05). However, for parents in the MMR group, there was a significant association between intention and whether or not they had taken their child for the first MMR, χ2(2, n = 144) = 10.182,

exact p = 0.002, two-sided (three cells had expected count less than five). Sequential logistic regression Modulators analyses were performed to identify significant predictors of intention for MMR and dTaP/IPV separately. This method was see more used as it is deemed most suitable for when there are theoretical grounds on which to predict the relative importance of variables [20], [24] and [25]. Direct predictors of intentions (attitude; subjective norm;

perceived behavioural control) were entered in the first block. The belief composites (behavioural beliefs; normative beliefs; control beliefs) were entered in the second block, along with the sociodemographic variables that had correlated significantly with intention (first MMR in the case of MMR and number of children in the case of dTaP/IPV). Conner et al. [23] report that by entering the variables in this way the researcher can test whether the effects of the belief composites are mediated by other TPB components. They also argue that by including all components in the model (including those that did not correlate significantly with intention), this provides a more stringent selleck screening library test of the role of any additional variables [23]. Assumptions of logistic regression were validated by examining residuals [24]. For both MMR and dTaP/IPV, there were only a small number of outliers. For MMR, their removal did not alter the results significantly. For dTaP/IPV, the removal of four outliers made a significant difference

to the results and the regression was re-run. For both vaccinations, tolerance values were >0.1 and VIF values were <10, indicating that there was no collinearity between the predictor variables [24]. A total of 144 cases were analysed (three much were deleted due to missing data). To determine the required sample size, Tabachnick and Fidell [20] advocate using N ≥ 50 + 8m (m is the number of predictors) to test the overall fit of the model and N ≥ 104 + m to test the individual predictors within the model. The researchers were interested in the overall correlation and the individual independent variables. In this case, Tabachnick and Fidell [20] recommend calculating N both ways and choosing the larger number of cases. In accordance with their recommendations, a minimum sample size of 111 was necessary. Using a criterion of p ≤ 0.