Relative Risk instead of Odds in Logistic Regression (or other procedure)?

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

Relative Risk instead of Odds in Logistic Regression (or other procedure)?

Ian Martin-3
Hi All,

It’s my understanding that reporting Odds Ratio (OR) for a common binomial outcome is less desirable — and misleading.   Rather, I should be reporting Relative Risk (RR) for a disease outcome with a frequency of about 20% in a population.  

I can’t use CROSSTABS to get RR because my independent variables are continuous, and can’t see how to get REGRESSION>Binary Logistic to produce RR instead of OR.

Any suggestions or workarounds?

TIA,
Ian


Ian D. Martin, Ph.D.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
Reply | Threaded
Open this post in threaded view
|

Re: Relative Risk instead of Odds in Logistic Regression (or other procedure)?

Bruce Weaver
Administrator
Hi Ian.  Try GENLIN with a log-link binomial error distribution. Here's an
example from one of my files:

GENLIN default2 (REFERENCE=LAST) BY ed2 (ORDER=ASCENDING)
  /MODEL ed2 INTERCEPT=YES
 DISTRIBUTION=BINOMIAL LINK=LOG
  /CRITERIA METHOD=FISHER(1) SCALE=1 COVB=MODEL
    MAXITERATIONS=100 MAXSTEPHALVING=5
    PCONVERGE=1E-006(ABSOLUTE) SINGULAR=1E-012
    ANALYSISTYPE=3(WALD) CILEVEL=95 CITYPE=WALD
    LIKELIHOOD=FULL
  /MISSING CLASSMISSING=EXCLUDE
  /PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION (EXPONENTIATED).

Although it is a Stata resource, this UCLA page suggests another approach:
Poisson regression with a robust covariance matrix.

https://stats.idre.ucla.edu/stata/faq/how-can-i-estimate-relative-risk-using-glm-for-common-outcomes-in-cohort-studies/

HTH.



Ian Martin-3 wrote

> Hi All,
>
> It’s my understanding that reporting Odds Ratio (OR) for a common binomial
> outcome is less desirable — and misleading.   Rather, I should be
> reporting Relative Risk (RR) for a disease outcome with a frequency of
> about 20% in a population.  
>
> I can’t use CROSSTABS to get RR because my independent variables are
> continuous, and can’t see how to get REGRESSION>Binary Logistic to produce
> RR instead of OR.
>
> Any suggestions or workarounds?
>
> TIA,
> Ian
>
>
> Ian D. Martin, Ph.D.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to

> LISTSERV@.UGA

>  (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD





-----
--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

--
Sent from: http://spssx-discussion.1045642.n5.nabble.com/

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
--
Bruce Weaver
bweaver@lakeheadu.ca
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.
Reply | Threaded
Open this post in threaded view
|

Re: Relative Risk instead of Odds in Logistic Regression (or other procedure)?

Rich Ulrich
Bruce - That's an interesting article. I'm impressed that mis-specifying the
model as Poisson or RR in that mid-range has so little effect on the solution.
Despite the original questioner, or that article, I generally recommend against
ever using RR.

I think that there are two questions - What do we solve? and What do we report?
  - That paper shows how to get out the RR, if that is what you want to report.

I think that I will always want to know that my solution used the logically-proper
model, the logistic, with the OR obtained and noted. The only powerful inducement
to use the RR is that it has a commonplace interpretation; its best justification is that
for very low fractions, it is similar to the OR. The fact is, the divergence (OR from RR)
at higher levels has the unfortunate effect of /mis-representing/ the contrasts of
effects, when results are compared across several studies or outcomes.  - Teach your
audience the OR, and use the OR.

If you don't have comparisons in mind, then you are just looking at the one study,
and if you think that the OR does not serve, I think you should probably use the raw numbers (or point estimates) to make your point.  "Lying with statistics" is what
comes to mind when I think of an RR computed, say, as 0.99/0.95, to represent the
OR which is near 0.2 (or 5.0, as reciprocal).

Rich Ulrich


From: SPSSX(r) Discussion <[hidden email]> on behalf of Bruce Weaver <[hidden email]>
Sent: Friday, November 9, 2018 7:25 PM
To: [hidden email]
Subject: Re: Relative Risk instead of Odds in Logistic Regression (or other procedure)?
 
Hi Ian.  Try GENLIN with a log-link binomial error distribution. Here's an
example from one of my files:

GENLIN default2 (REFERENCE=LAST) BY ed2 (ORDER=ASCENDING)
  /MODEL ed2 INTERCEPT=YES
 DISTRIBUTION=BINOMIAL LINK=LOG
  /CRITERIA METHOD=FISHER(1) SCALE=1 COVB=MODEL
    MAXITERATIONS=100 MAXSTEPHALVING=5
    PCONVERGE=1E-006(ABSOLUTE) SINGULAR=1E-012
    ANALYSISTYPE=3(WALD) CILEVEL=95 CITYPE=WALD
    LIKELIHOOD=FULL
  /MISSING CLASSMISSING=EXCLUDE
  /PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION (EXPONENTIATED).

Although it is a Stata resource, this UCLA page suggests another approach:
Poisson regression with a robust covariance matrix.

https://stats.idre.ucla.edu/stata/faq/how-can-i-estimate-relative-risk-using-glm-for-common-outcomes-in-cohort-studies/

HTH.



Ian Martin-3 wrote
> Hi All,
>
> It’s my understanding that reporting Odds Ratio (OR) for a common binomial
> outcome is less desirable — and misleading.   Rather, I should be
> reporting Relative Risk (RR) for a disease outcome with a frequency of
> about 20% in a population. 
>
> I can’t use CROSSTABS to get RR because my independent variables are
> continuous, and can’t see how to get REGRESSION>Binary Logistic to produce
> RR instead of OR.
>
> Any suggestions or workarounds?
>
> TIA,
> Ian
>
>
> Ian D. Martin, Ph.D.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to

> LISTSERV@.UGA

>  (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD





-----
--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

--
Sent from: http://spssx-discussion.1045642.n5.nabble.com/

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD