A note for ACAPS
After sudden-onset disasters, needs assessments may produce severity estimates repeatedly, apace with updated information. We expect updates to add new information; at the same time, if impacts last, we also expect that earlier severity estimates resonate in later ones.
The extent to which severity persists is not well known. To our knowledge, nowhere have severity measures from subsequent needs assessments been correlated. There is a practical interest in gauging this correlation. If needs change rapidly in nature, degree and direction, severity estimates should be repeated frequently. If the severity pattern is more persistent, estimates have a longer shelf-life. Subsequent assessment may then focus more on indicators with direct operational value.
Two measurements in Nepal, following the earthquakes in 2015, provide an opportunity to study the degree and structure of persistence. UNOCHA produced severity measures for low-level administrative units – Village Development Committees (VDCs) - in April and June. Over 600 VDCs were evaluated at both points in time. Although the indices used for the purpose employed different indicators, they can be correlated.
At face value, the correlation between the two estimates is moderate (+.49). It suggests that the impact on the affected communities changed considerably within two months. This is to be expected because a second earthquake struck Nepal during this period, with an epicenter in a different district from the first. However, some of the variability is a statistical artifact. Both indices incorporated indicators with fixed weights. When these are released, the connection between the indices shows up much stronger (+.76).
To do so, we rely on an established statistical technique – canonical correlation analysis – that tests how closely two sets of indicators measure the same concept. The same technique can be employed to express the degree to which a composite measure – severity in our case – has changed between two points in time. The indicators collected at the two points need not be the same – and, in fact, in Nepal they were largely different. For example, the strength of the earthquake was at first measured geophysically, as the Modified Mercalli Index; in the second assessment a combination of casualty, building damage, and food security indicators took its place.
The much higher correlation coefficient that we obtain by this method leads to a paradoxical conclusion. On one side the severity pattern shows persistence. On the other, UNOCHA and its partners collected several new indicators during the two months. These they combined in two sub-indices – physical and socio-economic vulnerability – that are scarcely correlated with the components of the first severity index. In other words, they contribute genuinely new information.
Still the question can be asked whether this second severity measure added much value, the way it was done. Both times, the assessment produced one global severity index value for each VDC. The second index, in June, included indicators with primary information about three sectors only – one each about shelter, health and food security. It was a considerable feat to collect them in over 600 VDCs. But it is hard to imagine that the humanitarian community was able to fine-tune its response, leaning on one combined severity index, to this much variability among small local communities.
The operational value of the second assessment might have been greater if instead more effort had been expended to estimate persons in need. These estimates might have been limited to the district level, or perhaps to district headquarter towns, plus and small samples of outlying communities. There were only twelve affected districts. Estimates, in sectors critical to survival and recovery, of persons in acute need as well as of those in moderate need would not only provide a basis for severity estimates, but they would also be of direct value to response planners and implementers.
For Nepal, that question is now academic. We cannot generalize from this one study to severity patterns in other places, disaster types or time intervals. But whenever we believe that the severity pattern is stable, then the initial estimates may buy us enough time to refine subsequent assessments to produce greater operational information value.