Question

I was wondering about some things in dummy.coef() which converts the estimated parameters (contrasts) in ANOVA models to the original ones. It only works for univariate models, but the changes required to also make it work for multivariate models seem minor. In dummy.coef.lm():

• line 52 coef <- object$coefficients would have to be coef <- as.matrix(object$coefficients) to accomodate univariate and multivariate models (coef(object) is a vector in the 1st case, and a matrix in the 2nd)
• line 60 ans <- drop(mm[rn == tl[j], keep, drop = FALSE] %*% coef[keep]) would have to be ans <- drop(mm[rn == tl[j], keep, drop = FALSE] %*% coef[keep, ]) to keep all columns in coef
• line 61 names(ans) <- rnn[rn == tl[j]] could be names(ans) <- rep(rnn[rn == tl[j]], ncol(coef)) to give names to the rows of all columns

The printing method would need some changes, but that seems to be it. Does anybody know why dummy.coef() was not designed to handle multivariate models?

Another thing I stumbled upon: Lines 20-22 are

for (i in vars) args[[i]] <- if (nxl[[i]] == 1)
    rep.int(1, nl)
else factor(rep.int(xl[[i]][1L], nl), levels = xl[[i]])

Is that safe? I.e., if the if() clause is TRUE, wouldn’t there be an unexpected else? I would have expected something like

for (i in vars) args[[i]] <- if (nxl[[i]] == 1) {
    rep.int(1, nl)
} else { factor(rep.int(xl[[i]][1L], nl), levels = xl[[i]]) }

Answer 1

This only addresses your second question (the one about the if() x else y statement at lines 20-22 in the code).

To start with, try cut-and-pasting these two blocks into an R session:

test <- TRUE

# Block 1 -- Doesn't work
if(test) 
    cat("test is TRUE\n")
    else
    cat("test is FALSE\n")

# Block 2 -- Works
{
if(test) 
    cat("test is TRUE\n")
    else
    cat("test is FALSE\n")
}

What’s going on? The {} make all the difference here. Block 1 ‘read-parse-evaluates’ the code line-by-line, causing just the problem you’d expect. Block 2, on the other hand, is read and completely parsed before any evaluation takes place. That’s part of what {} directs R to do. When it receives the block of code as a whole, the parser clearly parses the if() x else y block as one expression.

The code you quoted was taken from inside the body of a function (i.e. inside of a {} pair). In that context, it is handled correctly (i.e. like Block 2).

HTH