Exact p-values for the rows of a year-over-year table

Adds a two-sided exact p-value to each row of a period-over-period change table of counts from yoy, testing whether the count changed by more than chance between the two periods. Use it when you want a test to go with each change, and then pass the table to scan_adjust to correct for the number of rows tested.

Usage

yoy_pvalues(y)

Arguments

y

An rmbl_yoy object from yoy with units = "count".

Value

y, with its class and attributes, and an added numeric p_value column; NA where there is no comparison period or both counts are zero.

Details

For count units the interval in yoy() is the exact conditional binomial interval for the ratio of two Poisson means; this is the matching two-sided test. The test conditions on the total of the two periods and asks whether the split is compatible with equal means: with current count \(a\) and previous count \(b\), \(a \sim \mathrm{Binomial}(a + b, 1/2)\) under the null, and the p-value is that of stats::binom.test(a, a + b, p = 0.5). The two periods are treated as having equal exposure; for changing populations use rate_change and scan_adjust, which weight by the exposures. Counts are rounded to whole numbers.

Errors: y not a yoy() result; a yoy() result whose units are not counts (percentages and measured values have no Poisson test). A progress indicator is shown for long tables in an interactive session.

See also

scan_adjust, yoy.

Examples

seg <- data.frame(year = rep(2022:2023, each = 3),
                  site = rep(c("A", "B", "C"), 2),
                  n = c(30, 50, 80, 45, 52, 79))
y <- yoy(seg, value = "n", period = "year", by = "site")
yoy_pvalues(y)$p_value
#> [1]        NA 0.1053423        NA 0.9211910        NA 1.0000000

# the same test by hand for site A: 45 against 30
stats::binom.test(45, 75)$p.value
#> [1] 0.1053423

# anything that is not a yoy() table is refused
try(yoy_pvalues(seg))
#> Error : `y` must come from yoy()