knitr::opts_chunk$set(
  collapse = TRUE,
  comment = ">",
  fig.height = 1
)
library(mmre)

Overiew

R package to fit two-stare continuous-time discrete-space Markov models with individual level random effects. Methodology was developed with Enrico Pirotta to assess the effects of exposure to Navy sonar on marine mammal movement patterns.

Installation

To install the mmre package run

devtools::install_github("cmjt/mmre")
library(mmre)

All model likelihoods are coded using TMB; in oder to compile all TMB templates after installation run

compile.mmre()

and to load the templates run

dll.mmre()

once in each workspace.

Example

The mmre package contains an example dataset, example$data, of three individuals on the AUTEC Naval range (see figure below). The state variable indicate if an individual was off (state = 1) or on (state = 2) range (black polygon in figure). Id is the individual ID (individual_i for i = 1,2,3). The time column gives the relative times os the observations in days and the t.since column gives, in days, the time since an individual was esposed to Navy sonar activity.

knitr::include_graphics("figure/AUTEC.png")
data(example)

To fit a simple two state continuous-time Markov model run

mod.basic <- fit.mmre(data = example$data,parameters = list(log_baseline = log(c(0.5,0.5))))

and to get the estimated transition probability matrix P(t = 1)

get.probs(mod.basic,1)

To compare the results to the msm package run

library(msm)
msm.fit <- msm(state ~ time, subject = ID, data = example$data, qmatrix = rbind(c(0, 0.5), c(0.5, 0)),  
    exacttimes = FALSE)
pmatrix.msm(msm.fit)

Model with individual level random effects

mod.basic.re <- fit.mmre(data = example$data,parameters = example$parameters.basic.re)
get.probs(mod.basic.re,1)

Model with exponential decay effect of covariate and individual level random effects

mod.decay.re <- fit.mmre(data = example$data,parameters = example$parameters.decay.re, decay = TRUE, cov.names = "t.since")
get.coefs(mod.decay.re)


cmjt/mmre documentation built on Oct. 2, 2023, 11:24 p.m.