Page 120 - PhD GT

irrelevant due to the fact of their total abstinence from heroin during the past three months. Item analysis is based upon 115 respondents. The mean inter-item correlation for this section was lower at .31 and item-total correlations were all above .4 with the exception of the fifth item (‘During the past three months I started using drugs when I knew it would cause me problems’). This item had an item-total correlation of .3. Cronbach's alpha for this part of the scale was .82.
The third part of the scale, referred to as ICSPC and consisting of ten items, measures perceived control: respondents' beliefs about their current ability to control their use of heroin. Six respondents did not complete this part of the scale. The mean inter-item correlation was .38, item-total correlations were all above .43 and Cronbach's alpha was .86 (see Appendix 6). These scores compare closely with those found by Heather et al. 1993 in the original validation of the scale.
Further item analysis was conducted combining all 25 items in the scale; one hundred and eleven respondents completed all three parts of the scale. The first five items (constituting Part 1 of the scale) were found to have strong positive correlations with each other and negative correlations with all remaining items in the scale. The remaining 20 items had positive correlations with each other with two exceptions. Correlations were calculated between each item and the item total when that item had been deleted. Item total correlations for the first five items were similar to each other, all negative and ranging from -.25 to -.36 , while item total correlations for the remaining 20 items were also similar to each other, all positive and ranging from .36 to .65. Thus item analysis of the combined parts seems to support the view that Part 1 (AC) measured a distinct phenomenon which is negatively related to Parts 2 (FC) and 3 (PC). Cronbach's coefficient alpha for the whole scale was .81. Deletion of each of the first five items slightly raised Cronbach's alpha in each case, by approximately .02 whereas deletion of all other items resulted in a marginal (in the region of .01) lowering of Cronbach's alpha (see Appendix 7). This lends support to the finding that responses to the first five items were consistent with each other and distinct from responses to the remaining items in the scale.
5.7.6 Construct validity of the ICSdrg
“To the extent that a variable is abstract rather than concrete, we speak of it as being a construct” (Nunnally 1978 p. 96). Where a scale is devised to measure a construct, its validity for this purpose can be assessed by comparison with other scales that have been validated as being capable of measuring the same construct or a construct that can be argued to be theoretically positively correlated with it (Carmines and Zeller 1979). At the time of adapting and validating the
108