Re: row weights in co-inertia

From: Stephane Dray (dray@biomserv.univ-lyon1.fr)
Date: Tue Dec 21 2004 - 15:44:19 MET

  • Next message: Frank Dziock: "Re: row weights in co-inertia"

    Hi !

    At 07:55 21/12/2004, you wrote:

    >Dear ADE-colleagues,
    >
    >I am trying to perform a co-inertia in ADE-4 with R 2.0.0 based on a Fuzzy
    >CA of 155 species and 11 trait variables with 59 trait categories. The
    >other datatable has 155 species and 37 sites.
    >
    >FCA was performed on the traits-species table (traits) with traits in
    >columns and species in lines/rows
    >CA was performed on the transposed species-sites table (artentranspo) with
    >sites in columns and species in lines/rows
    >
    >######################
    >blocks <- c(4,11,3,4,3,4,4,7,3,9,7) ##no. of categories per variable
    >traits <- read.table ("c:/1
    >Daten/Sachsen-Anhalt/Ergebnisse/Analysen/traits.txt", sep="\t")
    >
    >artentranspo<-data.frame(t(arten))
    >ca<-dudi.coa(artentranspo, scan=F, nf=2)
    >traits2<-prep.fuzzy.var(traits, blocks, row.w=ca$lw) ## new FCA with row
    >weights ca$lw from the CA
    >
    >fca <- dudi.fca(traits2, scann=F, nf=4)
    >coinert<-coinertia(fca,ca,scannf=F, nf=4)
    >plot(coinert)
    >randtest.coinertia(coinert,nrepet=9999)
    >######################
    >
    >In the manual (p. 52) it says that in "dudi" cw are row weights, but this
    >doesnt work for my coinertia, so I took "lw", which is said to be column
    >weights, which I doubt!?
    >
    >The co-inertia seems to have worked with my data and using ca$li as row
    >weights, because I got a summary plot.

    I am not sure to really understand what you say. It seems there is an error
    which must be corrected but can you tell us about which "manual" are you
    talking ?

    row weights are in lw (line weights)
    column wights in cw (column weights)
    for all objects of the class dudi.
    Your example seems correct, but I suppose that there is a mistake in your
    text when you say ca$li as row weights, it is ca$lw I suppose ??

    >But in the Monte-Carlo test an error occured "non-equal row weights" with
    >eitther setting "fixed=0" or 1 or 2.
    >
    >I would be delighted if someone could help me by telling me what is "lw"
    >and "cw" and maybe a hint on what went wrong in the Monte-Carlo test.

    you do not receive an error but a warning, It is quite different. You must
    fix the table on which you apply COA because row weights are computed from
    this analysis. And the real message that you obtain is:

    > dudi2 <- dudi.coa(doubs$poi, scan = FALSE, nf = 2)
    > dudi1 <- dudi.pca(doubs$mil, scale = TRUE, row.w=dudi2$lw,scan =
    FALSE, nf = 3)
    > coin1 <- coinertia(dudi1,dudi2, scan = FALSE, nf = 2)

    > rnd1=randtest(coin1,fixed=2)
    Warning: non uniform weight. The results from permutations
    are valid only if the row weights come from the fixed table.
    The fixed table is table Y : doubs$poi

    Here is is correct, I fix the table on which coa is applied.

    >Thanks very much in advance,
    >
    >With best wishes from sunny Leipzig,
    >
    >Frank

    Here in Montreal it is sun, snow and cold (-25, -38 with the wind).

    Stéphane DRAY
    --------------------------------------------------------------------------------------------------

    Département des Sciences Biologiques
    Université de Montréal, C.P. 6128, succursale centre-ville
    Montréal, Québec H3C 3J7, Canada

    Tel : (514) 343-6111 poste 1233 Fax : (514) 343-2293
    E-mail : stephane.dray@umontreal.ca
    --------------------------------------------------------------------------------------------------

    Web http://www.steph280.freesurf.fr/
    --------------------------------------------------------------------------------------------------



    This archive was generated by hypermail 2b30 : Tue Dec 21 2004 - 15:44:29 MET