Individual-level analysis of DPOG was conducted using multiple regression analysis. Outcome gain scores had a low mean and were positively skewed; therefore the Poisson log-functional form with a GLM command was used [65].

The model was developed in line with a conceptual framework which hypothesized that outcomes would be influenced by a mixture of individual characteristics (dependency, how DPs where managed) and patterns of service provision (types of care received, direct payments support) (Fig. 2). Given the relatively small sample size, the model was conceived for explanatory purposes [66].

Factors influencing outcomes from direct payments

Explanatory variables included individuals’ characteristics, needs (IADL, ADL), dependency and services used. Information on types of support purchased, total care input and proportion contributed to total care input by each support type were included. Total care input represents the weekly sum of hours of: DP support, self-funded care and unpaid care. Hours of care were generally recorded as per the care plan/DP records, but if these differed from the daily diary, the latter took precedence, although the ‘official’ care package amount was recorded separately.

Although data on cognitive impairment was collected (by observation), it was not included as a variable in the model, as it was outside the capacity of the research to include a formal assessment of cognitive impairment, and because of its potential impact on other variables.

A number of variables initially included in the model were later discarded as not statistically significant. These included age > 80; PA turnover, package size (as hours per week and as £ per week), purchased care from a home care agency, percentage of package spent on/ total care input (for all care categories), use of and significance of accountancy service, IADL score and individual scores for the following IADL items: telephone, household tasks shopping, transport. The final set of variables included was a result of a step-wise process, in which attention was paid to avoid collinearity. Risks of overfitting were reduced due to the fact that there were almost no missing data points [67]. Overdispersion was discounted performing the likelihood ratio test of the over-dispersion parameter alpha using a negative binominal distribution.

Note: The content above has been extracted from a research article, so it may not display correctly.

Please log in to submit your questions online.
Your question will be posted on the Bio-101 website. We will send your questions to the authors of this protocol and Bio-protocol community members who are experienced with this method. you will be informed using the email address associated with your Bio-protocol account.

We use cookies on this site to enhance your user experience. By using our website, you are agreeing to allow the storage of cookies on your computer.