are contrasted, with levels of anxiety and coping skills being the dependent variables. In practice there may be a number of procedures for measuring these variables, all of which are likely to be intercorrelated. Each of these variables could be examined separately, though in reality there are only two hypotheses under investigation – the impact of the treatment on anxiety and its effect on coping skills. More than two statistical analyses are therefore redundant, and represent an overstatement of the data available to the researchers. A real-life example of this process is the much-cited National Institute of Mental Health study of treatments for depression (Elkin, 1994) which shows statistical significance on only some of a relatively large family of variables pertaining to dysfunctional emotional states. A consequence of multiply-sampling related data-sets is to increase the risk of Type I errors – rejecting the null-hypothesis when that hypothesis is false (in practice, for example, claiming that one treatment works better than another when in reality both work equally well). Because it is well recognised that a series of measures tapping similar domains may be inter-related, investigators often employ multivariate tests, which permit some understanding of relationships between dependent measures. Though this procedure overcomes some of the problems noted above, problems can arise where multivariate tests which indicate overall significance are then followed by univariate tests. Not only does this increase the risk of Type I error, but results can be difficult to interpret, once again because of possible relationships among variables under test. =(#0)%0($72
DP
.01230/1.40/5&&'67894/0/571.8/5&&/6648./1.40&
Made with FlippingBook Ebook Creator