Using 'mlm` object in `mtable` output

Question

Is there any way to work with mlm objects in mtable from the memisc package?

Without using multiple response matrix, what I want is something like:

library(car)
library(memisc)
lm1 = lm(Sepal.Length ~ Petal.Length + Petal.Width + Species, data=iris)
lm2 = lm(Sepal.Width ~ Petal.Length + Petal.Width + Species, data=iris)
mtable(lm1, lm2)

which produces

Calls:
  lm1: lm(formula = Sepal.Length ~ Petal.Length + Petal.Width + Species, 
          data = iris)
lm2: lm(formula = Sepal.Width ~ Petal.Length + Petal.Width + Species, 
        data = iris)

===============================================
  lm1       lm2   
-----------------------------------------------
  (Intercept)                  3.683***  3.048***
  (0.107)   (0.094)  
Petal.Length                 0.906***  0.155*  
  (0.074)   (0.065)  
Petal.Width                 -0.006     0.623***
  (0.156)   (0.136)  
Species: versicolor/setosa  -1.598*** -1.764***
  (0.206)   (0.180)  
Species: virginica/setosa   -2.113*** -2.196***
  (0.304)   (0.265)  
-----------------------------------------------
  R-squared                      0.837     0.551 
adj. R-squared                 0.832     0.539 
sigma                          0.339     0.296 
F                            185.769    44.496 
p                              0.000     0.000 
Log-likelihood               -48.116   -27.711 
Deviance                      16.681    12.708 
AIC                          108.231    67.423 
BIC                          126.295    85.486 
N                            150       150     
===============================================

But:

mlmIris = lm(cbind(Sepal.Length, Sepal.Width) ~ Petal.Length + Petal.Width + Species, data=iris)
mtable(mlmIris)

produces

Error in qt(p = alpha/2, df = dendf) : 
  Non-numeric argument to mathematical function

I'm not going to reproduce the ways I've tried to extract an lm object that I can use in mtable. Suffice it to say, none of them worked.

Answer 1

As suggested in the comments you need to write a getSummary method for mlm objects because there currently isn't any. The lm method that you get by inheritance does not work.

I have taken a quick stab at this and provide a getSummary.mlm below along with a suitable setSummaryTemplate. The coefficients and standard errors are now correctly extracted. What needs more work is the extraction of summary statistics (R-squared, residual sum of squares, F statistic, etc.) which I didn't work on. It should give you a good start though. If you improve the method further, please also consider providing it to Martin Elff (the memisc maintainer) so that it is directly available in memisc.

After sourcing the functions provided below, this works:

R> mtable(mlmIris)

Calls:
mlmIris: lm(formula = cbind(Sepal.Length, Sepal.Width) ~ Petal.Length + 
    Petal.Width + Species, data = iris)

=====================================================
                            Sepal.Length Sepal.Width 
-----------------------------------------------------
(Intercept)                   3.683***     3.048***  
                             (0.107)      (0.094)    
Petal.Length                  0.906***     0.155*    
                             (0.074)      (0.065)    
Petal.Width                  -0.006        0.623***  
                             (0.156)      (0.136)    
Species: versicolor/setosa   -1.598***    -1.764***  
                             (0.206)      (0.180)    
Species: virginica/setosa    -2.113***    -2.196***  
                             (0.304)      (0.265)    
-----------------------------------------------------
N                               150                  
=====================================================

The source code is:

getSummary.mlm <- function(obj, alpha = 0.05, ...)
{
## extract coefficient summary
cf <- lapply(summary(mlmIris), "[[", "coefficients")
## augment with confidence intervals
cval <- qnorm(1 - alpha/2)
for(i in seq_along(cf)) cf[[i]] <- cbind(cf[[i]],
  cf[[i]][, 1] - cval * cf[[i]][, 2],
  cf[[i]][, 1] + cval * cf[[i]][, 2])
## collect in array
nam <- unique(unlist(lapply(cf, rownames)))
acf <- array(dim = c(length(nam), 6, length(cf)),
  dimnames = list(nam, c("est", "se", "stat", "p", "lwr", "upr"), colnames(coef(obj))))
for(i in seq_along(cf)) acf[rownames(cf[[i]]), , i] <- cf[[i]]

## return everything
return(list(
  coef = acf,
  sumstat = c(
    "N" = nobs(obj)
  ),
  contrasts = obj$contrasts,
  xlevels = obj$xlevels,
  call = obj$call
))
}

setSummaryTemplate("mlm" = c(
  "N" = "($N:d)"
))