Skip to contents

Asymmetry Index Across a Family of Masks

Source: R/regional-symmetry.R
pr_symmetry_sensitivity.Rd

Recomputes pr_calc_symmetry_map() for every mask scheme in a list and returns one row per scheme, so that a symmetry claim can be reported together with how much it depends on where the region of interest was drawn.

Usage

pr_symmetry_sensitivity(
  trial,
  masks,
  statistic = "mean",
  denominator = "grid",
  balance = TRUE
)

Arguments

trial

A pr_trial object.

masks

A named list of mask schemes. Each element is a pr_mask, a logical matrix, or a list of exactly two of those giving the two sides explicitly — the same forms pr_calc_symmetry_map() accepts for its own masks.

statistic

Character. "mean" (default), "max" or "loaded"; see pr_calc_regional_map().

denominator

Character. "grid" (default) or "loaded"; see Details.

balance

Logical. Trim each scheme to its largest mirror-symmetric subset before computing the index. Default TRUE.

Value

A tibble::tibble with one row per scheme and columns scheme, balanced, statistic, denominator, n_left, n_right, left_value, right_value, asymmetry_pct.

Details

A left/right index is only as stable as its mask. Excluding dead sensors, restricting to a band, or thresholding on contact frequency all change which cells are compared, and each of those choices can move the index by more than the effect being reported. Running the family and printing the spread is cheap — the per-sensor map is computed once and reused for every scheme — and it turns an unstated choice into a reported one.

With balance = TRUE (the default) each scheme is passed through the mirror trim of pr_mask_mirror_balance() first, so every row compares equal, mirror-image sets of sensors and n_left == n_right throughout. Set it to FALSE to see what the raw masks give; the difference between the two runs is the part of the index that came from the mask rather than from the recording.

See also

pr_calc_symmetry_map() for a single scheme.

Other regional symmetry functions: pr_calc_cop_masked(), pr_calc_regional_map(), pr_calc_symmetry_map(), pr_mask_mirror_balance(), pr_mask_rowbands()

Examples

trial <- pr_example_trial("saddle_horse")
layout <- trial$layout
bands <- pr_mask_rowbands(layout, sides = FALSE)

schemes <- list(
  whole_mat = layout$active,
  cranial = bands$cranial,
  middle = bands$middle,
  caudal = bands$caudal
)
pr_symmetry_sensitivity(trial, schemes)[, c("scheme", "asymmetry_pct")]
#> # A tibble: 4 × 2
#>   scheme    asymmetry_pct
#>   <chr>             <dbl>
#> 1 whole_mat          13.1
#> 2 cranial            13.4
#> 3 middle             12.7
#> 4 caudal             13.1

# A data-driven mask need not be even-handed. Here the hotspot cells are
# 18 left against 12 right, and the raw index is a quarter of what the
# balanced 12/12 comparison reports.
hot <- matrix(FALSE, layout$grid_rows, layout$grid_cols)
co <- layout$coords_mm
hot[cbind(co$row, co$col)] <- pr_sensor_map(trial, "max")$value > 8
rbind(
  pr_symmetry_sensitivity(trial, list(hotspots = hot), balance = FALSE),
  pr_symmetry_sensitivity(trial, list(hotspots = hot))
)[, c("balanced", "n_left", "n_right", "asymmetry_pct")]
#> # A tibble: 2 × 4
#>   balanced n_left n_right asymmetry_pct
#>   <lgl>     <int>   <int>         <dbl>
#> 1 FALSE        18      12          2.92
#> 2 TRUE         12      12         14.6