2.1 Aims and overview of the methods
2.1.2 The research hypotheses
The following hypotheses concerning acceptability, comprehensibility, sensitivity, validity and reliability (see Appendix 1) were tested in three research settings: the community (study 1), a hospital clinic (study 2) and in general practice (study 3). It was hypothesised that the acceptability and sensitivity of the PIMS will be better than comparative instruments, and that the other psychometric properties will be no worse. Existing measures o f health status administered in conjunction with the PIMS are referred to as comparative measures (the SF-36, SF-12 and HSQ-12, see Chapter 1). Power calculations for sample size were carried out for the hypotheses, as indicated (see Section 2.4.7 on page 119). The methods used to test the psychometric properties of the PIMS are summarised in Box 17 at the end o f this chapter.
2.1.2. i Acceptability
1. The research hypothesis was that, as the items were derived from areas o f life originally generated by the public (Bowling, 1995b, 1995c), the PIMS will be more acceptable than comparative instruments: studies 1, 2 and 3 (see page 105).
2.1.2. a Comprehensibility
2. The research hypothesis was that respondents will be able to distinguish between the two sub-domains o f each PIMS item, that is between the impact o f any condition on an area o f life, and the importance of that area to the respondent: studies 1, 2 and 3
(comprehensibility, see page 106).
2.1.2. Hi Sensitivity
The hypotheses concerning sensitivity were that:
3. The floor and ceiling effects of PIMS i t e m s a n d scores will be lower than existing measures: studies 1, 2 and 3 (see page 108).
4. For those with a medical condition or a health problem (study 1), a cardiac condition, treatment by surgery or medicine (study 2), or a condition or longstanding illness (LSI) (study 3), the PIMS will be more sensitive and specific than comparative health status measures (classification, see page 108).
Ceiling effect: least im pact (PIM S) and best health (comparative scales), floor effect: greatest impact (PIM S) and worst health (com parative scales).
5. Mean PIMS scores will be higher for those described above, and corresponding differences will be less marked in the comparative scales: studies 1, 2 and 3
(discriminatory power, see page 109).
6. PIMS scores will discriminate between levels of severity. That is, mean PIMS scores for each level of severity (high, medium and low) will be significantly different: studies 2 and 3 (discriminatory power, see page 109).
7. In logistic regression models on the presence o f health problems, medical conditions or longstanding illness, models that include the PIMS will fit the data better than those that include comparative scales: studies 1 and 3 (discriminatory power, see page 109). 8. In multiple regression models on severity, a higher proportion of variation will be explained in models including the PIMS than in those including comparative scales: study 3 (discriminatory power, see page 109).
9. The responsiveness to change of the PIMS in study 2 will be better than the HSQ-12 in the outreach study (responsiveness to change, see page 109).
2.1.2.iv Validity
The hypotheses concerning validity were that:
10. The items included in the PIMS will be sufficient, and necessary, to cover the concept being measured, of the impact of any condition, or of health on life. That is, the proportion of individuals who endorsed particular items will be greater than the
proportion who added particular open items in the space provided at the end of the scale: studies 1, 2 and 3 (content validity, see page 111).
11. In ordinal regression models on ‘overall impact’ the contribution of the PIMS items (studies 1 and 3) will be as great as that of the comparative scales and other items (content validity, see page 111).
12. Item redundancy will not be detected by item-to-item correlations of the PIMS: studies 1, 2 an d 3 (content validity, see page 111).
13. Despite variation between individuals in meanings of impact on life of a condition, a common core of items relevant to most people will be found: studies 1 and 3 (content validity, see page 111).
14. The relevance of items will not differ for those whose condition is longstanding and those with a more recent condition: study 3*^ (content validity, see page 111).
The item on ‘overall im pact’, included in the PIMS questionnaire asks; ‘How much has your condition affected your quality o f life in the past four weeks?’
Chapter 2. Aims and validation methods
15. The concept o f impact of a condition incorporates, in this thesis, the distinct sub- domain of importance o f areas of life to individuals as well as the distinct sub-domain of limitation in the areas. More variation in impact will be explained in ordinal regression models that include both sub-domains: studies 1 and 3 (construct convergent validity, see page 113).
16. The impact of a condition will vary by socio-economic status, severity of condition and use of health services. Those in lower socio-economic groups, in poorer health, and higher users of health services will obtain significantly higher mean impact scores, indicating worse impact: studies 1, 2 and 3 (construct convergent validity, see page
113)'’.
17. PIMS scores will be significantly correlated with ‘overall impact’ and with two other proxy variables, general health status and condition severity. Correlations with the two proxy measures will be lower (but moderate) than with ‘overall impact’: studies 1, 2 and 3 (criterion concurrent validity, see page 115)**.
18. PIMS scores will be significantly correlated with recovery scores: study 2 (predictive validity, see page 115).
2.1.2. V Reliability
The hypotheses concerning reliability were that:
19. Items in the PIMS will exhibit internal consistency, that is, overall item correlations for each scale will be high: studies 1, 2 and 3 (internal consistency, see page 116). 20. Item-to-item and item-to-total correlations o f the PIMS will be high: studies 1, 2 and 3 (internal consistency, see page 116)*^.
21. Factor loadings obtained in principal component analysis will demonstrate high internal consistency: studies 1,2 and 3 (internal consistency, see page 116).
22. The reliability of the PIMS will be similar to the SF-36, based on reported values for the SF-36: studies 1 and 3 (test-retest reliability, see page 117).
23. If the PIMS was determined by factor analysis to be multi-dimensional, correlations between split, or multiple forms will be significant: studies 1 and 3 (internal
consistency, see page 116).
Pow er calculations to test the difference in proportions o f those with LSI and more recent conditions, who were limited in items, were carried out (difference in proportions test).
” Pow er calculations to test the correlation between PIM S scores and age were carried out. Pow er calculations to test the correlation between PIMS scores and impact were carried out. Pow er calculations to test the internal reliability o f the PIM S were carried out (C ronbach’s oc).