This page of links was constructed to accompany Chapter 6 ("Quantitative Aspects of Neuropsychological Assessment") of Goldstein & McNeil (2012). Clinical Neuropsychology: A Practical Guide to Assessment and Management for Clinicians, 2nd Edn (pp 129-155).Chicester: Wiley.
Throughout the chapter reference is made to computer programs. The relevant chapter headings are listed below together with links to the programs (in some cases the same programs are referred to under different chapter headings). All programs are written in Delphi and are for use with PCs (they will run on a Mac if suitable emulation software is installed).
All of these programs were written to accompany scientific papers. These papers give the technical background to the methods and provide worked examples. It is strongly recommended that these papers are read before using the programs
Click here for computer program(s) that calculate percentile ranks from raw scores and provide accompanying confidence/credible intervals (these confidence intervals capture the effects of using a normative sample to estimate the percentile rank of a score in the normative population)
For specific examples of the use of these latter methods click here for a program that provides percentile norms with accompanying interval estimates for commonly used self-report mood scales (e.g., HADS etc) based on large UK normative samples, or here for a program that provides percentile norms for other self-report scales (e.g., BDI) based on large Australian normative samples
Click here for computer program(s) that illustrate the method of expressing the endpoints of confidence intervals on test scores as percentile ranks (these confidence intervals capture the effects of measurement error on test scores). The programs are for use with the WAIS-III and WISC-IV. A version for the WAIS-IV is also available (click here).
Click here for computer program(s) that test for reliable differences between short-form Index scores on the WAIS-III, and here for a program that does the same for the WISC-IV (these programs also report the abnormality of the differences - see chapter for this crucial distinction). A version for the WAIS-IV is also available (click here)
Click here for computer program(s) that allow users to estimate the percentage of the normative population expected to exhibit at least as many abnormally low scores as a case (the user has a choice of criteria for defining an abnormally low score).
One of the programs is generic, it will provide base rate data for any test battery, provided the user has access to the correlation matrix for the tests. The others are specifically for use with the WAIS-III and WISC-IV. A version for the WAIS-IV is also available (click here)
Just as a neuropsychologists would like to know what percentage of the normative population would exhibit at least as many abnormally low scores as a case (see above), it is also useful to know what percentage would exhibit at least as many abnormally large differences between tests as a case. The programs referred to above also provide such base rate data (i.e., click here for programs). A version for the WAIS-IV is also available (click here)
A wide range of computer programs fall under this heading. These methods are frequently upgraded to incorporate new or improved features - the links below are limited to the most recent versions of each program.
Click here for a program (Singlims_ES.exe) that compares a case's score to a control or normative sample. The program provides a significance test, point and interval estimate of the effect size for the difference between the case and controls, and point and interval estimates of the abnormality of the case's score. A Bayesian approach to this problem gives the same results as the classical test (the Bayesian program, BTD_ES.exe, can be found at the same location)
The best classical test for such a comparison is the Revised Standardized Difference Test (RSDT). Click here for a program that performs this test (the current version of the program for this test is RSDT_ES.exe, which now provides point and interval estimates of effect sizes).
A Bayesian test for the standardized difference between a case's scores on two tasks is also available (BSDT; Bayesian Standardized Difference Test) . This test has advantages over the classical test and is recommended over it. It provides a significance test, point and interval estimate of the effect size for the difference, and point and interval estimates of the abnormality of the case's difference. Click here for this program, the link takes you to a list of programs, the program required is DiffBayes_ES.exe.
As noted in the chapter, the conventional criteria for a classical dissocation requires only that a case is signifcantly poorer than controls on one task but does not differ from controls on another. These criteria are not very rigorous (and are associated with high rates of false positives). It is better to also require that the case shows a significant difference between their performance on the two tasks. This additional criterion avoids the high rates of false postivies associated with the conventional criteria. The test on the difference can be conducted using either the RSDT or BSDT as outlned in the previous section. Two programs can be used to test whether a case meets the criteria for a classical or strong dissocation, using either classical methods (Dissocs_ES.exe), or Bayesian methods (DissocsBayes_ES.exe). Click here for these programs.
Sometimes the performance of case and each of the controls are expressed as correlation coefficients or the slopes of regression lines (i.e., in time or weight estimation tasks etc). When the data for a case are expressed as a correlation click here for a program that compares the case's correlation to those of controls. Click here for a program that compares slope of a case's regression line with those of controls.
The basic methods of testing for a deficit in the single case, or testing for a difference between a case's performance on two tasks, have now been extended to allow users to control for the effects of covariates (e.g., one can test for a deficit controlling for the effects of years of education, or processing speed).
Click here for these programs: the program BTD_Cov.exe tests for a deficit controlling for covariates; the program BSDT_Cov.exe tests for a standardized difference between two tasks controlling for covariates. These two programs take summary data from the controls as inputs. Companion programs (BTD_Cov_Raw.exe and BSDT_Cov_Raw.exe) perform the same analyses but take raw data from the controls as inputs.
Methods have recently been devloped that allow users to compare the scores of two single cases. The difference between the cases is referred to a control sample (i.e. a control sample is required to use these methods). The comparison of the two cases can be carried out in the absence of covariates, or controlling for the effect of covariates. Click here for these programs.
As noted in the book chapter, regression can be used to assess change (a case's score at Time 1 can be used to predict their score at Time 2; this predicted score is then compared to the score the case actually obtained at Time 2 - if the discrepancy is large this indicates change in functioning.
Two sets of programs are available: the first set assumes you have access to an existing regression equation for predicting scores at Time 2 from scores at Time 1. There are two such programs - one for use when there is a single predictor variable (i.e., Time 1 scores), the other for use when there is a vector of predictor variables (e.g., Time 1 scores, and age)
Click here for the programs.
As noted in the book chapter, regression equations can be built from summary data from a sample (i.e., means and SDs of predictor and criterion variables, and correlation between them). Therefore, a neuropsychologist does not need an existing equation to use regression for inference in the single case. The program regbuild.exe takes summary data from a sample, builds a regression equation, and then applies it to the data for a single case. Click here for the program. There is also a companion program that can allow users to build regression equations even when the correlation between variables has not been reported (the correlation can be "recovered" if the study presents the results of a paired t-test or ANOVA).
Since the chapter was written the work on regression from summary data has been extended to allow clinicians to build and apply multiple regression equations using summary data (Crawford et al., 2012). Click here for the multiple regression program; the link also includes updated versions of the bivariate (i.e., single predictor) regression programs (the new versions inlude point and interval estimates of effect sizes for the discrepancy between predicted and obtained scores).
Click here for the computer program described in the book chapter. The program applies Bayes' theorem: it provides point and interval estimates of the post-test probability that a case has a condition of interest (or for a negative test, the post-test probability that they are free of the COI). The pre-test and post-test probability distributions are plotted.
