multinomialTS fits state-space models to multinomially distributed data, such as community count data or palaeoecological time series. It estimates taxon-taxon interactions, driver-taxon relationships, and cross-correlated error, and it handles time gaps in the response where covariates are observed.
See the vignettes on the package website on how to use the model.
Quick start
library(multinomialTS)
data(story_pollen_matrix)
data(story_char_matrix)
# Rows of Y with counts; X must cover every time step the model iterates over.
sample_idx <- which(rowSums(story_pollen_matrix) != 0)
Y <- story_pollen_matrix[sample_idx, ]
X <- scale(story_char_matrix)
n <- ncol(Y)
# mnGLMM provides starting values for mnTS.
glmm_mod <- mnGLMM(Y = Y, X = X[sample_idx, , drop = FALSE], V.fixed = diag(n))
V.fixed <- matrix(NA, n, n); V.fixed[1] <- 1
B.fixed <- matrix(NA, ncol(X), n); B.fixed[, 1] <- 0
C.start <- 0.5 * diag(n); C.fixed <- C.start
C.fixed[C.fixed != 0] <- NA
ts_mod <- mnTS(Y = Y, X = X, Tsample = sample_idx,
B0.start = glmm_mod$B[1, , drop = FALSE],
B.start = glmm_mod$B[2, , drop = FALSE], B.fixed = B.fixed,
C.start = C.start, C.fixed = C.fixed,
V.start = glmm_mod$V, V.fixed = V.fixed,
sigma.start = glmm_mod$sigma)
summary(ts_mod)
# Bootstrapping gives better standard errors than the Wald approximation.
# boot_out <- boot(ts_mod, reps = 1000)Installation
Windows
The best way to install at the moment is to compile from source:
- If you have Rtools44 or (Rtools45 for R 4.5) and devtools installed, you can build the latest version of the package directly from github using:
devtools::install_github("https://github.com/QuinnAsena/multinomialTS")
Alternatively, if you are having issues with the build tools, install the binaries:
install.packages("https://github.com/QuinnAsena/multinomialTS/releases/download/v1.1.0/multinomialTS_1.1.0-windows.zip", repos = NULL, type = "win.binary")
macOS
The best way to install at the moment is to compile from source:
If you have xcode-select and devtools installed, you can build the latest version of the package directly from github using:
devtools::install_github("https://github.com/QuinnAsena/multinomialTS")On apple it is nice and easy to download xcode-select by opening a terminal and copying this code:
xcode-select --install. Then try and run:devtools::install_github("https://github.com/QuinnAsena/multinomialTS")
Alternatively, if you are having issues with the build tools, install the binary.
For Apple Silicon (M-series) macs, use:
install.packages("https://github.com/QuinnAsena/multinomialTS/releases/download/v1.1.0/multinomialTS_1.1.0-macOS-arm64.tgz", repos = NULL)
Intel macs: there is no pre-built binary. Every macOS machine GitHub Actions offers is now Apple Silicon, and R packages with compiled code cannot be cross-compiled, so please install from source using the devtools route above. Binaries labelled macOS-x86_64 in releases before v1.1.0 were in fact built on Apple Silicon and will not load on an Intel mac.
Linux
Compile from source. You need a C++ compiler (g++ or clang++) plus BLAS and LAPACK development headers, which most distributions package:
- Debian/Ubuntu:
sudo apt-get install build-essential libblas-dev liblapack-dev - Fedora/RHEL:
sudo dnf install gcc-c++ blas-devel lapack-devel
Then:
devtools::install_github("https://github.com/QuinnAsena/multinomialTS")