The composite scores in WRAP were created by averaging several scores from different neuropsychological tests, after first z-transforming each score to put it on a common scale. Such composite scores, which can be either global or domain-specific in nature, can reduce error variance, making it easier to detect subtle longitudinal change (Jonaitis et al., 2019). Fourteen composite scores are now available. Five of these are domain-specific composites structured after analyses done on WRAP data by Lindsay Clark (Clark et al., 2016), and nine are global composites inspired by Mike Donohue's Preclinical Alzheimer's Cognitive Composite (Donohue et al., 2014).

In Freeze 2023_May, we added four new composites in response to substantial missingness on some of the constituent raw tests. First, the Stroop test was discontinued in December 2019, and so Executive Function composite scores had not been available for recent years. To allow for greater continuity, we created two new Executive Function composites which substitute, variously, Animal Naming and CFL for the Stroop Color-Word score. Second, the advent of remote testing brought with it greater missingness on Digit Symbol and the Trail-Making Test, which had been used in all PACC scores. To allow for computation of a global composite based on the telephone battery (T-Cog), we created three-test PACC scores, again using, variously, Animal Naming and CFL to estimate executive function. A paper on the T-Cog evaluating the second of these sets of composites is in preparation. Four- and five-test composites rely on the MMSE, for which we do not yet have a crosswalk from the T-Cog version of the MoCA, and are not yet available.

Each composite is available with two scalings: one that has been z-transformed such that scores from cognitively unimpaired participants at their first observation have a mean of 0 and a standard deviation of 1 (these begin with z_); and one that has been transformed so that the same set of scores follows a distribution with a mean of 100 and a standard deviation of 15 (these begin with std_). Data in the paper linked above suggests that composites are more powerful than both individual raw test scores and our earlier factor scores that were based on similar inputs.

The following table lists the constituent tests in each composite score and, where available, provides a citation in which the composite is described more fully. I have also attached two text files containing the SQL code used to create these composites, as well as a table of coefficients necessary for their calculation.

Scores from Freeze 18 through Freeze May_2020 were computed without the final rescaling step. These scores, previously labeled with the suffix _old, are no longer available as of Freeze 2023_May. Prior to Freeze 18, composite scores were not computed in the WRAP database. Earlier freezes had included factor scores that had been developed on the basis of factor analyses of the first and second waves (Dowling et al., 2010; Koscik et al., 2014). Because these factor scores demonstrated more unwanted longitudinal variability compared to the cognitive composites (Jonaitis et al., 2019), they are no longer included in freezes.