The Effect of Early Flood Warnings on Mitigation and Recovery During the 2010 Pakistan Floods - Reducing Disaster: Early Warning Systems for Climate Change - page 246

the tehsil level. The most recent MICS surveys took place in 2007-2008 and 2011,
providing representative household data for the periods preceding and following
the 2010 floods. We obtained access to parts of the MICS data from the Punjab
Bureau of Statistics (PBOS), a provincial government agency that administers the
survey on behalf of UNICEF.
The 2011 MICS survey asked each respondent if the 2010 floods had affected the
household. Based on the responses to this question, the PBOS classified a cluster as
being “flood-affected” in 2010 if all of the randomly selected households in
the cluster responded “yes” to this question and “non-flood-affected” if any of the
households in the cluster responded with a “no” to this question. 4 From the listing
of flood-affected clusters, we determine the percentage of flood-affected clusters in
each district. 5
Based on information from both the MICS 2011 and the McRam survey 2010, the
five districts with the highest number of 2010 flood-affected clusters were Rajanpur,
Muzaffargarh, Layyah, Dera Ghazi Khan, and RahimYar Khan. Due to security
reasons,femalestaffandenumeratorscouldnotvisitRajanpurandDeraGhaziKhan,
sowethereforeconcentrateoursurveyinthethreeremainingdistricts:Muzaffargarh,
Layyah, and Rahim Yar Khan. Flood maps obtained from the United Nations McRam
survey, the Punjab Provincial Disaster Management Authority (PDMA), and LUMS
confirm that each of the three districts lie across the border of flooded and non-
flooded areas. According to the MICS 2011, 9 % of the clusters sampled in Rahim
Yar Khan can be classified as flooded; while 18 % of the clusters in Layyah and 51 %
oftheclustersinMuzaffargarhwere“looded”in2010.
Using a total listing of all villages in the three focus districts, we select villages
based on propensity scores. We use pre-flood data from the 2007-2008 MICS wave,
including household wealth and livestock, income, occupation of household head,
access to utilities, literacy, health, and access to public infrastructure, to create a
score of characteristics correlated to the propensity to be flooded. By matching pro-
pensity scores based on these characteristics, we obtain a control group that was not
flooded during 2010 but had similar propensity to be flooded based on socio-
economic factors. 6
4 A cluster was designated as flood affected only if all the households in the cluster responded to
the question of being affected by the flood in 2010 with a “Yes.” This was done to make sure there
are no errors due to the migration of households into and out of the cluster since 2010-2011, when
the survey was conducted and only clusters where there is minimum likelihood of migration in and
out are selected as flood affected.
5 Note that the MICS is a representative random sample of the total population, not a census of all
households, so the percentage of flood-affected clusters calculated is approximate but based on the
random sample.
6 Note, in using both the 2007-2008 and 2011 rounds of MICS, we have effectively restricted our
sample to villages that were common in both rounds. Since the samples in both years were com-
pletely random, any villages that have been sampled in both rounds are also random - there is no
reason to suspect any bias in the selection of these villages. Note also, that resampling the same
villages in 2011 that were sampled in 2007-2008 does not mean that the same households were
sampled, since the selection of households is random.