Journal of the Formosan Medical Association (Apr 2014)

Tests of data quality, scaling assumptions, reliability, and construct validity of the SF-36 health survey in people who abuse heroin

  • En-Chi Chiu,
  • I-Ping Hsueh,
  • Cheng-Hsi Hsieh,
  • Ching-Lin Hsieh

DOI
https://doi.org/10.1016/j.jfma.2012.05.010
Journal volume & issue
Vol. 113, no. 4
pp. 234 – 241

Abstract

Read online

Health-related quality of life (HRQOL) is considered an important outcome indicator in substances abuse studies. However, psychometric properties of HRQOL measures are largely unknown in people who abuse heroin. Therefore, the present study aimed to examine data quality, scaling properties, reliability, and construct validity of the 36-Item Short Form healthy survey (SF-36) in people who abuse heroin. Methods: A total of 469 people who abuse heroin participated in the study. Data quality was determined by data completeness. Scaling properties were evaluated by item frequency distribution, equivalence of item means and standard deviations, item-internal consistency, and item-discriminant validity (calculating scaling success). Internal consistency was examined using Cronbach's α. Construct validity was examined by investigating convergent validity and divergent validity among the eight scales of the SF-36. Results: The results of data quality showed low missing rates (0.0–3.8%) and high completion rates in the scales (91.9–98.7%). The results of scaling assumptions showed good item frequency distribution on each item, roughly equivalent item means and standard deviations within a scale, good item-internal consistency (>0.4) and good scaling success rates (77.5–100%), except on the two scales of bodily pain (BP) and social functioning (SF). Three scales showed ceiling and/or floor effects [i.e., physical functioning (PF), role limitations due to physical problems (RP), and role limitations due to emotional problems (RE)]. Cronbach's α was acceptable (>0.7), except for the BP and SF scales. Construct validity was partially supported by the results of convergent validity and divergent validity. Conclusion: The results confirmed good data quality; satisfactory scaling assumptions and internal consistency (except for the BP and SF scales); and generally acceptable construct validity. However, the PF, RP, and RE scales showed ceiling and/or floor effects. Therefore, the BP, SF, PF, RP, and RE scales should be used with cautions in measuring HRQOL in people who abuse heroin.

Keywords