Using some sort of bind() function to create a data.frame with coefficients from nested models?

Question

I think I'm on to a good idea, but don't know how to implement it. I am currently running a number of nested models. For example:

y<-c(1, 0, 1, 1, 1, 1, 1, 1, 0)
x1<-c(1, 0, 9, 9, 1, 9, 2, 1, 0)
x2<-c(1, 2, 3, 5, 4, 2, 4, 5, 1)
x3<-c(1, 2, 1, 4, 1, 3, 4, 8, 3)

model1 <-lm(y~x1)
model2 <-lm(y~x1+x2)
model3 <-lm(y~x1+x2+x3)

model1_output<-(summary(model1 )$coefficients[1])
model2_output<-(summary(model1 )$coefficients[1])
model3_output<-(summary(model1 )$coefficients[1])

I would then like to put all of my output in a data table that matches along the rows (variable names) but inserts the new coefficients in their own columns. I'd like the data.frame presenting the output from my example as:

             b(Model 1)  b(Model 2) b(Model 3)
 (Intercept)  0.59217    0.2555     0.27983
 x1           0.05220    0.04116    0.0375
 x2              NA      0.12530    0.15142
 x3              NA      NA        -0.02994

I'm sure there must be some smart way to do this using the plyr() package (or some other package!), but can't seem to figure it out. Thanks!

This answer is very relevant and likely more flexible than some of what has been discussed here: http://stackoverflow.com/questions/16962576/how-can-i-rbind-vectors-matching-their-column-names — Brandon, Feb 18 '15 at 22:29

score 4 · Accepted Answer · edited May 23 '17 at 11:44

Using this answer from another question:

cbind.fill<-function(...){
    nm <- list(...) 
    nm<-lapply(nm, as.matrix)
    n <- max(sapply(nm, nrow)) 
    do.call(cbind, lapply(nm, function (x) 
    rbind(x, matrix(, n-nrow(x), ncol(x))))) 
}

You could do this:

do.call(cbind.fill,lapply(list(model1,model2,model3),function(x){coef(x)}))

                  [,1]       [,2]        [,3]
(Intercept) 0.59216966 0.25551154  0.27982881
x1          0.05220228 0.04116431  0.03754457
                    NA 0.12530141  0.15142340
                    NA         NA -0.02993768

And then set the row and column names manually. (Note that this returns a matrix, not a data frame, if that matters to you.)

Quick question--How would I add multiple columns relating to my coefficients? I tried: `do.call(cbind.fill,lapply(list(model1,model2,model3),function(x){coef[1:4](x)}))` but that didn't seem to work. — roody, Jul 27 '12 at 18:17

score 0 · Answer 2 · edited May 23 '17 at 11:59

Try the following. I tried to keep it as general as possible, and borrowed some help from here.

models <- list(model1=model1,model2=model2,model3=model3)    

#Find out which model (by index) has the most coefficients.
max.model <- which.max(unlist(lapply(models, function(x) length(x[[1]]))))

#How many coefficients there are in the 'biggest' model (including intercept)
max <- length(models[[max.model]][[1]])

#Create list of coefficients, replacing blank spaces with NA
coeff <- lapply(models, function(x) c(x[[1]],rep(NA,max-length(x[[1]]))))

#See the link above
result <- do.call(cbind,coeff)

colnames(result) <- names(models)

rownames(result) <- names(models[[max.model]][[1]])

Result:

> result
                model1     model2      model3
(Intercept) 0.59216966 0.25551154  0.27982881
x1          0.05220228 0.04116431  0.03754457
x2                  NA 0.12530141  0.15142340
x3                  NA         NA -0.02993768

What I mean by trying to keep it general is that this would work with any number of models; only the first line of code needs to change.

Using some sort of bind() function to create a data.frame with coefficients from nested models?

2 Answers2