 0.902 0.692 1.174

var17 0.249 0.092 0.670

var19 4.860 1.312 18.013

var23 0.984 0.976 0.991

Thanks again for any insight!

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to BTAinVA

04-27-2011 09:59 PM

Do you try to use exact logistic reg.

notice there are lots of variables for only 2208 obs.

And If you can ,plz keep number of variables as few as you can.

Ksharp

notice there are lots of variables for only 2208 obs.

And If you can ,plz keep number of variables as few as you can.

Ksharp

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to Ksharp

04-28-2011 11:12 AM

Ksharp,

Thanks again for the response. No doubt there could be multicollinearity and a host of other problems with my model. But it turns out that they are not all equal to 1, it just seems to be a matter of rounding. When I use the method here (http://support.sas.com/kb/37/106.html) to get more decimal places displayed in the output, I get the following results:

Effect Response OddsRatioEst LowerCL UpperCL

var1 1 0.9999997933 0.9999992533 1.0000003333

Thanks again for the response. No doubt there could be multicollinearity and a host of other problems with my model. But it turns out that they are not all equal to 1, it just seems to be a matter of rounding. When I use the method here (http://support.sas.com/kb/37/106.html) to get more decimal places displayed in the output, I get the following results:

Effect Response OddsRatioEst LowerCL UpperCL

var1 1 0.9999997933 0.9999992533 1.0000003333

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to BTAinVA

04-28-2011 11:41 AM

Another approach is to rescale the predictor variable. For example, if the variable is in units of grams, rescale to kilograms.

Or use the UNITS statement, which accomplishes the same thing with more flexibility.

HTH,

Susan

Or use the UNITS statement, which accomplishes the same thing with more flexibility.

HTH,

Susan

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to Susan

04-28-2011 12:42 PM

Susan,

Thanks for the reply. I just tried rescaling var1 (home value) dividing it by 1000. It helped a little. The CL are now 0.999 and 1.000

Thanks for the reply. I just tried rescaling var1 (home value) dividing it by 1000. It helped a little. The CL are now 0.999 and 1.000

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to BTAinVA

04-29-2011 02:32 AM

Hi.

You talked about multicollinearity.But in logistic model ,there will not be consider in general,because we assume the residual have like binary distribution,so it is hard to measure multicollinearity.

But Another alternation way is that Maybe you can try to use multi-variable linear regression(encode the category variable with some proper method). Be honest, your result is really looks curious.

PS: using logic link function to map 0 and 1 into infinity

Ksharp Message was edited by: Ksharp

You talked about multicollinearity.But in logistic model ,there will not be consider in general,because we assume the residual have like binary distribution,so it is hard to measure multicollinearity.

But Another alternation way is that Maybe you can try to use multi-variable linear regression(encode the category variable with some proper method). Be honest, your result is really looks curious.

PS: using logic link function to map 0 and 1 into infinity

Ksharp Message was edited by: Ksharp