Combining Cluster and Factor Analysis to Examine the Effects of Social Vulnerability on Recovery

Tracking post-disaster repopulation is often limited to changes in structural foundations (e.g., houses rebuilt) or census survey data to estimate the rate of recovery. Recently, deliverable address counts at the census block group level from the US Postal Service, have served as a proxy for repopulation, to assess recovery from Hurricane Katrina in Orleans Parish. Although 83% of the Parish was flooded, 71% of the addresses were deliverable in the summer of 2008, increasing to 76% the following year.

Scholars have examined the effects of social vulnerability using principal components analysis based on census data. However, these analytic methods do not have measurement errors associated in the model; therefore, it is difficult to interpret factors in the measure. This study refines these measures by using the extended similarities tree, a nonparametric clustering algorithm that shows overlapping and hierarchical features among variables. This procedure selected key factors from census data that described a community's social vulnerability. Reliability and appropriate fit of the measures were tested using confirmatory factor analysis. Applying the newly derived social vulnerability indices, with deliverable addresses as the outcome, a random-effects growth model and geographically weighted regression were utilized to investigate non-spatial and spatial trends in New Orleans repopulation. Preliminary results revealed areas dominated by middle class, working poor, farming, and transient households had significantly lower rates of repopulation over time. This combination of cluster and factor analytic methods to measure social vulnerability and its effect on repopulation demonstrates a convenient and meaningful method to investigate community recovery.