Index construction: a pipeline approach for transparency and diagnostics

Indexes are commonly used to combine multivariate information into a single number for monitoring, communicating, and decision-making. They are applied in many areas including the environment (e.g. drought index, Southern Oscillation Index), and the economy (e.g. Consumer Price Index, FTSE). Developers, analysts, and policymakers tend to have their favorite indexes, but there is little transparency about their performance. Indexes are used like black boxes---raw data is entered and a single number is returned---with scarce attention paid to diagnostics. Interestingly, though, all indexes can be constructed using a data pipeline in a series of well-defined steps, regardless of their origin. This talk will explain this, and how you can use this structure to inspect the behavior of indexes in different scenarios. Our work coordinates the vast array of index research and development into a simple set of building blocks. This modular data pipeline is implemented in an R package, which contains some standard indexes, and allows others to be easily coded. We will illustrate the benefits of this framework using the drought index, and show how different versions, created by different parameter choices, can lead to potentially varied decisions.