exhaustive_tests: (main function) Runs exhaustive tests

In exhaustiveRasch: Item Selection and Exhaustive Search for Rasch Models

(main function) Runs exhaustive tests

Description

(main function) Runs exhaustive tests

Usage

exhaustive_tests(
  dset,
  modelType = "PCM",
  combos = NULL,
  scale_length = 4:length(dset),
  na.rm = TRUE,
  tests = c("no_test"),
  splitcr_mloef = NULL,
  splitcr_LR = NULL,
  splitcr_wald = NULL,
  icat_wald = FALSE,
  alpha = 0.1,
  bonf = FALSE,
  DIFvars = NULL,
  gap_prop = 0,
  extremes = TRUE,
  max_contrast = NULL,
  PSI = 0.8,
  ICs = FALSE,
  keep = FALSE,
  ignoreCores = 1,
  verbose = TRUE,
  ...,
  itemfit_param = NULL,
  estimation_param = NULL
)

Arguments

dset	a data.frame containing the data
modelType	a character value defining the rasch model to fit. Possible values: RM, PCM, RSM
combos	either 1) a list of item combinations to be tested, e.g. from apply_combo_rules or the passed_combos slot of an object of passed_exRa-class from a previous call to this function. Or 2) an object of passed_exRa-class. In this case, the previously fit models froms its passed_Models slot will also be used and will be passed to the test functions. This will speed up the analysis. If the parameter is NULL, all possible combinations of the items (columns) in dset will be tested
scale_length	a numeric vector defining the length of the item combinations to test
na.rm	a boolean value. If TRUE, in the respective item combination all cases with any NA are removed (na.omit). If FALSE, only cases with full NA responses are removed. NOTE: test_mloef currently does not allow for missing values. If test_mloef is under the tests to perform, na.rm will automatically be set TRUE for ALL tests.
tests	a vector of characters defining the tests to perform. Possible values: all_rawscores, test_itemfit, test_LR, test_mloef, test_waldtest, threshold_order, test_DIFtree, test_personsItems, test_respca, test_PSI. Tests will be performed in the given order. test_mloef is not available for pairwise estimation, threshold_order is not available (and not meaningful) for dichotomous models.
splitcr_mloef	Split criterion to define the item groups for test_mloef. "median" and "mean" split items in two groups based on their items' raw scores median or mean. "random" (only for 'psychotools' estimation) performs a random split (in this case, the seed can be set with the "splitseed" argument of itemfit_control. splitcr_mloef can also be a vector of length k (where k denotes the number of items) that takes two or more distinct values to define groups used for the Martin-Löf Test.
splitcr_LR	Split criterion for subject raw score splitting for test_LR. "median" uses the median as split criterion, "mean" performs a mean split, "random" (only for 'psychotools' or 'pairwise' estimation) performs a random split (in this case, the seed can be set with the "splitseed" argument of itemfit_control. splitcr_LR can also be a vector which assigns each person to a certain subgroup (typically an external criterion). This vector can be numeric, character or a factor.
splitcr_wald	Split criterion for subject raw score splitting for test_waldtest. "median" uses the median as split criterion, "mean" performs a mean-split, "random" (only for 'psychotools' or 'pairwise' estimation) performs a random split (in this case, the seed can be set with the "splitseed" argument of itemfit_control. Optionally splitcr_wald can also be a dichotomous vector which assigns each person to a certain subgroup (e.g., following an external criterion). This vector can be numeric, character or a factor.
icat_wald	a boolean value indicating if the waldtest will be conducted on item level (TRUE, default value) or on item category level. This parameter only effects estimations using 'psychotools' or 'pairwise' and will be ignored for eRm estimations.
alpha	a numeric value for the alpha level. Will be ignored for test_itemfit if use.pval in itemfit_control is FALSE
bonf	a boolean value whether to use a Bonferroni correction. Will be ignored if use.pval is FALSE
DIFvars	a data.frame containing the variables and their data to use for differential item functioning analysis with test_DIFtree
gap_prop	a numeric value between 0 and 1 that sets the criterion for the minimum proportion of neighboring person parameters with an item/threshold location in between. If set to 0, this criterion will not be checked (used in test_personsItems only)
extremes	a boolean value indicating if a check for the item/threshold locations left of the 2nd lowest and right of the 2nd highest person parameter (used in test_personsItems only).
max_contrast	a numeric value defining the maximum loading of a factor in the principal components analysis of the standardised residuals. Only relevant, if test_respca is one of the tests.
PSI	a numeric value defining the minimum value for the person- separation-index (separation reliablility).
ICs	a boolean value defining whether to compute information criteria for the remaining models. You can add these later to the object of class passed_exRa-class by using the add_ICs function.
keep	a boolean value difining whether des person parameters will be part of the passed_exRa-class results object (TRUE) or not (FALSE). Keeping the person parameters will result in shorter runtimes, if several tests that need these parameters are used. On the other hand there will be a largeramount of memeory usage.
ignoreCores	a numeric value for the number of cpu cores to hold out in parallelizing the test run.
verbose	a boolean value. If set to FALSE, all output during the analysis will be suppressed.
...	arguments for itemfit_control and estimation_controlcan be passed directly to this function.
itemfit_param	a list from itemfit_control with options for test_itemfit
estimation_param	options for parameter estimation using estimation_control

Value

an object of passed_exRa-class.

Examples

  library(exhaustiveRasch)
  data(ADL)
  passed <- exhaustive_tests(dset= ADL[c(1:4,7,12,14)],
    modelType= "RM", scale_length= 5, alpha=0.05,
    tests=c("test_LR", "test_itemfit", "test_respca"),
    splitcr_LR = ADL[,17], itemfit_param =
      itemfit_control(use.pval=FALSE, upperMSQ=1.5, lowerMSQ=0.5),
    estimation_param= estimation_control(
      est="psychotools"), verbose=FALSE)

exhaustiveRasch documentation built on April 3, 2025, 6:18 p.m.