Cat: Computerized Adaptive Testing Survey (catSurv) Object

In erossiter/catSurv: Computerized Adaptive Testing for Survey Research

Computerized Adaptive Testing Survey (catSurv) Object

Description

Creates an object of class Cat. Cat objects are used in administering Computerized Adaptive Testing (CAT) Surveys. These objects contain several pieces of information relevant for CAT surveys, and are used as input in the main functions of the catSurv package.

Details

Assume we have a survey battery with I questions. An object of the class Cat has the following slots:

• ids A vector of length I of unique question identifiers. Default is the column names of response data frame used to calibrate Cat object.

• guessing A vector of length I of guessing parameters. Guessing parameters are only applicable for Cat objects fit with the "tpm" model, using the tpmCat function.

• discrimination A vector of length I of discrimination parameters.

• difficulty A vector or list of length I of difficulty parameters. For Cat objects of the "ltm" or "tpm" model, difficulty is a vector that contains a parameter for each item. For Cat objects of the "grm" or "gpcm" model, difficulty is a list that contains a vector for each item, and each vector contains parameters for each response option.

• answers A vector of length I of answers to questions as given by the survey respondent. Unanswered questions have the value NA. Questions respondent has skipped or refused to answer have a value of -1.

• priorName A character vector of length one giving the prior distribution to use for the ability parameter estimates. The options are "NORMAL" for the normal distribution, "STUDENT_T" for the student's t distribution, and "UNIFORM" for the uniform distribution. The default value is "NORMAL".

• priorParams A numeric vector of length two of parameters for the distribution specified in the priorName slot. When priorName is set to "NORMAL", the first element of priorParams is the mean, and the second element is the standard deviation. When priorName is set to "STUDENT_T", the first element of priorParams is the location parameter and the second is degrees of freedom. When priorName is set to "UNIFORM", the elements of priorParams are lower and upper bound, respectively. Note that the uniform distribution is only applicable for the "EAP" estimation method. The default values are 0,1.

• lowerBound A numeric indicating the lower bound of the interval of the latent scale used in estimation. The default value is -5.

• upperBound A numeric indicating the upper bound of the interval of the latent scale used in estimation. The default value is 5.

• model A string indicating the model fit to the data. The options are "ltm" for the latent trait model, "tpm" for Birnbaum's three parameter model, "grm" for the graded response model, and "gpcm" for the generalized partial credit model.

• estimation A string indicating the approach to estimating ability parameters. The options are "EAP" for the expected a posteriori approach, "MAP" for the modal a posteriori approach, "MLE" for the maximum likelihood approach, and "WLE" for the weighted maximum likelihood approach. The default value is "EAP".

• estimationDefault A string indicating the approach to estimating ability parameters when the primary estimation choice indicated in the estimation slot is "MLE" or "WLE" and this estimation fails to converge. The options are "EAP" and "MAP". The default value is "MAP".

• selection A string indicating the approach for selecting the next item. The options are "EPV" for minimum expected posterior variance, "MEI" for maximum expected information, "MFI" for maximum Fisher information, "MPWI" for maximum posterior weighted information, "MLWI" for maximum likelihood weighted information, "KL" for the maximum expected Kullback-Leibler (KL) information, "LKL" maximum likelihood weighted KL information, "PKL" maximum posterior weighted KL information, "MFII" for maximum Fisher interval information, and "RANDOM" where the next item is chosen randomly. The default value is "EPV".

• z A numeric used in calculating δ. δ is used in determining the bounds of integration for some selectItem methods. See selectItem for more information. Default value is 0.9.

• lengthThreshold A numeric. The number of questions answered must be greater than or equal to this threshold to stop administering items. The default value is NA.

• seThreshold A numeric. The standard error estimate of the latent trait must be less than this threshold to stop administering items. The default value is NA.

• infoThreshold A numeric. The Fisher's information for all remaining items must be less than this threshold to stop administering items. The default value is NA.

• gainThreshold A numeric. The absolute value of the difference between the standard error of the latent trait estimate and the square root of the expected posterior variance for each item must be less than this threshold to stop administering items. The default value is NA.

• lengthOverride A numeric. The number of questions answered must be less than this override to continue administering items. The default value is NA.

• gainOverride A numeric. The absolute value of the difference between the standard error of the latent trait estimate and the square root of the expected posterior variance for each item must be less than this override to continue administering items. The default value is NA.

Author(s)

Haley Acevedo, Ryden Butler, Josh W. Cutler, Matt Malis, Jacob M. Montgomery, Tom Wilkinson, Erin Rossiter, Min Hee Seo, Alex Weil

See Also

checkStopRules, estimateTheta, gpcmCat, grmCat, ltmCat, selectItem, tpmCat

erossiter/catSurv documentation built on Dec. 11, 2022, 6:36 p.m.