12 Robustness and Sensitivity analysis

12.1 Assessing exclusion & Inclusion errors

Targeting approaches 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.