|
http://education.novell.com/faq/testfaq.htm : FAQ regarding Novell certification exam (sketches some details of the adaptive algorithm) |
|
http://www.eu.microsoft.com/mcp/articles/tesinn.htm : information about the Microsoft certification examines; includes links to a sample exam and a white paper on adaptive testing (includes faq) |
|
http://www.gmat.org/catmain.htm : information about the Graduate Management Admissions Test (GMAT) (includes faq) |
|
http://www.prometric.com/testingcandidates/candidates.htm : Sylvian Prometric (formerly Drake Prometric) is a firm specializing in the administration of computer based tests and training. They provide the computers for a large number (maybe all?) national, commercial adaptive licensure exams; this page is, unfortunately, the only interesting page I could find on their site; it describes a little about their tests and has some interesting links. |
|
http://www.galisys.com/sylvan.html : white paper on computer-based testing (esp. adaptive testing); more thorough and scholarly than most marketing hype. |
|
http://www.cci.unl.edu/CVs/Dissertations/ReedCastleDiss.html : A dissertation with (IMHO) modestly interesting research questions but, more importantly, an excellant introduction to licensure, adaptive testing, etc. |
|
http://www.ncsbn.org/pfiles/nclex/cattop.html : information about the NCLEX(r) licensure exam (includes faq) |
|
http://www.musc.edu/cdap/roberts : information about unfolding IRT models, including research, software, etc. Unfolding models are those which assume an ideal point; for example, "I occationally help others. True..False" People who are very helpful might say false becaus ethey ALWAYS help others. Sociopaths might also say false because they NEVER help people. You and I (!?) would probably say true. |
|
http://www.spss.com/cool/papers/index.html : White papers by the good folk at SPSS including information about test development, data mining, a comparison of several decision-tree methods of analyzing data, etc. |
|
decision trees : Decision trees are a "black box" substitute for regression analysis. That is, they perform the same function but I don't want to tell you much about their internals (i.e., from the little *I* know). However, their significant advantage is that thay are "model-less". Whereas the standard linear *and* non-linear procedures require you to specify the model in advance (at least in some way), these procedures handle (1) the selection of the variables (if applicable), (2) the fitting of the model parameters, and (3) the creation of the model. If this sounds like magic, consider this class of models: Instead of constructing a linear (or non-linear curve) model of the data, these procedures construct a tree (perhaps binary) of splits based on rules which, ultimately, produce a classification decision. In some procedures, a (linear) regression is substituted for the classification (i.e., classification is essentially multi-dimensional subsetting followed by piece-wise regression). This white paper was written by the manufacturer of the CHAID method which has been criticized for not having a pruning function. Unfortunately, this sort of analysis has captured the "data mining" community's interest and little substantive information is available on-line. You might be able to find a copy of CART by Breimen, et al. |
|
Details of mastery testing: Although there is a fairly detailed literature on mastery models, there is virtually no discussion of this on the web. |
|
Details of the existing licensure exams: The implementation details of the exams are not widely advertised; e.g., Can you skip items? review? change previous answers? What are the psychometrics of the exams (reliability, validity, etc.)? Details of the item trials and calibration (sample size, motivation, representativeness, estimation methods and outcomes)? How is the inherent multi- dimensionality handled? What form do the items and responses take (multiple choice- single answer, MC-multiple answer, constructed response, simulation, ...)? |