News and Events

A common problem in life-span developmental research is that the measure for a given construct is not repeated over time, due to changes in “test-forms” or the need for “age-appropriate” scales. Structural equation modeling (SEM) analyses can potentially be utilized in such situations, similar to their use when data is incomplete due to subject attrition. Using longitudinal data from the IHD archives (Berkeley Growth Study) and from our own longitudinal study (McArdle, Hamagami, Meredith & Bradway, 2000), we discussed several options for modeling such data. If two tests (e.g., WIAS, Stanford-Binet) can be assumed to measure the same construct, then data can be recalibrated when only one measure has been collected at a given point. One approach is to build in measurement overlap (e.g., administer several forms of a test at one point in time), and then SEM techniques could be utilized, if other assumptions are met. However, as these assumptions are often not met in such data sets and this type of overlap is rarely present, another option is to use a specialized procedure based on Item Response Theory (IRT). In examining individual items on the WAIS and Stanford-Binet, we found that 45 of the 248 words appeared on both tests. Then, mapping of items versus persons, we found support for a single construct assessing vocabulary skills, and thus recalibrated the measures into a new scoring system based on a common metric. Utilizing this new measure, we used growth modeling techniques and found that there was growth in vocabulary knowledge until about age 20 and then no subsequent decline (up to age 70+). Small, but important differences in this change were found across gender, samples, and education level. In sum, in the absence of exact repeated measures over time, by building in some overlap at the item or scale level, constructs can be recalibrated and a consistent measure of the construct can be analyzed.

John J. McArdle's presentation "Modeling latent growth curves using longitudinal data with non-repeated measurements" can be viewed in PDF format, using Adobe® Acrobat® Reader®.