Data Harmonization in Aging Research: Not so Fast

Abstract

Background/Study Context: Harmonizing measures in order to conduct pooled data analyses has become a scientific priority in aging research. Retrospective harmonization where different studies lack common measures of comparable constructs presents a major challenge. This study compared different approaches to harmonization with a crosswalk sample who completed multiple versions of the measures to be harmonized.Methods: Through online recruitment, 1061 participants aged 30 to 98 answered two different depression scales, and 1065 participants answered multiple measures of subjective health. Rational and configural methods of harmonization were applied, using the crosswalk sample to determine their success; and empirical item response theory (IRT) methods were applied in order empirically to compare items from different measures as answered by the same person.Results: For depression, IRT worked well to provide a conversion table between different measures. The rational method of extracting semantically matched items from each of the two scales proved an acceptable alternative to IRT. For subjective health, only configural harmonization was supported. The subjective health items used in most studies form a single robust factor.Conclusion: Caution is required in aging research when pooling data across studies using different measures of the same construct. Of special concern are response scales that vary widely in the number of response options, especially if the anchors are asymmetrical. A crosswalk sample that has completed items from each of the measures being harmonized allows the investigator to use empirical approaches to identify flawed assumptions in rational or configural approaches to harmonizing.