stochCurtail: Simulates stochastic curtailment on item scores of a test

In marjoleinF/curtail: A package for test curtailment

Description

Simulates stochastic curtailment on an existing test dataset of item scores, based on an existing training dataset of item scores, and a user-specified cut-off value.

Usage

  stochCurtail(dataset.train, dataset.test = NULL, Xstar, gamma0=.95, 
  gamma1=.95, plot = TRUE)

Arguments

dataset.train	A dataframe, containing item scores only, which will be used to derive the probabilities of obtaining a final test score greater than, or equal to, the cut-off value, based on the current cumulative score
dataset.test	A dataframe containing item scores only. Curtailment will be simulated on these observations. When no test dataset is specified, curtailment will be simulated on the training dataset.
Xstar	Cut-off value to be used for classifying observations as 'at risk' (test-score values greater than or equal to the cut-off value) or 'not at risk' (test-score values less than cut-off value)
gamma0	The threshold for the probability, calculated using the 'not-at-risk' training observations, that the classification decision based on the stochastically curtailed version will match that of the full-length instrument.
gamma1	The threshold for the probability, calculated using the 'at-risk' training observations, that the classification decision based on the stochastically curtailed version will match that of the full-length instrument.
plot	Should a histogram of test lengths be plotted?

Details

The code is still under development and might change in future versions.

Value

The function prints accuracy estimates to the command line, and plots the curtailed test length distribution. In addition, the function invisibly returns a list with the following elements:

test.results	data.frame with columns full.lenght.decision (classification decsion according to full-length test); curtailed.decision (classification decision according to curtailed test administration); current.item (item at which testing was halted); current.score (cumulative testscore at item at which testing was halted).
curtailed.test.length.distribution	Descriptive statistics of number of items administered and number of tests curtailed.
confusion.martrix	Confusion matrix of full-length and curtailed test classification decisions.
accuracy	Correct classification rate (accuracy), sensitivity and specificity.

References

Fokkema, M., Smits, N., Finkelman, M. D., Kelderman, H., & Cuijpers, P. (2014). Curtailment: A method to reduce the length of self-report questionnaires while maintaining diagnostic accuracy. Psychiatry Research 215, 477-482.

Fokkema, M., Smits, N., Kelderman, H., Carlier, I.V. & Van Hemert, A.M. (2014). Combining decision trees and stochastic curtailment for assessment length reduction of test batteries used for classification. Applied Psychological Measurement, 38(1), 3-17.

Finkelman, M.D., Smits, N., Kim, W. & Riley, B. (2012). Curtailment and stochastic curtailment to shorten the CES-D. Applied Psychological Measurement, 36(8), 632-658.

Examples

  ## obtain a test and training dataset
  set.seed(32061983)
  samp <- sample(1:1000, 500); train <- samp[1:500]
  trainingdata <- itemscores[train,]
  testdata <- itemscores[-train,]
  tmp1 <- stochCurtail(trainingdata, testdata, 19)
  tmp1$curtailed.test.length.distribution
  ## try lower gamma values for earlier stopping, but lower accuracy:
  tmp2 <- stochCurtail(trainingdata, testdata, 19, gamma0=.75, gamma1=.75)
  tmp2$curtailed.test.length.distribution

marjoleinF/curtail documentation built on May 21, 2019, 11:47 a.m.