Skip to contents

Detect response-code inversion

Source: R/check_response_inversion.R
check_response_inversion.Rd

Some participants (or whole batches of exported data) may have their response codes flipped relative to the convention rcicr expects: +1 means "inverted version chosen" when the analyst intended "original chosen", or the CSV export inadvertently negated the response column. A straightforward way to detect this is to compute the infoVal twice per participant — once with the original codes and once with every response negated — and compare. For correctly coded data, the original should score higher; if the flipped CI scores meaningfully higher, the response column is probably inverted.

Usage

check_response_inversion(
  responses,
  method = c("2ifc", "briefrc"),
  rdata = NULL,
  stimuli = NULL,
  baseimage = "base",
  col_participant = "participant_id",
  col_stimulus = "stimulus",
  col_response = "response",
  margin = 1.96,
  iter = 10000L,
  ...
)

Arguments

responses

Data frame with one row per trial. Required columns: participant_id, stimulus, response (values in {-1, +1}). Load yours from CSV via read_responses() or utils::read.csv(); column names are configurable via the col_* arguments.

method

"2ifc" or "briefrc".

rdata

Path to the rcicr .RData file. Either rdata or stimuli must be supplied for the 2IFC path.

stimuli

In-memory stimuli list (the $stimuli element of an rcisignal_sim object). Use in place of rdata when the file path no longer resolves (e.g. after saveRDS()/ readRDS() across R sessions).

baseimage

Name of the base image in rdata$base_face_files.

col_participant, col_stimulus, col_response

Column names.

margin

Numeric. Flagging threshold: flipped infoVal must exceed original infoVal by at least this amount. Default 1.96.

iter

Reference-distribution iterations. Default 10000.

...

Passed through to compute_infoval_summary().

Value

An rcisignal_diag_result() object. data$per_participant has participant_id, infoval_original, infoval_flipped, and likely_inverted (logical).

Details

This check calls compute_infoval_summary() twice and reports participants whose flipped infoVal exceeds the original by margin or more. A non-zero count is a strong signal that the response column is miscoded for those participants.

Requires rcicr (same as compute_infoval_summary()). Runs two infoVal sweeps, so it takes roughly twice as long.

Examples

if (FALSE) { # \dontrun{
sim <- simulate_2ifc_data(n_per_condition = 10, n_trials = 60, seed = 1)
check_response_inversion(sim$data, method = "2ifc",
                         rdata = sim$rdata_path)
} # }