Transform the coefficient and confidence intervals in texreg output

Question

I am fitting several logistic regression models, and trying to use the package texreg to create a nicely looked table to present all models.

As I know, texreg::screenreg can only report the coefficients (the betas) and the correspondings CIs, but for logistic regression, but it is more common to report the exponent of the coefficients (the odds ratios).

I know that I can use override.coef, override.ci.low and override.ci.up to get what I want, but the output table is not desirable, it gives an asterisk if a CI cover 0 which should be 1 after transformation.

Is there a better and simpler way to transform the coefficients and the CIs? Also, is it possible for me to override the asterisk, I want to provide asterisk to represent the magnitude of p-value (*** p < 0.001, ** p < 0.01, * p < 0.05) ? Thanks!

Here is what I have tried

> set.seed(123)
> x1 <- rnorm(1000)
> x2 <- rnorm(1000)
> y <- runif(1000) < (1 / (1 + exp(-(0.3 + 0.5*x1))))
> mod1 <- glm(y~x1, binomial())
> mod2 <- glm(y~x2, binomial())
> mod3 <- glm(y~x1+x2, binomial())
> 
> tex1 <- extract(mod1)
> tex2 <- extract(mod2)
> tex3 <- extract(mod3)
> 
> screenreg(list(tex1, tex2, tex3), ci.force=T)

==========================================================
                Model 1       Model 2        Model 3      
----------------------------------------------------------
(Intercept)        0.30 *        0.28 *         0.30 *    
                [0.17; 0.43]  [ 0.15; 0.41]  [ 0.17; 0.43]
x1                 0.60 *                       0.60 *    
                [0.45; 0.74]                 [ 0.45; 0.74]
x2                               0.05           0.01      
                              [-0.07; 0.18]  [-0.12; 0.14]
----------------------------------------------------------
AIC             1294.48       1369.92        1296.47      
BIC             1304.30       1379.74        1311.19      
Log Likelihood  -645.24       -682.96        -645.23      
Deviance        1290.48       1365.92        1290.47      
Num. obs.       1000          1000           1000         
==========================================================
* 0 outside the confidence interval

After overriding,

> tex1@coef <- exp(tex1@coef)
> tex2@coef <- exp(tex2@coef)
> tex3@coef <- exp(tex3@coef)
> 
> ci1 <- confint(mod1)
Waiting for profiling to be done...
> ci2 <- confint(mod2)
Waiting for profiling to be done...
> ci3 <- confint(mod3)
Waiting for profiling to be done...
> 
> tex1@ci.low <- exp(ci1[, 1])
> tex2@ci.low <- exp(ci2[, 1])
> tex3@ci.low <- exp(ci3[, 1])
> tex1@ci.up <- exp(ci1[, 2])
> tex2@ci.up <- exp(ci2[, 2])
> tex3@ci.up <- exp(ci3[, 2])
> 
> screenreg(list(tex1, tex2, tex3))

========================================================
            Model 1       Model 2       Model 3     
--------------------------------------------------------
(Intercept)        1.34 *        1.32 *        1.34 *   
                [1.18; 1.53]  [1.17; 1.50]  [1.18; 1.53]
x1                 1.81 *                      1.81 *   
                [1.58; 2.10]                [1.58; 2.10]
x2                               1.05 *        1.01 *   
                              [0.93; 1.19]  [0.89; 1.15]
--------------------------------------------------------
AIC             1294.48       1369.92       1296.47     
BIC             1304.30       1379.74       1311.19     
Log Likelihood  -645.24       -682.96       -645.23     
Deviance        1290.48       1365.92       1290.47     
Num. obs.       1000          1000          1000        
========================================================
* 0 outside the confidence interval

Just rewrite the function to do what you want. That's the beauty of OSS. — IRTFM, Dec 23 '15 at 04:54

IRTFM · Accepted Answer · 2016-01-03T17:44:54.577

There is a ci.test parameter which can be set as the "null value" as would be appropriate for the transformed parameters in this case. It should be set at 1.0 rather than 0. So this succeeds:

  screenreg(list(tex1, tex2, tex3), ci.test=1)

#------output--------
========================================================
                Model 1       Model 2       Model 3     
--------------------------------------------------------
(Intercept)        1.34 *        1.32 *        1.34 *   
                [1.18; 1.53]  [1.17; 1.50]  [1.18; 1.53]
x1                 1.81 *                      1.81 *   
                [1.58; 2.10]                [1.58; 2.10]
x2                               1.05          1.01     
                              [0.93; 1.19]  [0.89; 1.15]
--------------------------------------------------------
AIC             1294.48       1369.92       1296.47     
BIC             1304.30       1379.74       1311.19     
Log Likelihood  -645.24       -682.96       -645.23     
Deviance        1290.48       1365.92       1290.47     
Num. obs.       1000          1000          1000        
========================================================
* 1 outside the confidence interval

Notice that 2 of the 6 parameter estimates are no longer starred.

Thanks! Just curious, is there another way to do the same thing but not using the override method? — pe-perry, Dec 23 '15 at 06:42
I read the source for `extract.glm` and the package vignette/JSS article and do not see any transform methods for coefficients. It should be possible to write a function that did all three steps on a single texreg object and then use lapply across a list of them. That would be more the R-way. I wonder if the package author might accept it as a feature request. I get the impression from the JSS article that he intends that the package have the maximum generalizability. — IRTFM, Dec 23 '15 at 17:04

Transform the coefficient and confidence intervals in texreg output

1 Answers1