Page 232 - PhD GT

scores and dependence scores that were found in the validation study of the LDQ were replicated in the present study both at intake and at the follow-up points, lending support to the reliability of this method.
10.4 Analysis of the data
Analysis of the data was conducted in a number of steps. In Chapter 7, the participants who were seen for follow-up at three months and at twelve months were described and compared them with those not seen; their scores and results of short interviews at each of the follow-up points were also presented. Differences in mean scores were presented in order to investigate whether any change in these had occurred. There followed a closer examination of the nature of the change and in Chapter 9 the correlates and predictors of change in dependence were explored. Bi-variate correlational analysis was used to examine the relationships between dependence and other variables, followed by regression analyses to examine the effect of these different variables on the outcome of change in dependence when the relationship between the different variables was accounted for. Issues arising from this analysis plan are discussed below.
10.4.1 Problems with the use of difference scores for cohort analytical studies
Cronbach and Furby (1970) warned against the use of change scores on several grounds: change scores have different meanings for different individuals depending upon the starting point or pretest score; measurement errors or the unreliability of the instrument are compounded in change scores. However, Rogosa, Brandt and Zimowski (1982) argued that this is not necessarily the case, claiming that “the difference between two fallible measures can be nearly as reliable as the measures themselves” (p. 744). They used a growth model to show that the difference score can be an accurate and useful measure of individual change, even in situations where reliability is low (p. 730). Their analyses are restricted to the special case of linear growth whereas in the present study differences in dependence scores across the three data collection points are complex and not unidirectional, as described in Chapter 9.
In the present study, individuals with low dependence scores at the outset have less scope for change than individuals who start out with high dependence. Thus people with small change scores may be low dependence individuals who had achieved the maximum possible change for them as measured by the LDQ or high dependence individuals who did not change very much. If an absence of dependence or low dependence is the outcome which is sought, then the dependence
227