README.md

In nklange/fullDPROC: Fitting Dual-Process model to recognition and source memory data

fullDPROC: Rpackage

• author: Nicholas Lange
• contact: [e-mail]
• cran link: tba
• Source: fullDPROC_0.1.0.tar.gz

This package fits the univariate dual-process model (Yonelinas, 1999) to recognition or source memory data by fitting the model to the ROC curve (minimizing SSE) or to the frequencies of responses by minimizing log-likelihood.

All fitting can be done by using the DPSD wrapper function and specifying data and model/parameter restrictions in the function arguments. The predicted ROC curve can be produced using the predictedROC function.

DPDS.R function

This is the main wrapper function. The package requires individual item-level confidence rating responses from recognition judgments or source judgments (for experiments with a single source dimension, i.e. a single rating scale ranging from Source A to Source B).

Arguments

• responseScale: Vector containing possible levels of confidence rating responses ordered from highest to lowest (e.g. 6:1).
• confidenceRatings: Vector containing participant confidence rating responses to individual items.
• TargetLure: Vector coding containing information whether each individual item is a Target or Lure item (e.g. Old, New, Old, Old).
• targetLabel: A string/integer designating the label for the target items in TargetLure. If fitSource = TRUE, one source has to be designated target, the other lure.
• lureLabel: A string/integer designating the label for the lure items in TargetLure. If fitSource = TRUE, one source has to be designated target, the other lure.
• iterations (default = 200) A numeric value specifying the number of iterations.
• eqRecollection (default = FALSE) A boolean value specifying if recollection is set equal for the target and lure source (TRUE) or is estimates separately for both sources (FALSE).
• eqVar (default = TRUE) A boolean value specifying if the standard deviation of the target distribution is equal to that of the lure distribution (i.e. = 1) (TRUE) or estimated separately (FALSE).
• fitSource (default = TRUE) A boolean value specifying if the model is being fitted to source memory data (TRUE) or recognition memory data (FALSE)
• fitROC (default = FALSE) A boolean specifying if the model is fitted by minimizing SSE to ROC (hit/fa rates) or by MLE to rating frequencies

Parameter restrictions

Recollection is bounded to be between 0 and 1, Familiarity to be positive. Criteria are ordered.

• eqVar = FALSE: sd_target is estimated, bounded to be positive.
• fitSource = FALSE: recollection_lure = 0 (overrides eqRecollection argument)

Optimization

• fit with BFGS in optim
• Starting parameters:
  • for MLE: for 40% of iterations, starting parameters are estimated (mostly) from data and 60% are drawn randomly from uniform distributions
  • for SSE minimization: drawn randomly from uniform distributions with reasonable range

Output

• Parameter estimates (recollection_target, recollection_lure, familiarity, sd_target, criteria) for the iteration with the lowest SSE or likelihood)
  • if eqVar = TRUE -> sd_target = 1
  • if eqRecollection = TRUE -> recollection_lure = recollection_target
  • if fitSource = FALSE -> recollection_lure = 0
• minimized value:
  • SSE if fitROC = TRUE
  • negative log-likelihood if fitROC = FALSE

predictedROC function

This function provides the output for the predicted ROC curve from parameter values (for graphs and such).

Arguments

• recollection_target
• recollection_lure
• familiarity
• sd_target (default = 1)
• fitSource (default = TRUE)

Output

• dataframe with false alarms (fa) in first column for approximately continuous 0 - 1, and corresponding hit rates (hit) in second column

Auxilliary functions

These are functions that are called from the wrapper function depending on arguments, but do not need to be called manually.

• fitDPSD.R: Function that fits recognition data (fitSource = FALSE) using MLE on rating frequencies
• fitDPSDsource.R: Function that fits source data (fitSource = TRUE) using MLE on rating frequencies
• fitDPSDROC.R: Function that fits recognition data (fitSource = FALSE) by minimizing SSE on cumulative rates
• fitDPSDROCsource.R: Function that fits source data (fitSource = TRUE) by minimizing SSE on cumulative rates
• cumRates.R: Function that transforms the raw rating data into cumulative hit/fa rates
• ratingFreq.R: Function that transforms the raw rating data into frequencies of ratings per rating bin and target/lure
• makeMLEstartparameters.R: Function that estimates starting parameters for MLE optimization. For 2/5 of iterations, starting parameters are chosen from normal distributions centred on parameter estimates taken from the observed data; for remaining iterations, starting parameters are drawn from uniform distributions.

To Do:

• Integrate Bayesian estimation from Hierarchical Bayes version
• Tidy for cran
  • vignette
  • example data or simulation function for some data
• Also accept already summarized cumrates or frequency data rather than only item-level data
• Make predictedROC talk directly to output from DPSD

nklange/fullDPROC documentation built on May 26, 2019, 2:34 a.m.