6.4 Multivariate analysis Table 6.24 summarises the variables found in the bi-variate analyses to be significantly associated in some way with the vulnerability measures. In order to identify the key variables that may explain vulnerability to the effects of flooding, among the many that were associated with the vulnerability measures, backwards regression analyses were carried out for two of the vulnerability measures: the overall severity measure and the subjective stress measure included in the Intangibles Survey for this report. Such analyses had been undertaken for the two GHQ12 measures, both current and worst (Tunstall et al., 2006) and the results are reproduced here.
Table 6.17 was used to guide the selection of variables for inclusion in the multiple regressions. In all cases the dependent variables were transformed to normality using a log normal transformation to ensure that their distribution more closely conformed to the normal distribution required for regression analyses. The .010 probability level was used as the cut-off for the inclusion of variables in the models although most of the variables included were significant at the .05 level.
6.4.1 Multi-variate analyses: flood characteristics and vulnerability
Although the focus of Task 11 is upon the social dimensions of vulnerability, it is recognised that the characteristics of a flood event will have an effect on the impacts of flooding (Section 1.2.1). The bi-variate analyses in Table 6.24 show this to be the case across all the vulnerability variables for three flood characteristics, maximum depth of main room flooding, number of main rooms flooded and contamination of the floodwaters. Therefore, it is important to take account of how the flood characteristics affects vulnerability. In initial multi-variate analyses, we examined the impact that the flood event characteristics taken on their own had on the vulnerability variable as indicated in the model shown in Figure 6.1
NS = Not statistically significant on any measure. * indicates some statistical significance
When the worst time GHQ12 was considered as a measure of short term health impacts of flooding, four of the six factors emerged as explanatory factors (Table 6.18). Contamination of the floodwaters made the most significant contribution followed by the maximum depth of main room flooding. The extent of flooding as measured by the number of main rooms flooded and the length of flood warning lead time also emerging as significant factors. As predicted in the model, the greater the depth of main room flooding, the higher were the worst time GHQ12 scores. Where pollution of the flood waters was detected, the scores were also higher. A longer warning lead time was associated with lower scores and thus did lead to reduced mental health and stress effects at the time when those impacts were at their worst. Flood characteristics such as duration of flooding and the speed of onset were not significant factors. However the key finding is that these flood characteristics alone explained very little of the variance in the effect of flooding on the mental health and well being of flood victims at the time they judged the effects to be at their worst.
In considering models for the current GHQ12 Likert scores (Table 6.18), it must be emphasised again that these scores reflect the mental health and well being of respondents at the time of the interview and this occurred in most cases some years after the flood event. During this time many circumstances and events may have intervened to change the respondents’ state of health. Therefore, it is not surprising that the current state of health and well being as measured by the GHQ12 Likert scores is more difficult to predict with flood characteristics variables. Taking the current GHQ12 scores as an indication of the long term effects and vulnerability to flooding, it appears that, when flood characteristics were taken on their own, the same two characteristics of the flood event experienced that were key factors in the immediate worst time effects, depth and contamination, had a significant and lasting impact (Table 6.18). However in the long term, the length of a flood warning received was no longer a factor. It is notable that the characteristics of the event experienced taken on their own explained even less of the long term variability in the vulnerability measure, the current GHQ12 scores.
It can be argued that the current GHQ12 score is likely to be influenced by the worst time experience as indicated by the worst time GHQ12 score. These two variable GHQ12 Likert variables were quite strongly correlated (Correlation: 0.54). Therefore, a further regression analysis was undertaken including the worst time GHQ12 Likert as an explanatory variable for current GHQ12 Likert scores along with the flood characteristics variables. When this was done, the worst time GHQ12 Likert emerged as the only significant factor indicating that the effects of the flood event are reflected in the GHQ12 worst time scores and do not add any explanatory power when that score is included. Indeed the worst time GHQ12 score explains a significant proportion of the variance in the current GHQ12 scores (R2 = 0.319, R2 (adjusted) = 0.318).
The factors that emerged when the stress of the flood event and the overall subjective severity were considered in relation to the nature of the event experienced were similar to those that were predictors of the GHQ12 measures (Table 6.19). Again flood event characteristics alone explained little of the variance in the subjective responses to flooding. Contamination of the flood waters emerged as a very significant predictor of both stress and overall severity. Then either the extent or depth of main room flooding and in the case of overall severity, both, were factors. The length of flood warning lead time, however, featured as an explanatory variable only for the stress of the flood event albeit not a very significant predictor ( not significant at the p<0.05 level) suggesting that a longer warning lead time may make a small contribution to reducing the stress experienced during and after a flood event.
Table 6.18: Flood event characteristics and GHQ12 Likert scores
Worst time GHQ12 Likert
Number of cases = 611, R2 = 0.075, R2 (adjusted) = 0.069