The GEORISK deliverable on risk assessment explains in more details the data collection (link). We collected a little more than 60 questionnaires from the seven GEORISK countries where each risk from the register are placed on a 4×4 risk matrix. This page explains the method used for ranking the risks in the risk assessment results.
For the charts on the website, the methodology has evolved from the deliverable in order to have a more robust treatment of the data:
- First, since for some country we received less than 10 questionnaires, a smoothing method based on a hierarchical bayesian model was used. Basically, each answer gives for each risk a location on the 4×4 matrix. Taking into account all the answers we instead compute for each risk and each country the probability for a respondent to choose a cell.
- Since the data collected is semi-quantitative only, we do not compute a risk index anymore because it was based on an average of scores. Computing averages should instead only be performed on fully quantitative data.
- The new indicator is an « odd». We compute an « average» risk by summing of all the answers for all the risks, we then compare each risk to this average risk and we compute the odd : theta = P(risk > average) / P(risk < average); This indicator is thus a robust way to represent the ordering of each risk : the larger the odd, the higher the risk is.
- For computing the probability P(risk > average) and P(risk < average), we perform a cell by cell comparison of the two risks with the rule that a cell i,j is superior to a cell k,l if and only if i ≥ k and j ≥ l (and (i,j) ≠ (k,l)) ; For instance the cell (1,4) (in yellow) is superior to the cells (1,1), (1,2) and (1,3) (in green); and it is inferior to the cells (2,4), (3,4), and (4,4) (in orange). However we do not compare with the other cells (in light grey), for instance we do not choose between the cells (1,4) and (4,1);