Field-Study Statistics: A Reviewable 2 By 2 Table

Field-Study Statistics: A Reviewable 2 By 2 Table

animation
testing
validation
field-trials
Try Change Exposed group: outcome count to 0.
Observe Watch Observed risks, Risk ratio, Odds ratio, and Approximate 95% interval.
Explain Explain why a 0.5 continuity correction is applied when at least one cell is zero.

Technical boundaries. This simple analysis does not repair confounding, selection bias, missing telemetry, or changed outcome definitions. An interval spanning 1 is compatible with no multiplicative difference; a narrow interval does not prove causality.

Field-Study Statistics: A Reviewable 2 By 2 Table

Field evidence often starts as a count table. For example, compare a candidate gateway firmware (exposed) with the current firmware (control) and define the outcome before looking at the data, such as “a site-day with packet loss above the acceptance threshold.” Odds ratio and risk ratio answer different questions, and neither repairs confounding, selection bias, missing telemetry, or an outcome definition changed after collection.

An interval spanning 1 is compatible with no multiplicative difference under this simple analysis; a narrow interval does not prove causality. Preserve the four counts, denominators, site allocation, observation window, exclusions, missingness, firmware identity, outcome definition, and the code that produced the estimate. When exposure is not randomized, investigate site and time differences before attributing an effect to firmware.

Warning

Do not label an odds ratio as “times the risk.” When outcomes are common, odds and risk can diverge substantially. Choose the effect measure before analysis, show absolute risks beside it, and send small samples, repeated observations, clustering, or safety decisions for statistical review.

Back to the chapter