Reliability procedures for categorical data in Performance Analysis

Abstract

A number of papers have discussed reliability assessment techniques but none have previously limited themselves to methods appropriate to the collection of categorical data on the nominal scale i.e. where events are recorded within defined categories, as in many instances for performance analysis. This paper makes a number of recommendations including the use of a gold standard coding to compare the typical analyst’s codes against. Reliability measures should reflect the way in which the notation data is analysed so that assessments can be made regarding the extent to which each variable presented in the results was coded accurately. The Yule’s Q statistic, based on the odds ratio, is recommended as the reliability statistic of choice for categorical data due to the intuitiveness of the measure, its ease of calculation and its apparent applicability to the decision making of performance analysts. Furthermore a value of 0.95 or above should be deemed an acceptable value to determine confidence in the analyst’s ability to place an event into a category reliably.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 108 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.