Complex Samples - Odds Ratios, where are the p-values hiding?

[mcrow87]

Hi There!

I keep hitting my head on the wall with SPSS Complex Samples & Odds Ratios.

- In the Odds Ratios output, all that is provided are the Odds Ratios and 95% CI. Some of these answers do not include OR = 1 and thus are significant, but I cannot report that alone. I need the p-value - do you know where these can be found?

- In the parameters output table, I have predictor factors with multiple levels (i.e. Age strata 1. 18-30, 2. 31-60, 3. 61-85). Why is every last level deemed "redundant" and set at 0?

- In the parameters output table, the predictor factors' exp(B) for multiple levels are different than the Odds Ratios. Do you know why the exp(B) are different than the formal "Odds Ratios" outputs?

Thanks for your time, all.

[Bruce Weaver]

DISCLAIMER:  I have never used CSLOGISTIC.  But...have you tried adding EXP to the /STATISTICS sub-command?  According to the FM (see link below), this will add exponentiated coefficient estimates (i.e., odds ratios) to your table of parameter estimates.  Presumably, that table also has a p-value column (or Sig., as SPSS labels it).  If you include CINTERVAL on the /STATISTICS sub-command, it should show the confidence intervals too.  

http://www-01.ibm.com/support/knowledgecenter/SSLVMB_22.0.0/com.ibm.spss.statistics.reference/spss/complex_samples/syn_cslogistic_statistics.htm

HTH.

mcrow87 wrote

Hi There!

I keep hitting my head on the wall with SPSS Complex Samples & Odds Ratios.

- In the Odds Ratios output, all that is provided are the Odds Ratios and 95% CI. Some of these answers do not include OR = 1 and thus are significant, but I cannot report that alone. I need the p-value - do you know where these can be found?

- In the parameters output table, I have predictor factors with multiple levels (i.e. Age strata 1. 18-30, 2. 31-60, 3. 61-85). Why is every last level deemed "redundant" and set at 0?

- In the parameters output table, the predictor factors' exp(B) for multiple levels are different than the Odds Ratios. Do you know why the exp(B) are different than the formal "Odds Ratios" outputs?

Thanks for your time, all.

[mcrow87]

Thanks for your reply!

I am getting the EXP values with p-values. When you run complex samples, you can also obtain separate Odds Ratios output relative to specified reference levels within your factors. In my case, my EXPs are different than the OR output. In my Parameters table, I am also getting one “redundant parameter” per factor. Each of my factors have a minimum of 2 levels (sometimes up to 5 with my ‘ethnicity’ variable). I’m not sure if SPSS is automatically assigning a reference class to one level (seems to be the “lowest” level) - there is no option to change this in the Wizard. 

From: "Bruce Weaver [via SPSSX Discussion]"
Date: Thursday, October 15, 2015 at 3:09 PM
To: Matthew Crowson
Subject: Re: Complex Samples - Odds Ratios, where are the p-values hiding?

DISCLAIMER:  I have never used CSLOGISTIC.  But...have you tried adding EXP to the /STATISTICS sub-command?  According to the FM (see link below), this will add exponentiated coefficient estimates (i.e., odds ratios) to your table of parameter estimates.  Presumably, that table also has a p-value column (or Sig., as SPSS labels it).  If you include CINTERVAL on the /STATISTICS sub-command, it should show the confidence intervals too.  

http://www-01.ibm.com/support/knowledgecenter/SSLVMB_22.0.0/com.ibm.spss.statistics.reference/spss/complex_samples/syn_cslogistic_statistics.htm

HTH.

mcrow87 wrote

Hi There!

I keep hitting my head on the wall with SPSS Complex Samples & Odds Ratios.

- In the Odds Ratios output, all that is provided are the Odds Ratios and 95% CI. Some of these answers do not include OR = 1 and thus are significant, but I cannot report that alone. I need the p-value - do you know where these can be found?

- In the parameters output table, I have predictor factors with multiple levels (i.e. Age strata 1. 18-30, 2. 31-60, 3. 61-85). Why is every last level deemed "redundant" and set at 0?

- In the parameters output table, the predictor factors' exp(B) for multiple levels are different than the Odds Ratios. Do you know why the exp(B) are different than the formal "Odds Ratios" outputs?

Thanks for your time, all.

---

If you reply to this email, your message will be added to the discussion below:

http://spssx-discussion.1045642.n5.nabble.com/Complex-Samples-Odds-Ratios-where-are-the-p-values-hiding-tp5730771p5730773.html

To unsubscribe from Complex Samples - Odds Ratios, where are the p-values hiding?, click here.
NAML

[Bruce Weaver]

Oops, never mind.  I've just looked at your original post again, and this time noticed that you said:

"- In the parameters output table, the predictor factors' exp(B) for multiple levels are different than the Odds Ratios. Do you know why the exp(B) are different than the formal "Odds Ratios" outputs?"

So clearly, you ~are~ seeing Exp(B) in your table of parameters, which means you did have EXP on your /STATISTICS sub-command.  (I should have read more carefully before responding.)  

If there truly is no option to include p-values in the output from /ODDSRATIOS, you could always use OMS to send that output to another data set, and then compute the p-values yourself.  The steps would be something like this, I think (with variable names modified as necessary):

COMPUTE #LogOR = ln(OddsRatio).
COMPUTE #LogUL = ln(Upper).  
* The 1.96 in the next line assumes a 95% CI.
COMPUTE #SE = (#LogUL - LogOR) / 1.96.
COMPUTE #zNeg = ABS(#LogOR) / #SE * -1.
COMPUTE p = CDFNORM(#zNeg)*2.

HTH.

Bruce Weaver wrote

DISCLAIMER:  I have never used CSLOGISTIC.  But...have you tried adding EXP to the /STATISTICS sub-command?  According to the FM (see link below), this will add exponentiated coefficient estimates (i.e., odds ratios) to your table of parameter estimates.  Presumably, that table also has a p-value column (or Sig., as SPSS labels it).  If you include CINTERVAL on the /STATISTICS sub-command, it should show the confidence intervals too.  

http://www-01.ibm.com/support/knowledgecenter/SSLVMB_22.0.0/com.ibm.spss.statistics.reference/spss/complex_samples/syn_cslogistic_statistics.htm

HTH.

mcrow87 wrote

Hi There!

I keep hitting my head on the wall with SPSS Complex Samples & Odds Ratios.

- In the Odds Ratios output, all that is provided are the Odds Ratios and 95% CI. Some of these answers do not include OR = 1 and thus are significant, but I cannot report that alone. I need the p-value - do you know where these can be found?

- In the parameters output table, I have predictor factors with multiple levels (i.e. Age strata 1. 18-30, 2. 31-60, 3. 61-85). Why is every last level deemed "redundant" and set at 0?

- In the parameters output table, the predictor factors' exp(B) for multiple levels are different than the Odds Ratios. Do you know why the exp(B) are different than the formal "Odds Ratios" outputs?

Thanks for your time, all.