bayesplain.compare_rates

bayesplain.compare_rates(events, exposure=(1.0, 1.0), prior='jeffreys', labels=None, estimand='ratio', unit='', n_draws=None, seed='unset')[source]

Compare the rate of events between two groups.

The rate ratio has an exact posterior – a scaled beta-prime distribution – so the headline answer involves no sampling. The difference in rates is drawn from the two exact Gamma posteriors and subtracted.

Parameters:
events : array_like

Two event counts, [y1, y2].

exposure : array_like, default (1.0, 1.0)

Two exposures, [t1, t2], in the same units. Unequal exposure is the whole reason to compare rates rather than counts: 40 crashes in ten years is a lower rate than 12 in two.

prior : str, tuple, or RatePrior, default 'jeffreys'

Prior applied to each group’s rate.

labels : sequence of str, optional

Group names. Defaults to ('group 1', 'group 2').

estimand : {'ratio', 'difference'}, default 'ratio'

Which comparison to report, oriented as group 2 relative to group 1. The ratio is the conventional one for rates, and is exact.

unit : str, optional

How to print a rate, e.g. "per year". Used for the difference and for each group’s own rate.

n_draws : int, optional

Number of draws. Defaults to the package setting.

seed : int or None, optional

Seed for the draws.

Returns:

Call .summary() for the full report.

Return type:

Result

Examples

Pedestrian collisions before and after a road diet: 31 in the four years before, 9 in the two years since.

>>> import bayesplain as bp
>>> res = bp.compare_rates(
...     events=[31, 9], exposure=[4, 2], labels=["before", "after"],
...     unit="per year",
... )
>>> round(res.point(), 2)
0.59
>>> round(res.probability("<", 1.0), 3)
0.931