200 likes | 208 Views
This study explores the challenges in assessing disability in aged care, particularly when using neutral or "not applicable" response options. It discusses the implications and potential biases associated with different scoring methods and proposes alternative approaches such as Item Response Theory. 8 Relevant
E N D
Complications with Assessing Disability in Aged Care: When Does “Does Not Apply” Apply? E. Helmes & A. Campbell Department of Psychology James Cook University
Disabilities are common in residential care facilities • Severe levels of disability may affect other domains of function • Example: impaired mobility limits social interactions
Many rating scales and self-report instruments include a neutral, or “Cannot Say” option Examples: early MMPI, 16PF (5th edition), Likert scales with uneven number of options (5-, 7-, or 9-point scales)
Responses to such neutral points are ambiguous: • Neutral? • Indifferent? • Lacks understanding of content? • Lacks knowledge need to answer? • Hostility?
What of seemingly more direct “Does Not Apply” or “Not Applicable” options? • Bristol Activities of Daily Living Scale (Bucks et al., 1996) – all 19 items • MOSES (Helmes, et al., 1987) – 18 of 40 items
11. FINDING WAY AROUND INSIDE: (For example, ability to find his room, the washroom, the dining room) How often during the daytime in the past week did the resident become disoriented (confused) in finding his or her way around the inside of the residence? 1. Not at all 2. Seldom (only one to three times during the week) 3. At times (either once or twice a day on more than three days, or several times a day on one to three days) 4. Often (several times a day or on more than three days) 5. Question does not apply ‑‑ the resident never moved around inside the building without assistance from the staff
Content Interpretation? • Pruchno et al. (1988): • 5 of 18 items – inability to speak implies greater levels of disability, so score as 5>4 • 11 items equivalent to non-occurrence, so equate with “Not at All”, so score as 5=0 • 536 nursing home residents, 24/40 items retained after changed scoring & confirmatory factor analysis
Samples • Norming sample; 2921 unique cases • Psychogeriatric – 397 • Nursing home – 918 • Home for the Aged – 563 • Continuing Care – 447 • 924 (31.6%) males, 1985 (68%) females • Mean age 78.9 (SD = 10.9) • 490 Single, 688 Married, 1588 Widowed, 123 Divorced or Separated
Scoring Variations • All “Does Not Apply” coded as “5” • Pruchno et al. variation • Listwise deletion of any case with a “Does Not Apply” score (as in 1987 components analysis)
Analysis • Scoring key as target: 8 items on each of 5 dimensions • 12 covariance matrices (3 scoring variations x 4 samples) • EQS confirmatory factor analysis • M-Plus distribution-free confirmatory analysis
Results • All solutions not optimal: cross-loading items • Fewer model mis-specifications with M-Plus • No clear pattern: M-Plus suggests poorer fit with ‘Exclude’ scoring option (CFI; but not RMSEA)
Results • Pruchno approach more model mis-specifications • Pruchno approach more marginal loadings • Exclude approach fewest marginal loadings, mis-specifications with EQS (not so with M-Plus) • Deletion method results in fewer items with low loadings (i.e. clearer structure)
Conclusions • Minimal differences across methods of compensation for “Does Not Apply” option • No method gives univocally better fit • Listwise deletion gives clearer structure but at cost of smaller and likely biased sample
Alternative: Item Response Theory • IRT provides information on performance of response options • Preliminary results of analysis of nursing home data using GGUM (Roberts et al., 2004): Generalized Graded Unfolding Model
Disorientation Item 16: 5 more extreme than 4 8 of 18 items
Withdrawal Item 40: “Does not Apply” = Most Severe 10 of 18 items