nhmgrid: An R package for visualization of non-homogeneous Markov model probabilities

The nhmgrid R package provides an easy-to-use interface for estimation and visualization of non-homogeneous Markov model transition probabilities. The main features in the package are:

• Supports a variety of model types in transition probability estimation using marginaleffects.
• Is able to calculate state transition proportions without a model.
• Produces highly customizable plots using ggplot2.

Installation

You can install the most recent version of nhmgrid by running the following lines:

# install.packages("devtools")
devtools::install_github("mirojantti/nhmgrid")

Example

The package contains a simulated panel data set health. The data set consists of 100 individuals whose health state has been measured every year for 10 years.

library(nhmgrid)

head(health, 12)
#>    id    sex age   state lagstate
#> 1   1 female  15 healthy     <NA>
#> 2   1 female  16    sick  healthy
#> 3   1 female  17    sick     sick
#> 4   1 female  18 healthy     sick
#> 5   1 female  19 healthy  healthy
#> 6   1 female  20 healthy  healthy
#> 7   1 female  21    sick  healthy
#> 8   1 female  22    sick     sick
#> 9   1 female  23 healthy     sick
#> 10  1 female  24    sick  healthy
#> 11  2 female  20 healthy     <NA>
#> 12  2 female  21 healthy  healthy
nrow(health)
#> [1] 1000

One can calculate the transition proportions between the health states using stprops and plot them. Because the plot is constructed using ggplot2, it is highly customizable.

props <- stprops(health, x = "age", state = "state", id = "id")
plot(props) +
  ggplot2::geom_smooth(se = FALSE)

The transition probabilities between the states can be estimated with a Markov model using stprobs. In this example we fit a multinomial logistic regression model and estimate the probabilities separately for men and women.

fit <- nnet::multinom(state ~ lagstate + age + sex, data = health)
probs <- stprobs(fit, x = "age", group = "sex")
plot(probs)

For more examples, see the package documentation.