Summarise a correlation recovery into per-pair metrics — summary.bmmtools_cor

One row per condition, pair, estimator and scale, summarised across replications. Errors, coverage and the correlations are reported against both truths: the generating correlation (bias, rmse, coverage) and the correlation the simulated subjects had (bias_sample, rmse_sample, coverage_sample, r, ccc).

Usage

# S3 method for class 'bmmtools_cor_recovery'
summary(object, ...)

# S3 method for class 'bmmtools_cor_recovery_summary'
summary(object, ...)

Arguments

object: A bmmtools_cor_recovery object from recover_correlations().
...: Not used.

Value

A bmmtools_cor_recovery_summary tibble with the columns term, estimator, scale, n_replications, n_converged, true_value, sample_sd, mean_estimate, bias, rmse, bias_sample, rmse_sample, coverage, coverage_50, coverage_sample, ci_width, ci_width_50, rejection_rate, false_positive_rate, power, r, r_low, r_high, ccc, ccc_low, ccc_high, z_mean, z_sd, z_mean_mcse and z_sd_mcse, preceded by condition when the object came from a grid. The z columns summarise the rows' z, as in summary.bmmtools_recovery(); a correlation has no contraction in this version. coverage_50 and ci_width_50 are the coverage of the generating correlation and the mean width of the central 50 % interval; they are NA when the estimates carry no such interval.

Details

true_value is the generating correlation when it is the same in every replication and NA otherwise. sample_sd is the standard deviation of the in-sample correlations across replications.

rejection_rate is the share of intervals that exclude zero. It is reported as false_positive_rate when every true_value is 0 and as power when none is. A true_value that is NA, which is what a nonzero correlation becomes on the natural scale, counts as not 0, so power is defined there. With a mix of zero and nonzero values both are NA.

r (with a Fisher-z interval) and ccc (Lin's concordance, see recovery_ccc()) compare the estimates with sample_value across replications, one pair per replication. They are NA with fewer than three replications or when either side has no spread.