Our study was conducted in 2010 among older people who had recently been admitted to a hospital in the context of the 'Prevention and Reactivation Care Programme', which was designed to prevent loss of function in older patients due to hospitalisation and targeted older hospital patients (≥ 65 years of age) who were vulnerable to loss of function after hospital admission. Three months after hospital admission is known to be a good moment to assess effects of a programme [20, 21]. Therefore, patients were interviewed three months after hospital admission. Our research is based on the pilot study of 456 patients (≥ 65 years old) prior to implementation of the 'Prevention and Reactivation Care Programme'. The results of the pilot study have been used to identify possible practical implementation problems in preparation for the main evaluation study and serve as a base for power calculations for the main study. We interviewed 296 patients in their homes (response rate 65%). This work was supported by Netherlands organisation for health research and development (ZonMw) grant number: 60-61900-98-130.
The 30-item SMAS consists of six five-item subscales. The scale's overall internal consistency is 0.90 . Within the subscales of taking initiative, investing, self-efficacy, variety, and multifunctionality, abilities are related to the physical and social dimensions of well-being in the SPF theory [13, 14]. The ability to have a positive frame of mind is considered a more general cognitive frame; its subscale is thus not directly related to specific dimensions of well-being. Average overall SMAS scores range from 5 to 30, with higher scores indicating higher SMA.
Overall subjective well-being was measured with the SPF-IL(s) (15-item Social Production Function Instrument for the Level of well-being) . The scale integrates both affective and cognitive components of well-being, and measures levels of physical and social well-being. Cronbach's alpha of the SPF-IL in our study was 0.72, indicating a reliable instrument.
Cantrill's Ladder was used to assess satisfaction with life and reflects a general, cognitive evaluation of a person's overall well-being .
The analyses included the following seven steps.
1. The sample characteristics were analysed using descriptive statistics.
2. We data-screened the items by examining the number of missing items and each item's mean and standard deviation.
3. To verify the factor structure of the questionnaire and to test whether the relationship between observed variables and their underlying latent constructs existed, confirmatory factor analysis was executed using the LISREL program version 8.80 . By using structural equation modelling the overlap between items and dimensions can be traced via modification indices that were used to further refine the measurement model and eliminate potential overlap between items. No correlation errors either within or across sets of items were allowed in the model.
4. Item reduction analysis was performed to develop a short version of the questionnaire. Item removal following several criteria: (i) items were excluded following modification indices provided by LISREL and the strength of the factor loadings; (ii) item elimination stopped when the reliability of each subscale dropped below 0.65; (iii) subscales were left with as few items as possible (but a minimum of three) without loss of content and psychometric quality; and (iv) at least one physical well-being item (comfort or stimulation) and one social well-being item (affection, behavioural confirmation or status) was kept in each subscale while maintaining validity and reliability. Listwise deletion of cases with missing data on the 30 items resulted in N = 204. Imputation was done by replacing missing values with the mean of the data, restoring the original sample of N = 296.
We used four indices of model fit to test the measurement models, with cut-off criteria proposed by Hu and Bentler . First, the overall test of goodness-of-fit assesses the discrepancy between the implied model and the sample covariance matrix by means of a normal-theory weighted least squares test. A plausible model has low, preferably non-significant χ2 values. Chi-square is, however, overly sensitive when the sample size is large (over 200) , leading to difficulty in obtaining a desired non-significant level . Second, the Root Means Square Error of Approximation (RMSEA) reflects the estimation error divided by the degrees of freedom as a penalty function. Values on RMSEA below 0.06 indicate small differences between the estimated and observed model. Values of up to 0.08 suggest a reasonable fit of the model in the population. Third, we used the Standardized Root Means square Residual (SRMR), which is a scale invariant index for global fit that ranges between 0 and 1. Values on SRMR lower than 0.08 indicate a good fit. Fourth, we calculated the Incremental Fit Index (IFI), which compares the independent model (i.e., observed variables are unrelated) to the estimated model. Values on IFI are preferably larger than 0.95.
5. After item reduction analyses the first full version and final short version of the instrument were tested on the non-imputed dataset (N = 204). Listwise deletion of missing data on the basis of the 18 items in the short version resulted in N = 221. We re-ran the final short version on this sample.
6. Internal consistency of the subscales was assessed by calculating Cronbach's alphas, inter-item correlations within each subscale, and correlations between subscales.
7. Validity is the degree to which a scale measures what it is intended to measure; here we focused on the construct validity of the questionnaire. Construct validity is supported if instruments purported to assess the same concept correlate substantially with one another. Since the SPF-IL and SMAS are both based on the SPF theory we evaluated construct validity by comparing the SMAS scale scores with well-being measured by the SPF-IL scale. In addition, we will compare the SMAS scale scores with well-being measured by Cantril's ladder.