A team from Statistics without Borders (SWB) and the Humanitarian Data Exchange (HDX) collaborated to analyze casualty from the Nepal earthquake of 25 April 2015. The objective was to better understand the factors that contribute to mortality and to help in focusing response efforts through predictive analytics.

While indices such as INFORM are used in prioritizing humanitarian responses, our analysis is the first application of statistical methodologies for Nepalese earthquakes. The result is the Nepal Earthquake Hazard Index, which can be applied in future Nepalese earthquakes to focus on where help is needed the most.

We used the data on HDX including the types of housing construction, demographics, and the degree of shake measured by the Modified Mercalli Intensity index (MMI) to study the relationship with mortality through the development of a statistical model. This was transformed into a score (i.e., index) for prioritizing future earthquake response activities.

The Nepal Earthquake Hazard Index, ranging from 1 to 10 where 10 indicates the highest risk, comes in three categories: death risk in counts, death risk in rate (death count divided by total population), and death risk in density (death count divided by area). The maps below shows the differences based on the three categories.


According to the Index, Sindhupalchok and Kathmandu were expected to have the highest death counts (Death Count Index = 10). Kathmandu was expected to have the highest concentration of deaths per unit area (Death Density Index = 10) and Manang was expected to have the highest concentration of deaths per population. This reflects the larger total population and smaller total area of Kathmandu and the smaller total population of Manang, and may have implications in the planning and execution of response efforts.

The key insights from this analysis are:

  • Predicting deaths also predicts injuries. The analysis focused on deaths due to the ambiguity around injuries. The severity of the injuries was not available in the data, and an injury could only be a transitional state to the eventual quake-related fatality. However, deaths and injuries were highly correlated. A model predicting deaths equally predicted injuries.
  • The determining factor in whether there will be quake-related deaths is the degree of shake. While this is intuitively obvious – if there is no shake, there are no quake-related deaths – this validated that the model identified factors that logically made sense.
  • Mortality is related primarily to roof construction and the degree of shake and is less related to wall construction. This may be because certain types of roofs generally go with only certain types of walls; however, the wall construction variables by themselves were inadequate inconsistent in predicting deaths.
  • The risk associated with housing structures vary highly, and areas with many high-risk walls do not necessarily have many high-risk roofs. The top two maps of this dashboard show the actual death count and MMI, while the subsequent maps show the aggregate roof and wall risks. Houses with riskier roof constructions are concentrated along the northern border of the country, whereas houses with riskier wall constructions are concentrated in the western part of the country. This implies that, for example, the impact of a strong earthquake in the northwestern part of the country may be different from the same severity of earthquake in the southeastern region. This finding may affect earthquake preparedness planning, especially given the importance of roof construction.


While the analysis was done primarily at the district level, the resulting Nepal Earthquake Hazard Index can account for gradation in MMI at the Village Development Committee (VDC) level. Should another earthquake hit Nepal, this index should prove useful in prioritizing emergency response efforts.

The full research paper has been submitted to the Statistical Journal of the International Association for Official Statistics and it is expected to be published later this year.