trend_cor: Estimate the correlation between a DFA trend and some other...

Description Usage Arguments Details Value Examples

View source: R/trend_cor.R

Description

Fully incorporates the uncertainty from the posterior of the DFA trend

Usage

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
trend_cor(
  rotated_modelfit,
  y,
  trend = 1,
  time_window = seq_len(length(y)),
  trend_samples = 100,
  stan_iter = 300,
  stan_chains = 1,
  ...
)

Arguments

rotated_modelfit

Output from rotate_trends().

y

A numeric vector to correlate with the DFA trend. Must be the same length as the DFA trend.

trend

A number corresponding to which trend to use, defaults to 1.

time_window

Indices indicating a time window slice to use in the correlation. Defaults to using the entire time window. Can be used to walk through the timeseries and test the cross correlations.

trend_samples

The number of samples from the trend posterior to use. A model will be run for each trend sample so this value shouldn't be too large. Defaults to 100.

stan_iter

The number of samples from the posterior with each Stan model run, defaults to 300.

stan_chains

The number of chains for each Stan model run, defaults to 1.

...

Other arguments to pass to sampling

Details

Uses a sigma ~ half_t(3, 0, 2) prior on the residual standard deviation and a uniform(-1, 1) prior on the correlation coefficient. Fitted as a linear regression of y ~ x, where y represents the y argument to trend_cor() and x represents the DFA trend, and both y and x have been scaled by subtracting their means and dividing by their standard deviations. Samples are drawn from the posterior of the trend and repeatedly fed through the Stan regression to come up with a combined posterior of the correlation.

Value

A numeric vector of samples from the correlation coefficient posterior.

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
set.seed(1)
s <- sim_dfa(num_trends = 1, num_years = 15)
m <- fit_dfa(y = s$y_sim, num_trends = 1, iter = 50, chains = 1)
r <- rotate_trends(m)
n_years <- ncol(r$trends[, 1, ])
fake_dat <- rnorm(n_years, 0, 1)
correlation <- trend_cor(r, fake_dat, trend_samples = 25)
hist(correlation)
correlation <- trend_cor(r,
  y = fake_dat, time_window = 5:15,
  trend_samples = 25
)
hist(correlation)

bayesdfa documentation built on May 29, 2021, 1:06 a.m.