12-Robustness-Sensitivity-Analysis.Rmd

Robustness and Sensitivity analysis
======

## Assessing exclusion & Inclusion errors


[Targeting approaches](http://pubs.iied.org/pdfs/10796IIED.pdf) are numerous and all come with pros and cons. A good practice is often to combine them through an hybrid approach in order to reduce both exclusion and inclusion errors).


 *  __Exclusion__: % of "false negative", i.e. the individuals who have a predicted indicator equal to 0 while the actual value is 1
 
 *  __Inclusion__: % of "false positive", i.e. the individuals who have a predicted indicator equal to 1 while the actual value is 0

In the case of targeting, prediction should be analysed in terms of benefit and what is preferred at field level:

* high exclusion means that **appeal process will be overloaded** (which will also implies hidden cost in terms of staffing to address those appeals), 

 * while high inclusion may results in **higher programme cost**. In other terms, where is the higher cost: assistance with false positive or case management with false negative.