Dear all, I am trying to recreate a discriminant analysis in R based on the article from "Dong,J.-J.,etal.,Discriminant analysis of the geomorphic characteristics and stability of landslide dams, Geomorphology (2009)". I used lda (MASS) to determine the discriminant functions but I noticed that it is not the same as in the paper. I have three questions (1) Why does results from lda() does not show a constant? Isnt the discriminant function supposed to be D = a + b1*x1 + b2*x2 + ... + bm*xm? If there is one, where can I find it? (2) Why are the linear discriminant coefficients different from the paper? The discriminant function in the paper is: D = ? 2.62*log10(Peak.flow) ? 4.67*log10(Dam.height) + 4.57*log10(Dam.width) +2.67*log10(Dam.Length) +8.26 (He used SPSS for the analysis) (3) I used manova to perform the Wilks test. However, I am missing the significant values for the Wilks test. How come? I know these are newbie questions but I hope someone out there may have the answer. Thanks all Here is the code I used for the Linear Discriminant Analysis:
criteria<-c("Catchment.area", "Stream.order", "Mean.flow", "Peak.flow","UCG", "DCG","Landslide.volume","Landslide.area","HTD",
"Slope.height","Dam.height", "Dam.width", "Dam.length", "Lake.depth", "Lake.area", "Dam.volume","SClass")
tabcrit<-subset(tabata, rowSums(is.na(tabata[criteria]))==0) tabcrit<-tabcrit[criteria]
stabledams<-subset(tabcrit, SClass=="Stable") unstabledams<-subset(tabcrit, SClass=="Unstable")
st<-sample(nrow(stabledams)) ust<-sample(nrow(unstabledams))
training <- rbind(stabledams[st[1:5], ], unstabledams[ust[1:17],]) tr.lda<-lda(SClass~log10(Catchment.area)+log10(Dam.height)+log10(Dam.width)+log10(Dam.length),
data=training)
tr.lda
Coefficients of linear discriminants:
LD1
log10(Catchment.area) 1.0967609
log10(Dam.height) 0.9818473
log10(Dam.width) -1.9813511
log10(Dam.length) -0.7131808
For the Wilks test:
tr.matrix<-as.matrix(training[-17]) tr.manova<-manova(tr.matrix~training$SClass) tr.wilks<-summary(tr.manova, test="Wilks") tr.wilks
Df Wilks approx F num Df den Df Pr(>F) training$SClass 1 0.29328 0.75303 16 5 0.6979 Residuals 20 Cheers, Julius Tesoro