p_value {keyperm}

Convert results of permutation test for keyness to p-values

Description

Calculate p-values from the results of keyperm() with output = "counts".

Usage

p_value(results, alternative = NULL)

Arguments

results

results from permutation test. Must be of class keyperm_results_counts (obtained by setting output = "counts" in keyperm())

alternative

direction of p-value to calculate, one of "two.sided", "greater", "less". Defaults depend on the scores used. See details.

Details

Valid (slightly conservative) p-values are calculated from an object of class keyperm_results_counts that is obtained by running keyperm() with output = "counts". keyperm_results_counts is a matrix with three columns that contain the counts of generated permutations that resulted in a score strictly less than, equal to and strictly greater that the observed score.

For a one-sided p-value we use

pvalue_greater = (no. greater + no. equal + 1)/(no. of perms + 1)

or

pvalue_less = (no. less + no. equal + 1)/(no. of perms + 1)

Adding 1 in both the numerator and denominator amounts to including the observed values. This results in a slightly conservative p-value, but guarantees that the test is valid for any number of random permutations. It also means that never a p-value of zero is returned but the minimum possible p-value is 1/(no. perms + 1).

The two-sided p-value is calculated by

pvalue_twosided = 2 * min(pvalue_less, pvalue_greater)

(values larger than 1 are set to 1).

If alternative is not specified by the user, different defaults are used depending on the scoretype (which is included as an attribute in the keyperm_results_counts object). Since for llr and chisq, large values indicate a great deviation from equal frequencies without indicating the direction, alternative == "greater" is basically the only alternative of interest and is used as a default. For diff and logratio large absolute values indicate a great deviation from equal frequencies, and positive values correspond to higher frequencies in A, negative frequencies correspond to a higher frequency in B. For these scoretypes, the default is alternative = "two.sided". If only "positive" keywords for A with respect to B are desired, use alternative = "less".

Value

a numeric vector of p-values.