Cross-sectional research can only show associations rather than causality. To solve
this prospective studies are used which separate the independent and dependent
variables by time. Sutton (2002a) argues that both these designs are problematic and
do not allow inferences about causality to be made. He suggests that randomized
experimental designs are the best solution to this problem.
Hankins et al. (2000a) provide some detailed guidelines on how data using the TRA
and TPB should be analysed and state that much research uses inappropriate analy-
sis. They state that if multiple regression analysis is used adjusted R^2 should be the
measure of explained variance, that residuals should be assessed and that semi par-
tial correlations should be used to assess the unique contribution of each variable.
They also state that ‘Structural Equation Modelling’ might be a better approach as
this makes explicit the assumptions of the models.
Much psychological research does not involve a sample size calculation or a consider-
ation of the power of the study. Hankins et al. (2000) argue that research using
social cognition models should do this if the results are to be meaningful.
The TRA involves a generalized measure of attitude which is reflected in the inter-
action between ‘expectancy beliefs’ about the likelihood of the given behaviour lead-
ing to particular consequences and evaluations about the desirability of these con-
sequences. For example an attitude to smoking is made up of the belief ‘smoking will
lead to lung cancer’ and the belief ‘lung cancer is unpleasant’. This is calculated by
multiplying one belief with the other to create a ‘multiplicative composite’. This is
called the ‘expectancy value’ belief. In subsequent analysis this new variable is sim-
ply correlated with other variables. French and Hankins (2003) argue that this is
problematic as the correlation between a multiplicative composite and other vari-
ables requires a ratio scale with a true zero. As with other psychological constructs,
the ‘expectancy value belief has no true zero only an arbitrary was chosen by the
researcher’. Therefore they argue that the expectancy value belief should not be
used.
Predictive problems
Models such as the TRA, TPB, HBM and PMT are designed to predict behavioural
intentions and actual behaviour. However, two main observations have been
made. First, it has been suggested these models are not that successful at predicting
behavioural intentions and that they should be expanded to incorporate new cog-
nitions. Second, it has been argued that they are even less successful in predicting
actual behaviour. This second criticism has resulted in research exploring the intention–
behaviour gap.
Predicting intentions: the need to incorporate new cognitions
Sutton (1998a) argued that studies using models of health beliefs only manage to
predict between 40 and 50 per cent of the variance in behavioural intentions. Therefore,
HEALTH BELIEFS 37