bayesplain.contingency

bayesplain.contingency(table, prior='uninformed', scheme='independent_multinomial', fixed='rows', effect='auto', threshold=None, row_labels=None, col_labels=None, n_draws=None, seed='unset')[source]

Test whether the rows and columns of a table are related.

Parameters:
table : array_like

Counts, shape (R, C). Rows are the groups you sampled; columns are the categories you counted within each group. A pandas crosstab works directly, and its labels are picked up automatically.

prior : str, float, or ConcentrationPrior, default 'uninformed'

Dirichlet concentration over the cells. See bayesplain.priors.describe().

scheme : str, default 'independent_multinomial'

Which Gunel-Dickey sampling scheme the study design justifies; one of bayesplain.core.dirichlet_multinomial.SCHEMES. The default matches “we sampled a fixed number of cases in each group”, which is how planning data is usually gathered. The choice only affects the Bayes factor, never the posterior over the effect size.

fixed : {'rows', 'columns'}, default 'rows'

Which margin the design fixed, for the independent-multinomial scheme.

effect : {'auto', 'cramers_v', 'log_odds_ratio'}, default 'auto'

The effect size to put a posterior on. 'auto' uses the log odds ratio for a 2x2 table, where zero means exactly no association, and Cramer’s V otherwise.

threshold : float, optional

The value the reported probability is measured against. Defaults to 0 for the log odds ratio and to SMALL_EFFECT_V for Cramer’s V.

row_labels : sequence of str, optional

Names for the rows and columns. Read from a pandas object when not supplied.

col_labels : sequence of str, optional

Names for the rows and columns. Read from a pandas object when not supplied.

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

The doll-preference study of Hraba and Grant (1970), as analysed by Jamil et al. (2017): 62 of 89 African American children preferred a black doll, while 60 of 71 white children preferred a white doll.

>>> import bayesplain as bp
>>> res = bp.contingency(
...     [[62, 27], [11, 60]],
...     row_labels=["Black children", "white children"],
...     col_labels=["black doll", "white doll"],
... )
>>> round(res.bayes_factor().log_bf10, 2)
23.03
>>> res.frequentist.pvalue < 0.001
True

Notes

The Bayes factor is validated against the published values in Jamil et al. (2017) for all four sampling schemes.