About Rapfish‎ > ‎

Uncertainty

Rapfish currently takes account of two types of uncertainty
1. errors in estimating individual scores , using a Monte Carlo technique, and
2. the contribution of each attribute to the results of an evaluation field, using a jackknife leverage technique. 


1. Errors in scores
Upper and lower limits are assigned to individual scores as each score is evaluated.
For example, "the most likely score is 6/10; it could not be lower than 5/10, but could be as high as 8/10". The most likely value does not have to be centrally located within the error range. For Monte Carlo runs, probability distributions are normally taken as square wave: but triangular or Normal distributions can be used if desired. Usually 100 or 1000 Monte Carlo runs are performed and upper and lower 95% or 90% tiles taken as the results. An example from the Code of Conduct analysis Management Objectives field is given below


2. Attribute leverage
The leverage (or influence) of each attribute on the overall results is tested using a jacknife-like technique by dropping each attribute from the analysis in turn. The percentage contribution of each attribute is calculated from the change in ordination position when the attribute is removed . Where most attributes in a field contribute more or less equally to the results, that evaluation field is genuinely multivariate. On the other hand, if only one attribute strongly determines the results, it is likely a important single indicator, and, possibly, some of the attributes may be dropped. An example of a leverage plot from the social-economic field, one of the six evaluation fields of the Code of Conduct analysis, is provided below: the six attributes are very close in leverage; of these meeting the needs of coastal communities and indigenous people are the most discriminating among fisheries.


Comments