tidyst_kde_local_test {eks}

Tidy and geospatial kernel density based local two-sample comparison tests

Description

Tidy and geospatial versions of kernel density based local two-sample comparison tests for 1- and 2-dimensional data.

Usage

tidy_kde_local_test(data1, data2, labels, ...)
st_kde_local_test(x1, x2, labels, ...)

Arguments

data1, data2

data frames/tibbles of data values

x1, x2

sf objects with point geometry

labels

flag or vector of strings for legend labels

...

other parameters in ks::kde.local.test function

Details

A kernel local density based two-sample comparison is a modification of the standard kernel density estimate where the two data samples are compared. A Hochberg procedure is employed to control the significance level for multiple comparison tests.

For details of the computation of the kernel local density based two-sample comparison test and the bandwidth selector procedure, see ?ks::kde.local.test. The bandwidth matrix of smoothing parameters is computed as in ks::kde per data sample.

If labels is missing, then the first sample label is taken from x1, and the second sample label from x2. If labels="default" then these are "f1" and "f2". Otherwise, they are assigned to the values of the input vector of strings.

Value

The output has the same structure as the kernel density estimate from *_kde, except that estimate is the difference between the density values data1-data2 rather than the density values, and label becomes an indicator factor of the local comparison test result: "f1<f2" = data1 < data2, 0 = data1 = data2, "f2>f1" = data1 > data2.

The output from st_kde_local_test has two contours, with contlabel=-50 (for f1<f2) and contlabel=50 (for f1>f2), as multipolygons which delimit the significant difference regions.

Examples

## tidy local test between unsuccessful and successful grafts
data(hsct, package="ks")
hsct <- dplyr::as_tibble(hsct)
hsct <- dplyr::filter(hsct, PE.Ly65Mac1 >0 & APC.CD45.2>0)
hsct6 <- dplyr::filter(hsct, subject==6)   ## unsuccessful graft 
hsct6 <- dplyr::select(hsct6, PE.Ly65Mac1, APC.CD45.2)
hsct12 <- dplyr::filter(hsct, subject==12) ## successful graft 
hsct12 <- dplyr::select(hsct12, PE.Ly65Mac1, APC.CD45.2)
t1 <- tidy_kde_local_test(data1=hsct6, data2=hsct12)
gt <- ggplot2::ggplot(t1, ggplot2::aes(x=PE.Ly65Mac1, y=APC.CD45.2)) 
gt + geom_contour_filled_ks() + 
    scale_transparent(colorspace::scale_fill_discrete_qualitative(
        palette="Dark2", rev=TRUE, breaks=c("hsct6<hsct12","hsct6>hsct12"), order=c(2,1,3)))

## geospatial local test between Grevillea species
data(wa)
data(grevilleasf)
hakeoides <- dplyr::filter(grevilleasf, species=="hakeoides")
paradoxa <- dplyr::filter(grevilleasf, species=="paradoxa")
s1 <- st_kde_local_test(x1=hakeoides, x2=paradoxa)

## base R plot
xlim <- c(1.2e5, 1.1e6); ylim <- c(6.1e6, 7.2e6)
plot(wa, xlim=xlim, ylim=ylim)
plot(s1, add=TRUE)

## geom_sf plot
gs <- ggplot2::ggplot(s1) + ggplot2::geom_sf(data=wa, fill=NA)  + 
    ggthemes::theme_map() 
gs + ggplot2::geom_sf(data=st_get_contour(s1), ggplot2::aes(fill=label)) +
    colorspace::scale_fill_discrete_qualitative(palette="Dark2", rev=TRUE) +
    ggplot2::coord_sf(xlim=xlim, ylim=ylim)