We used longitudinal measurement invariance tests to identify the items in an abbreviated version of the Dysfunctional Beliefs and Attitudes about Sleep Scale that are non-invariant (response shifted) against CBT-I based on data from 114 insomnia patients. The partial invariance model built accordingly was then used as a population model for simulations to examine the impacts of the response-shift items on follow-up paired t-tests.
Invariance tests indicate CBT-I would lift the intercept of one item in DBAS-10 and cause non-uniform calibrations in three items. The following up simulations showed that failing to exclude the intercept-lifted item from the calculations of the subscale scores would lower the probability of using paired t-test to correctly detect the treatment effect by up to 53%.
We recommend sleep researchers to consider the issues of response-shift when assessing sleep-related constructs in interventional studies for insomnia.
Copyright © 2020 Elsevier B.V. All rights reserved.