alastairrushworth / badlm

📈 Bayesian adaptive distributed lag models

Overview

badlm is a small R package for inferring the temporally delayed dependence between a predictor and a response variable. See the paper

• Rushworth, A. Bayesian adaptive distributed lag models. (2018). Arxiv link.

Installation

To install the development version of the package, use

devtools::install_github("alastairrushworth/badlm")

# load the package
library(badlm)

badlm example

Generate a distributed lag function

# a nice distributed lag function - hump
x          <- 0:50
dlfunction <- -0.1 + (0.01*exp(-0.2*x) + exp(dnorm(x, sd = 4, mean = 10))) / 10

plot(x, dlfunction, type = "l")

Generate predictor and response under the distributed lag function

# response is an AR(1) process
expose        <- arima.sim(model = list(ar = 0.5), n = 500, sd = 0.1)
expose        <- (expose - mean(expose)) / sd(expose)
lag_mat       <- lag_matrix(expose, p = 50)
deaths.sig    <- lag_mat %*% dlfunction
deaths        <- deaths.sig + rnorm(450, sd = 0.01)

Try to recover the distribution lag function

dlm_est <- badlm(x = expose, y = deaths, 
           nlag = 50, k = 30, samples = 10000)

Plot the resulting lag curve

plot_lagcurve(dlm_est)