adonis.R

`adonis2` <-
    function(formula, data, permutations = 999, method = "bray",
             sqrt.dist = FALSE, add = FALSE, by = NULL,
             parallel = getOption("mc.cores"), na.action = na.fail,
             strata = NULL, ...)
{
    ## handle missing data
    if (missing(data))
        data <- model.frame(delete.response(terms(formula)),
                            na.action = na.action)
    ## we accept only by = "terms", "margin", "onedf" or NULL
    if (!is.null(by))
        by <- match.arg(by, c("terms", "margin", "onedf"))
    ## evaluate lhs
    YVAR <- formula[[2]]
    lhs <- eval(YVAR, parent.frame(), environment(formula))
    environment(formula) <- environment()
    ## Take care that input lhs are dissimilarities
    if ((is.matrix(lhs) || is.data.frame(lhs)) &&
        isSymmetric(unname(as.matrix(lhs))))
        lhs <- as.dist(lhs)
    if (!inherits(lhs, "dist"))
        lhs <- vegdist(as.matrix(lhs), method=method, ...)
    ## adjust distances if requested
    if (sqrt.dist)
        lhs <- sqrt(lhs)
    if (is.logical(add) && isTRUE(add))
        add <- "lingoes"
    if (is.character(add)) {
        add <- match.arg(add, c("lingoes", "cailliez"))
        if (add == "lingoes") {
            ac <- addLingoes(as.matrix(lhs))
            lhs <- sqrt(lhs^2 + 2 * ac)
        }
        else if (add == "cailliez") {
            ac <- addCailliez(as.matrix(lhs))
            lhs <- lhs + ac
        }
    }
    ## adonis0 & anova.cca should see only dissimilarities (lhs)
    if (!missing(data)) # expand and check terms
        formula <- terms(formula, data=data)
    if (is.null(attr(data, "terms"))) # not yet a model.frame?
        data <- model.frame(delete.response(terms(formula)), data,
                            na.action = na.action)
    formula <- update(formula, lhs ~ .)
    sol <- adonis0(formula, data = data, method = method)
    ## handle permutations
    perm <- getPermuteMatrix(permutations, NROW(data), strata = strata)
    out <- anova(sol, permutations = perm, by = by,
                 parallel = parallel)
    ## attributes will be lost when adding a new column
    att <- attributes(out)
    ## add traditional adonis output on R2
    out <- rbind(out, "Total" = c(nobs(sol)-1, sol$tot.chi, NA, NA))
    out <- cbind(out[,1:2], "R2" = out[,2]/sol$tot.chi, out[,3:4])
    ## Fix output header to show the adonis2() call instead of adonis0()
    att$heading[2] <- deparse(match.call(), width.cutoff = 500L)
    att$names <- names(out)
    att$row.names <- rownames(out)
    attributes(out) <- att
    out
}

`adonis0` <-
    function(formula, data=NULL, method="bray")
{
    ## First we collect info for the uppermost level of the analysed
    ## object
    Trms <- terms(data)
    sol <- list(call = match.call(),
                method = "adonis",
                terms = Trms,
                terminfo = list(terms = Trms))
    sol$call$formula <- formula(Trms)
    TOL <- 1e-7
    lhs <- formula[[2]]
    lhs <- eval(lhs, environment(formula)) # to force evaluation
    formula[[2]] <- NULL                # to remove the lhs
    rhs <- model.matrix(formula, data) # and finally the model.matrix
    assign <- attr(rhs, "assign") ## assign attribute
    sol$terminfo$assign <- assign[assign > 0]
    rhs <- rhs[,-1, drop=FALSE] # remove the (Intercept) to get rank right
    rhs <- scale(rhs, scale = FALSE, center = TRUE) # center
    qrhs <- qr(rhs)
    ## input lhs should always be dissimilarities
    if (!inherits(lhs, "dist"))
        stop("internal error: contact developers")
    if (any(lhs < -TOL))
        stop("dissimilarities must be non-negative")
    ## if there was an na.action for rhs, we must remove the same rows
    ## and columns from the lhs (initDBRDA later will work similarly
    ## for distances and matrices of distances).
    if (!is.null(nas <- na.action(data))) {
        lhs <- as.matrix(lhs)[-nas,-nas, drop=FALSE]
        n <- nrow(lhs)
    } else
        n <- attr(lhs, "Size")
    ## G is -dmat/2 centred
    G <- initDBRDA(lhs)
    ## preliminaries are over: start working
    Gfit <- qr.fitted(qrhs, G)
    Gres <- qr.resid(qrhs, G)
    ## collect data for the fit
    if(!is.null(qrhs$rank) && qrhs$rank > 0)
        CCA <- list(rank = qrhs$rank,
                    qrank = qrhs$rank,
                    tot.chi = sum(diag(Gfit)),
                    QR = qrhs)
    else
        CCA <- NULL # empty model
    ## collect data for the residuals
    CA <- list(rank = n - max(qrhs$rank, 0) - 1,
               u = matrix(0, nrow=n),
               tot.chi = sum(diag(Gres)))
    ## all together
    sol$tot.chi <- sum(diag(G))
    sol$adjust <- 1
    sol$Ybar <- G
    sol$CCA <- CCA
    sol$CA <- CA
    class(sol) <- c("adonis2", "dbrda", "rda", "cca")
    sol
}