Community Resilience to Climate Change: Theory, Research and Practice

131 [7] The Wilcoxon-Mann Whitney statistic (for two groups) and the Kruskal-Wallis test (for more than two groups) were selected as the nonparametric test best suited to ordinal responses (following Marusteri and Bacarea 2010), although it does not permit incorporating the complex stratified survey design. Nonparametric tests do not make assumptions about the underlying distribution of a variable but are less powerful than parametric tests. [8] The sample size is not large enough to permit analysis by subregion apart from by urban-rural zone (Sana Jaffers, personal communication). [9] The belief that flooding posed a problem to the community and the household were highly correlated; we chose to include the former because the bivariate analysis revealed stronger relationships with the resilience-related capacities. We restricted the focus to flooding occurring in the last two years to ensure a relatively recent and consistent frame of reference. [10] Because almost all respondents were household heads, we focused our analysis on the gender of the respondent rather than female versus male headship. [11] The wealth index was generated by principal components analysis using the following household assets: radio, mobile phone, fridge, TV, sofa set, electric/gas cooker, motor vehicle, livestock, and water pump (Twaweza, personal communication). [12] The statistics presented here are for the population, i.e., they incorporate the complex sampling design, while Appendix 1 presents the unweighted data and test statistics for these data. In practice, the differences between the averages derived from weighted and unweighted data are very slight. [13] It is not feasible to compute a measure of goodness of fit for the ordinal logit that takes into account complex sampling design in STATA. To give an indication of the fit, we computed the Pseudo R2 for unweighted specifications of these regression, which yields values of about 2%.