Working at the wrong zoom level?

In my previous blog post I described the three different uses of social media during disasters. One of them was utilizing social media for crisis information management purposes. In crisis information management we commonly split it up into three different phases, data collection, data processing and data analysis. While I see tremendous opportunities in outsourcing data processing and data analysis to crowds through social media, then I disagree with the focus we have had on data collection through social media during disasters.

Part of the problem we face is that this is being approached from two different sides that don’t understand each other well enough. As I mentioned in my earlier blog post then we must find the common middle ground between what the technology groups think can be done and what the humanitarian groups need to have done.

I had the great opportunity to take part in a research study commissioned to figure out exactly how to get over this gap between those two communities. Although the study will not get released until end of March, there were some interesting things I learned while doing interviews with a number of response organizations.

One of the key point I heard from both the humanitarian and military side was that reports coming in through social media turned out to be quite inaccurate at the street level. In other words, trying to use social media as a replacement for a 911 system does not work. It however did provide good situational overview at the neighborhood level.

Again one thing to realize for those that push for detailed collection of reports is that during a large scale disaster you cannot process timely every single request that comes in. Those that have worked in a 911 center during a large scale disaster can confirm this. You try to prioritize the reports and a great deal of reports end up in a queue that gets serviced a long time later, at which time citizens may have handled the issue themselves.

This may cause the fact that up to 90% of these reports may not be accurate when a first responder finally gets to the area. Given that high of a rate of inaccuracy shouldn’t we just ignore this social media reporting stuff all together?

First of all lets exclude the high priority reports from the discussion. These are reports like “person stuck under a rubble”, “ambulance needed for injured person”, etc. I do feel these warrant a different channel than other reports. I would like to see those reports handled quickly and then prioritized by a team of experienced responders which are NOT located in the field. They then forward the prioritized requests to the appropriate response organization on the ground. For the case of people trapped following an earthquake, the process is already being looked at and established.

Secondly there are all the situational reports, reports of needs (water, sanitation, etc.). For this kind of information we need to understand two important things:

1) Humans have a drive for helping themselves. This means they will try to find the things they need. This can mean that when the help finally arrives, the person reporting the need may have already met that need.

2) Humans have two legs. They will move between areas. A person that reports a need from a particular location will most likely not be there one hour later. As such we may not find the person reporting the need.

Due to these (and other) reasons it is important for us not to track those requests for needs on an individual basis, but rather to look at them at the neighborhood or area level and identify trends. Are we seeing a large number of requests come from a particular area? Are we seeing changes in the what is being requested? Is there an increase or decrease in particular areas? These are all important questions that crowd sourced data can answer us, but my feeling is we have tried to look at both the high priority reports and the situational reports at too deep a zoom level so to speak.

The tools we develop must therefore in my mind have two ways of tracking information. One is for the high priority reports and for those it is important to be able to prioritize them and then mark when they have been taken care of. The second one is taking all the other situational reports and aggregating them into logical administrative boundaries (neighborhoods, villages, etc.) and tracking them through time. We need to be able to identify these based on sectors/clusters so we can see where the need is for each sector. At the same time we must be able to “slide” a timeline to see the trends in the reports for each area.

We must also think seriously whether putting all of this together on a map like we do today in Ushahidi is the right approach. There are certainly other visualizations that would make more sense for some of this data.

It is my hope that this blog entry causes some good discussion amongst the two communities and we figure out ways to really make social media an effective tool in large scale disasters.

Published February 23rd 2011 at DisasterExpert