Quick multiple ANOVA testing using a matrix as input

[nuno-agostinho]

Useful to quickly calculate multiple Levene tests:

# x is a matrix of values (including NAs) from two or more groups
# g contains the group for each column
leveneTest <- function(x, g, centers=c("median", "mean")) {
    centers <- match.arg(centers)
    dname <- paste(deparse(substitute(x)), "and", deparse(substitute(g)))

    # Convert groups to factors
    g <- factor(g)

    if (centers == "median") {
        res <- calculateRowMediansByGroups(x, g)
    } else if (centers == "mean") {
        res <- calculateRowMeansByGroups(x, g)
    }
    spread <- abs(t(x) - res[rep(g, nrow(x))])

    fit <- lm.fit(model.matrix(~ g), spread)
    # Modified `stats:::anova.lm()` to accept `stats::lm.fit()` with multiple results
    # All the attributes needed to run `stats::anova.lm()` are available in `fit`
    var <- modANOVA(fit)

    # Prepare results
    statistic <- var$`F value`
    pval <- var$`Pr(>F)`

    centers <- deparse(substitute(centers))
    rval <- list(statistic=c("W"=statistic), p.value=pval, data.name=dname,
                 method=paste0("Levene's test (using the ", centers, ")"))
    class(rval) <- "htest"
    return(rval)
}