| NEWS | R Documentation |
Wrote all of the probability and derivatives of probability functions in C for speed and code clarity.
Wrote a quick integration function (that takes function arguments) to calculate Sympson's rule and a quick integration function (that takes vector arguments) to calculate the trapezoidal rule. The former is now implemented in eapEst, and the latter is implemented to calculate weighted Fisher and KL information for itChoose.
Changed simIrt to calculate probabilities in C before simulating responses (rather than calculating probabilities and simulating responses at the same time). This change can result in a slightly different matrix of responses to items due to the alteration of simulation order.
Changed FI and KL functions so that item information would be presented in a person by item dimension matrix rather than an item by person dimension matrix.
Fixed a bug in catIrt, so that if in non-interactive/batch mode, the CAT would not automatically set n.max to nrow(params).
Fixed a bug in print.catIrt, so that it would display the correct precision-based stopping rule.
Fixed a bug in wleEst, so that the interval would naturally extend if the first derivative of the log-likelihood evaluated to the same sign.
Fixed manual links to be consistent with updates to the catR package, namely the deletion of the createItemBank function.
Fixed another bug in WFI (weighted Fisher information), so that it would not error if only 0-1 items had been taken in the CAT.
Added NEWS.Rd to be compatible with R's news function.
Made sure that a new line would appear if progress = TRUE.
Fixed a bug in WFI (weighted Fisher information), so that it would not error if only 0-1 items had been taken in the CAT.
Added "it.range" argument to the catStart/catMiddle list and to the itChoose function to restrict the difficulty parameters (in a binary response model) to be within a particular range.
Changed some code to be neater, cleaner, etc.
Fixed a bug in catIrt, so that letting catStart$n.select or catMiddle$n.select be anything other than 1 would actually select between the top "n.select" items.
Fixed a bug in simIrt so that the parameters did not need to have column names for the function to correctly work.
Fixed a bug in termGLR, so that the stopping rule would use the "range" sub-argument of catMiddle rather than the depricated "int" sub-argument to calculate the likelihoods.
Fixed a bug in termCI, so that the stopping rule would only consider the last value of SEM when forming a confidence interval.
Fixed a small bug in print.summary.catIrt, so that the function would check the n.start argument of catStart (rather than the n.it argument) to determine how many starting items that the CAT administered.
Fixed several bugs in catMiddle, so that the CAT would choose the starting scoring method if the MLE wasn't achieved (and catMiddle$select = "MLE"), and wouldn't error if the starting scoring method was "step".
Fixed a bug whereby the catIrt function was not adequately determining if the first vector of item parameters was the item number. Now, the item number can ONLY be attached if the parameters are outputted from a catIrt function.
Added more termination criteria, including when the Fisher Information exceeds a particular threshold or when either the Fisher Information or SEM change less than a particular amount. Note that the original v.term argument is now obsolete and replaced with p.term and i.term arguments.
Added WFI (Weighted Fisher Information) function (not exported) and added two item selection criteria (Posterior WFI and Likelihood WFI) which both use the WFI function.
Added IKL (Integrated KL Divergence) function (not exported) and added two item selection criteria (Fixed Integral KL and Variable Integral KL) which both use the IKL function.
Changed the arguments to itChoose and exported the function so that anybody can use itChoose in his/her own CAT algorithm.
Changed simIrt to return params with the item numbers cbind-ed onto the front of them, which is required for the itChoose function.
The "var" option of catTerm$term no longer exists and is replaced with "precision" and "info".
Changed catMiddle$int to catMiddle$range to reduce confusion.
Changed the argument CI to conf.lev in plot.catIrt so that catIrt and plot.catIrt have consistent names for confidence intervals.
Changed thetas in catIrt and simIrt to theta to keep consistency with other functions.
Changed catStart$init.thet to catStar$init.theta to reduce confusion.
Plotting the individual CAT progression is more customizable.
The bank info functions/SEM functions are no longer able to be customized to prevent plotting conflicts.
Exported [.mod and [<-.mod functions outside of the catIrt function, where mod is any of the IRT models.
Changed a lot of the manual entries for clarity and consistency, and added a manual entry for the itChoose function.
Fixed the SEM for WLE/BME/EAP estimation to be appropriate for the particular estimation methods, and added numDeriv as a required package for BME variance estimation.
The format of the catIrt input function was rewritten a bit.
Rather than putting everything as arguments into catSim, modules (startCat, middleCat, termCat) were created so that the package can be easily updated in the future.
A Sympson-Hetter addendum to the catIrt function was added, in the hope of eventually writting a Sympson-Hetter function.
The simple version of the GLR stopping rule was added for classification CAT rather than just the original SPRT method.
The confidence interval stopping rule was added for classification CAT.
The call was deleted to the output of catIrt because the process of updating the function is too tedious to make it worth while.
Added an option for theta estimation from the full bank of items. If EAP was selected and failed to estimate theta (due to underflow), full-test estimation proceeds using the BME method.
Most of the functions were completely rewritten for speed.
summary.catIrt and plot.catIrt were rewritten, and print commands were added.
summary.catIrt and print.catIrt now can display each simulees CAT progression.
S3 method FI was added to make the code clearer.
S3 classes "brm" and "grm" were added, and class "cat" was changed to class "catIrt."
The argument "int" was added to catIrt, and a tiny bit of code was rewritten so that cat.thet estimates could not exceed int on either side.
The call was added to the output of catIrt so that the update command would work.
The eapSE command was added, and now catIrt gives the posterior standard deviation if score = "EAP".