Premium
P1‐192: Identification of Spline Functions and Cohort Differences in Psychometric Measures of the Long Beach Longitudinal Study
Author(s) -
Kennison Robert F.,
Zelinski Elizabeth,
Petway Kevin
Publication year - 2011
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1016/j.jalz.2011.05.472
Subject(s) - multivariate adaptive regression splines , spline (mechanical) , cohort , demography , cohort effect , rasch model , longitudinal study , linear regression , recall , regression , statistics , psychology , mathematics , medicine , bayesian multivariate linear regression , cognitive psychology , structural engineering , sociology , engineering
Background:Although declines in psychometric measures with age are often reported, few studies have examined changes with models that allow for identification of multiple splines where the rate of average decline may shift in old age. Zelinski and Kennison (2007) examined age and cohort effects in five psychometric ability measures with linear growth curve models, where two-piece splines with a common knot fixed at age 74 were fit to all five measures. It is likely, however, that the different measures are each associated with unique change functions that may include more than two splines andmultiple knot points. Thus, in the present study, unique change functions were identified for each of the measures with Multivariate Adaptive Regression Splines (MARS).Methods: Five Long Beach Longitudinal Studymeasures, reasoning, space, list recall, text recall, and vocabulary, representing two cohorts born 16 years apart, were modeled over ages 60 to 85 (n € Eœ 950). Scores were calibrated to Rasch scales with a possible 0-100 range. Empirically determined age-related changes in each measure were examined with MARS analyses. Possible cohort interactions in identified change functions were also identified.Results: Age:As shown in Figure 1, declines were universally observed, yet change rates and spline functions differed betweenmeasures. Reasoning showed steady declines beginning at the earliest age modeled (age 60), but the decline rate slowed after age 74. Space showed linear declines after age 66.8. List showed declines after age 64.3. Text showed declines after age 66.4. Vocabulary was stable up to age 73.9, declined from 73.9 to 83.4, and then recovered. Cohort: The later born cohort performed better than the earlier born cohort at the intercept for all measures except vocabulary, which showed no cohort effect. For reasoning and list recall the advantage maintained by the later born cohort diminished in later age. Conclusions: The use of an exploratory approach to identifying multiple linear functions with MARS produced results similar to those of Zelinski & Kennison (2007). However, these new findings also showed that the empirically identified deflection points varied by measure. Vocabulary showed relative stability up to the mid 70s before it declined. Space, list recall, and text recall were associated with initial declines beginning in the mid to late 60s. Reasoning was the only measure associated with initial declines at age 60. While the later born cohort outperformed the earlier born cohort at younger ages, this advantagewas reduced or eliminated in very old age suggesting greater cognitive reserve enjoyed by the later born cohort may be reduced late in life.