Correlograms of Moran's I with progressive Bonferroni correction
Hi Roger, According to the book "Numerical Ecology" by Legendre&Legendre in 2012, a progressive (sequential) Bonferroni correction is a modified version of Bonferroni correction, where the Bonferroni-corrected level is computed for each distance class separately instead of using a consistent level for all distance classes. This is done by using for each distance class the number of tests actually performed up to that distance class. For example, if we have 10 distance classes, and we use a=0.05 as a probability level, then the adjusted a (a') for the first distance class is a'=a/1=0.05, for the second distance class is a'=a/2=0.025 ... for the 10th distance class is a'=a/10=0.005. I've checked the mpmcorrelogram package, it seems that this package is for creating multivariate correlograms. But what I hope to get is correlograms of autocorrelation over one variate (e.g. Moran's I). Hope this is clearer. Thank you for your help. Mingke
On Mon, Mar 18, 2019 at 5:41 AM Roger Bivand <Roger.Bivand at nhh.no> wrote:
On Sun, 17 Mar 2019, Mingke Li wrote:
Dear list, I am trying to generate a correlogram of Moran's I, with symbols showing
if
the coefficient is significant or not after the progressive Bonferroni correction. Now I'm using the "correlog" function in the package ncf to calculate the coefficients among all distance classes. I symbolized the dots in the plot to show if they are significant or not (after
progressive
Bonferroni correction), say, open circles for non-significant
coefficients,
and solid circles for significant coefficients.
Explain what you mean here by progressive Bonferroni correction, please. Could you rather see what is done in the mpmcorrelogram package, and report back if it makes sense? Roger
These are the codes I used in R:
correlog.L2.2014new <- correlog(Grid.2014.all$Lng, Grid.2014.all$Lat,
Grid.2014.all$L2_Average,latlon = T, na.rm=T, increment=5, resamp=200)
for (i in list(1:length(correlog.L2.2014new$p))) {
correlog.L2.2014new$adjp[i]=correlog.L2.2014new$p[i]*i }
plot(correlog.L2.2014new$correlation[1:80], type="b", cex=1.5, lwd=1.5,
ylim=c(-0.1,0.7),
pch=ifelse(((correlog.L2.2014new$adjp>0.05)),1,16),
xlab="Distance (km)", ylab="Moran's I", cex.lab=2, cex.axis=1.5);
abline(h=0)
I also found the function "sp.correlogram" in the package spdep to make
the
correlogram with Bonferroni correction. Now I have three questions: 1. Can I set a "up-limit" to the "correlog" function? For example, if I just need the first 80 distance classes to generate the graph (as shown
in
the plot code), is it possible to let the "correlog" function skip the
rest
distance classes to save the computation time? 2. Is my code of generating p values after the progressive Bonferroni correction correct? If not, how should I fix it? 3. Now I'm kind of correcting the p value manually based on my understanding of progressive Bonferroni correction. Is there any function in any R packages for creating correlograms with progressive Bonferroni correction? "sp.correlogram" has a parameter setting "p.adj.method" but
it
seems do not have a progressive Bonferroni correction option?
I just start to work with the correlograms recently; any advice or
solutions are welcome. Thank you in advance!
Erin
[[alternative HTML version deleted]]
_______________________________________________ R-sig-Geo mailing list R-sig-Geo at r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-geo
-- Roger Bivand Department of Economics, Norwegian School of Economics, Helleveien 30, N-5045 Bergen, Norway. voice: +47 55 95 93 55; e-mail: Roger.Bivand at nhh.no https://orcid.org/0000-0003-2392-6140 https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en