Survey index residuals — plot_index

Plots survey index residuals by year, faceted by survey fleet (residual_type = "pearson", the default), or one-step-ahead (OSA) residual diagnostics for a single fit (residual_type = "osa", via osa_residuals() / plot.rceattle_osa()).

Usage

plot_index_residual(
  Rceattle,
  file = NULL,
  model_names = NULL,
  species = NULL,
  width = 7,
  height = 6.5,
  residual_type = c("pearson", "osa"),
  line_col = NULL,
  right_adj = 0,
  top_adj = 0.05,
  incl_proj = FALSE,
  single.plots = FALSE
)

Arguments

Rceattle: A single fit_mod() object or a list of them (overlaid).
file: Optional file stem; if given the figure is written to <file>_survey_indices.png.
model_names: Legend labels for the models.
species: Species (indices) to include. Default all.
width, height: Saved figure size (inches).
residual_type: "pearson" (standardized index residuals) or "osa".
line_col, right_adj, top_adj, single.plots: Deprecated base-graphics arguments, retained for back-compatibility and ignored.
incl_proj: Include projection years.

Value

A ggplot object.

Details

The residual is observed minus predicted, standardized by the sd the row was fitted with – the same quantity residuals.Rceattle() returns for type = "pearson". On a log-scale fleet it is measured against the likelihood's centre, log(hat) - bias_adjust_obs * sd^2 / 2; on a natural-scale fleet ("MVN", "MVNORM", "Normal", "TruncatedNormal" in Index_distribution), whose sd is absolute, it is (obs - hat) / sd. A covariance fleet ("MVN", "MVNORM") takes that sd from sqrt(diag(index_cov)), which its likelihood actually scores, so the residual is marginal rather than whitened. A positive residual means the survey saw more than the model predicted. The y scale stays free per fleet, since residual spread differs by fleet.