2
$\begingroup$

I am using rms and can't understand the difference between orm and lrm when used with contrasts. For example:

x <- factor(rbinom(100,2,0.6), labels = c("a","b","c"), ordered = TRUE)
y <- factor(rbinom(100,1,0.5), labels=c("no","yes"))
l <- lrm(x~y)
l

Logistic Regression Model

lrm(formula = x ~ y)
                     Model Likelihood     Discrimination    Rank Discrim.    
                        Ratio Test            Indexes          Indexes       
Obs           100    LR chi2      0.51    R2       0.006    C       0.529    
 a             24    d.f.            1    g        0.133    Dxy     0.059    
 b             40    Pr(> chi2) 0.4764    gr       1.143    gamma   0.117    
 c             36                         gp       0.024    tau-a   0.039    
max |deriv| 1e-10                         Brier    0.181                     

      Coef    S.E.   Wald Z Pr(>|Z|)
y>=b   1.0188 0.2988  3.41  0.0007  
y>=c  -0.7162 0.2884 -2.48  0.0130  
y=yes  0.2642 0.3715  0.71  0.4769

o <- orm(x~y)
l;o

Logistic (Proportional Odds) Ordinal Regression Model

orm(formula = x ~ y)
                     Model Likelihood          Discrimination          Rank Discrim.    
                        Ratio Test                 Indexes                Indexes       
Obs           100    LR chi2      0.51    R2                  0.006    rho     0.071    
 a             24    d.f.            1    g                   0.133                     
 b             40    Pr(> chi2) 0.4764    gr                  1.143                     
 c             36    Score chi2   0.51    |Pr(Y>=median)-0.5| 0.259                     
Unique Y        3    Pr(> chi2) 0.4766                                                  
Median Y        2                                                                       
max |deriv| 7e-05                                                                       

      Coef    S.E.   Wald Z Pr(>|Z|)
y>=b   1.0188 0.2988  3.41  0.0007  
y>=c  -0.7162 0.2884 -2.48  0.0130  
y=yes  0.2642 0.3715  0.71  0.4769  

We can see, that results of orm and lrm are equal. But when we use contrasts the results are different:

contrast(l,list(y="no"),list(y="yes"))

     Contrast      S.E.      Lower     Upper     Z Pr(>|z|)
11 -0.2642454 0.3714673 -0.9923081 0.4638172 -0.71   0.4769
Confidence intervals are 0.95 individual intervals

and

contrast(o,list(y="no"),list(y="yes"))

Contrast      S.E.      Lower   Upper    Z Pr(>|z|)
11 0.7545878 0.3714672 0.02652544 1.48265 2.03   0.0422

Confidence intervals are 0.95 individual intervals

Why orm contrast aren't equal beta regression coefficient as lrm contrast?

$\endgroup$
1
$\begingroup$

This is a significant error that has now been fixed on github and will be in the next release to CRAN. It's best to report rms package issues through https://github.com/harrelfe/rms. But thanks for reporting the error. You can get a temporary fix by typing

require(rms)
getRs('contrast.s', grepo='rms', dir='R', put='source')
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.