18 Bridging the Information Gap: Mapping Data Sets on Information …
215
voice SMS and developing an app and dashboard to enable two-way information
exchange between affected communities and responders. The pilot areas are riverine
islands in northwest Bangladesh, so-called char-islands, which are part of the densely
populated floodplains where many poor and vulnerable people live. The focus was
on the most recent and severe river flood of the last years, namely, the floods of 2014
that affected almost two million poor and vulnerable people living in nine districts
in northwest Bangladesh (Wahed et al. 2014). About 1 year after these floods, we
performed 13 oral history semi-structured interviews of which 11 in Dhaka (national
NGOs (active in the JNA consortium) and Department of Disaster Management) and
two in Sirajganj (one with a farmer and fisherman, and one with the director and
his two co-directors of the local NGO, MMS). We held one focus group discussion
with seven disaster responders of MMS, one focus group with 15 people living on
the chars (imam, teachers, entrepreneurs, part of the volunteer disaster management
committees), and one focus group with 13 local government officials [Upazila and
Union Disaster Management Committee, civil defense organization (Ansar VDP)].
So, in total, we got input from 51 people. We arranged the first batch of interviewees
based on our existing network and such that we would have a representative cross
section. Subsequently, we used a snowballing approach to grow our sample considering the availability of respondents and useful references. Although focal point in
these sessions was the flooding of 2014, we did allow interviewees also to draw from
their earlier or more recent disaster management experiences. All interviews were
transcribed. The focus group discussions were done with an interpreter, usually at
an open noisy marketplace, and could not be literally transcribed. Instead, we used
the notes taken. All interviews and notes were subsequently labeled using NVIVO
10 for Windows and coded based on three themes, i.e., Activity, Decision, and Information Need. We used inductive coding to have subthemes emerge from the data.
For each of these themes, clustering was done based on experience emerging from
the familiarization phase, domain knowledge, and literature study. In addition, we
asked the interviewees to validate our transcribed interviews. We asked two domain
experts to validate and expand on the list of needs. We also used the lists of Activities
and Decisions to identify possible discrepancies. For the second research question,
we used, in addition to the interviews, Internet search and literature study. In that
way, we could make an inventory of the data sets that were available during the
flooding of 2014. For the last research question, we singled out all the indicators per
data file and manually determined the match with a subtheme information need. We
scored the match as Yes, No, or Partly. Afterwards, we used constrained COUNT
formulae to calculate the coverage per disaster data source of the subtheme information needs. We used approximately the phases as defined in the Multi-Sector Initial
Rapid Assessment (MIRA) (MIRA 2015) to label both the data sets as well as the
information needs. The phases consisted of before (1), the first 72 h (2), the first 2
weeks (3), and the first 2 months (4). Table 18.1 gives an example for three data
sets and information needs. Data B covers 33% of the information needs if no time
constraints are considered. With time constraints, none of the information needs are
met, since the information was needed already in phase 1 but came only available in
phase 4.
215
voice SMS and developing an app and dashboard to enable two-way information
exchange between affected communities and responders. The pilot areas are riverine
islands in northwest Bangladesh, so-called char-islands, which are part of the densely
populated floodplains where many poor and vulnerable people live. The focus was
on the most recent and severe river flood of the last years, namely, the floods of 2014
that affected almost two million poor and vulnerable people living in nine districts
in northwest Bangladesh (Wahed et al. 2014). About 1 year after these floods, we
performed 13 oral history semi-structured interviews of which 11 in Dhaka (national
NGOs (active in the JNA consortium) and Department of Disaster Management) and
two in Sirajganj (one with a farmer and fisherman, and one with the director and
his two co-directors of the local NGO, MMS). We held one focus group discussion
with seven disaster responders of MMS, one focus group with 15 people living on
the chars (imam, teachers, entrepreneurs, part of the volunteer disaster management
committees), and one focus group with 13 local government officials [Upazila and
Union Disaster Management Committee, civil defense organization (Ansar VDP)].
So, in total, we got input from 51 people. We arranged the first batch of interviewees
based on our existing network and such that we would have a representative cross
section. Subsequently, we used a snowballing approach to grow our sample considering the availability of respondents and useful references. Although focal point in
these sessions was the flooding of 2014, we did allow interviewees also to draw from
their earlier or more recent disaster management experiences. All interviews were
transcribed. The focus group discussions were done with an interpreter, usually at
an open noisy marketplace, and could not be literally transcribed. Instead, we used
the notes taken. All interviews and notes were subsequently labeled using NVIVO
10 for Windows and coded based on three themes, i.e., Activity, Decision, and Information Need. We used inductive coding to have subthemes emerge from the data.
For each of these themes, clustering was done based on experience emerging from
the familiarization phase, domain knowledge, and literature study. In addition, we
asked the interviewees to validate our transcribed interviews. We asked two domain
experts to validate and expand on the list of needs. We also used the lists of Activities
and Decisions to identify possible discrepancies. For the second research question,
we used, in addition to the interviews, Internet search and literature study. In that
way, we could make an inventory of the data sets that were available during the
flooding of 2014. For the last research question, we singled out all the indicators per
data file and manually determined the match with a subtheme information need. We
scored the match as Yes, No, or Partly. Afterwards, we used constrained COUNT
formulae to calculate the coverage per disaster data source of the subtheme information needs. We used approximately the phases as defined in the Multi-Sector Initial
Rapid Assessment (MIRA) (MIRA 2015) to label both the data sets as well as the
information needs. The phases consisted of before (1), the first 72 h (2), the first 2
weeks (3), and the first 2 months (4). Table 18.1 gives an example for three data
sets and information needs. Data B covers 33% of the information needs if no time
constraints are considered. With time constraints, none of the information needs are
met, since the information was needed already in phase 1 but came only available in
phase 4.