Exploring response shift in longitudinal data

Abstract

OBJECTIVE: To evaluate recall bias in the thentest and to explore response shifts in chronically ill patients over time. METHODS: In a five-year study of 93 multiple sclerosis patients, kappa statistics examined recall bias in thentest scores. The thentest examined changes in internal standards. Longitudinal factor analysis investigated changes in conceptualization in evaluating quality of life (QOL). RESULTS: Standard analyses revealed deteriorating physical role performance but stability in other QOL indicators over time. Agreement between baseline indicators and thentest scores was lower for fatigue than for ambulation ( p<0.02), suggesting that the thentest captures recalibration response shift (17%) along with recall bias (22%). Patients changed internal standards for physical role limitations ( p<0.05) and fatigue ( p<0.0001), and reconceptualized QOL with respect to well-being and physical role. CONCLUSIONS: Underlying response shifts allowed patients to maintain homeostasis in reported QOL despite physical deterioration. These shifts interrelate but are not redundant. More variance in the thentest was explained by recall bias than by recalibration response shift.

Keywords:

• Quality of life
• Response shift
• Recall bias
• Longitudinal
• Psychosocial
• Multiple sclerosis
• Acronyms: EDSS = Expanded Disability Status Scale
• MAF = Multidimensional Assessment of Fatigue
• MS = multiple sclerosis
• MSSE = Multiple Sclerosis Self-efficacy
• SIP = Sickness Impact Profile
• QOL = quality of life
• ns = not significant

Acknowledgments

This work was funded in part by a grant to Dr. Schwartz from the Agency for Health Care Policy and Research (RO1 HSO8582-03).

Notes

¹The present work used existing methods, and thus did not assess the reprioritization component of response shift.

²Changes in internal standards may be ineluctably related to reconceptualization. If the response scale is expanded or contracted based on experience, then the anchors and the response intervals will have different meaning. See Schwartz and Sprangers (1999) for detailed discussion.

³The structure of the SIP precluded item selection in this way since total scores are based on weighted combinations of individual items. Thentest items from the SIP were thus selected based only upon their face validity.

⁴Donner et al. (1996) provide two tests of the ‘intraclass kappa’, which yielded almost identical p-values, kappa estimates, and confidence intervals. We report results of the intraclass kappa test labeled , because it provides better coverage based on simulations, and is asymptotically equivalent to Cohen's kappa (Blackman and Koval, 2000).