Detect potential change points in a dynamic_beta posterior path

For each dynamic coefficient in a dynamic_beta fit, compares the posterior distribution of the maximum scaled first-difference \(M = \max_t |\beta_t - \beta_{t-1}| / \sigma_\beta\) to a Monte-Carlo approximation of the same statistic under the AR(1) prior. The returned bf column is the tail-ratio score \(\Pr(M^{post} > m^*) / 0.05\), where \(m^*\) is the 95\ quantile of \(M^{prior}\). It is kept under the historical column name for compatibility, but it is not a marginal-likelihood Bayes factor. Use it to surface posterior temporal jumps that the AR(1) prior cannot comfortably accommodate.

Usage

detect_change_point(
  fit,
  coefs = NULL,
  n_prior_sims = 2000L,
  threshold_bf = 10,
  seed = NULL
)

Arguments

fit

A fitted lame object with dynamic_beta.

coefs

Optional character vector of coefficient names to check. Defaults to every dynamic coefficient on the fit.

n_prior_sims

Number of prior simulations for the null distribution. Default 2000.

threshold_bf

Cutoff for the "warn" column. Default 10; lower to 5 if the false-positive rate on AR(1) data is too high.

seed

Optional seed for reproducibility.

Value

A data frame with one row per dynamic coefficient: coef, bf (the heuristic tail-ratio score), m_post_mean (posterior mean of \(M\)), m_prior_q95 (95\ \(M\)), t_hat (period of the largest scaled jump), warn (logical, bf > threshold_bf).

Examples

# \donttest{
data(YX_bin_list)
fit <- lame(YX_bin_list$Y, YX_bin_list$X, family = "binary", R = 0,
            dynamic_beta = "dyad",
            nscan = 60, burn = 15, odens = 5, verbose = FALSE)
#> Warning: `family` = "binary" but `Y` contains values other than 0/1.
#> ℹ `Y` will be thresholded to `1 * (Y > 0)`; if you meant counts, use "poisson",
#>   or "ordinal"/"normal" as appropriate.
detect_change_point(fit)
#>      coef bf m_post_mean m_prior_q95 t_hat  warn
#> 1 X1_dyad  0   0.1319158    2.578110     2 FALSE
#> 2 X2_dyad  0   0.1068509    2.608174     2 FALSE
#> 3 X3_dyad  0   0.1456707    2.538016     2 FALSE
# }