Careless responding in internet-based quality of life assessments.

Abstract

Quality of life (QoL) measurement relies upon participants providing meaningful responses, but not all respondents may pay sufficient attention when completing self-reported QoL measures. This study examined the impact of careless responding on the reliability and validity of Internet-based QoL assessments. Internet panelists (n = 2000) completed Patient-Reported Outcomes Measurement Information System (PROMIS®) short-forms (depression, fatigue, pain impact, applied cognitive abilities) and single-item QoL measures (global health, pain intensity) as part of a larger survey that included multiple checks of whether participants paid attention to the items. Latent class analysis was used to identify groups of non-careless and careless responders from the attentiveness checks. Analyses compared psychometric properties of the QoL measures (reliability of PROMIS short-forms, correlations among QoL scores, "known-groups" validity) between non-careless and careless responder groups. Whether person-fit statistics derived from PROMIS measures accurately discriminated careless and non-careless responders was also examined. About 7.4% of participants were classified as careless responders. No substantial differences in the reliability of PROMIS measures between non-careless and careless responder groups were observed. However, careless responding meaningfully and significantly affected the correlations among QoL domains, as well as the magnitude of differences in QoL between medical and disability groups (presence or absence of disability, depression diagnosis, chronic pain diagnosis). Person-fit statistics significantly and moderately distinguished between non-careless and careless responders. The results support the importance of identifying and screening out careless responders to ensure high-quality self-report data in Internet-based QoL research.