Ignorability

In statistics, ignorability refers to an experiment design where the method of data collection (and the nature of missing data) do not have a significant influence on the intrepretation of the data. These designs are favorable because the inferences produced tend to be less influenced by the model chosen by the experimenter.

The idea of ignorability can refer also to observational data in which the predictors of "assignment" to a condition have been balanced between the two groups. For instance, although no experiment of the effect of attending a public versus private university could feasibly be done, it's possible to construct comparable groups of students at private and public universities; if the two groups cannot be distinguished in all of their observable characteristics, the assignment to public or private university can be said to be ignorable.

This idea is part of the Rubin Causal Inference Model, developed by Donald Rubin in collaboration with Paul Rosenbaum in the early 1970's.

External links

* [http://www.jair.org/media/1657/live-1657-2558-jair.pdf Ignorability in Statistical and Probabilistic Inference] by Manfred Jaeger

References

* Andrew Gelman, John B. Carlin, Hal S. Stern and Donald B. Rubin. Bayesian Data Analysis. Chapman & Hall/CRC: New York, 2004.