Data Envelopment Analysis ( Dea )

Data Envelopment Analysis (DEA)

Data Envelopment Analysis calculates the best-practice frontier for a given sample using piecewise linear programming. It then indicates the relative inefficiency of other units by measuring the distance between these units and the best practice frontier. These models can be input oriented (seeking to minimize inputs while retaining a constant output) or output oriented (seeking to maximize outputs while holding inputs constant). In this instance outputs would be factors indicating general wellbeing, while inputs would be the resources available to the state. One advantage to this approach is that it is highly modular and flexible for use with multi-output and multi-input treatments. Additionally, DEA can be extended to construct dynamic indices by use of the Malmquist Index technique. However use of this technique depends upon access on an inclusive set ‘output’ and ‘input’ variables necessary to model the relevant technology.

Input vs. Output Oriented Indices

When constructing an index the decision must be made to included inputs and/or outputs of wellbeing in the list of indicators. Inputs constitute any variable which contributes to (or detriments from) a society’s wellbeing, such as money spent on education or security. Outputs are any variables which are the result of a level of development, such as life expectancy or percentage of population with a high school degree. Models which include inputs have the disadvantage of greater incidence of collinearity and double counting. For example, the amount of spending on police force could be a function of the amount of crime in a state. By analyzing output variables exclusively these issues are minimized. This isn’t to say that output variables ...

... middle of paper ...

... level.

Within each of the dimensions weights would need to be assigned differently, as survey data does not exist on the relative importance of every indicator. However here we could incorporate various nested weighting schemes without compromising the model. We should include in the report a test of robustness to demonstrate how consistently rankings hold under shifting indicator and dimension weights.

The novelty of this approach is another strong mark in its favor of its adoption. Until very recently it would have been impossible to implement this methodology at the scope of an international index, however the Gallup poll should provide all the variables necessary to do so. Adopting a stated preference weighting scheme would make our index among the first of its kind and would constitute a serious contribution to the body of international development research.