If we apply more stringent criteria and only include events for which there was data for at least three months following the onset of the disaster, we would arrive at a figure of around 17 million. This of course excludes any displacements that occurred in the last three months of 2016, but given that it is also based on data that is out of date in some cases, it may also be an overestimate – albeit a less significant one. At the same time, the partial nature of the data available for most events means that some displaced people, particularly those in protracted or chronic situations, are currently off our radar screen. The data is simply too scant to allow us to gauge the overall situation with any accuracy. Having said that, updated data indicates that at least 3.6 million people remain displaced due to three large disasters in recent years – the earthquakes in Haiti (2010) and Nepal (2015), and Typhoon Haiyan in the Philippines (2013). Given the considerable difference between an estimate of 89 million and 17 million, and the fact that our analysis is based on only 130 out of more than 2,000 events in our dataset, we have decided not to endorse either figure or anything in between. Instead, we have described the diffi- culty we faced as a clear call for more followup and data collection on people displaced by disasters. As our previous reporting has shown and as our displacement risk modelling suggests, many people are exposed and vulnerable to frequently occurring, low-intensity hazards (see part 1). This increases the risk of repeated displacement and undermines long-term development gains for these communities. Despite our lack of confidence in stock figures for displacement associated with disasters, our analysis of the time-series data available serves several purposes. It provides a useful stocktake of how much data has been captured, for how long and by whom. This has led to the identification of some good practices (see Philippines box below). It also sheds light on patterns of displacement for different hazards, which helps explain how we have estimated the total number of new displacements for those events. Flooding in the Indian state of Bihar in July and August 2016, for example, generated two distinct waves of new displacements which resulted in 1,670,000 displacements. A group of children play near the edge of the IDP camp on a dirt road in Khmer, Amran, Yemen. Photo © UNHCR/Rawan Shaif, February 2016 INSIDE THE GRID: Overcoming data shortfalls 81