A hypothesis test on an observation record

Runs one of the ten DataFrame hypothesis-test facades over the exact series of an analysis_data() frame (or a plain numeric vector), or all ten at once with method = "summary_hypothesis". The four two-sample tests split the record at split_index, comparing observations with a data index below it against those at or above it – the split is on the record's INDEX, not on array position, so it agrees with the split a caller would get by re-running the test on a record whose observations were supplied out of order.

Usage

analysis_data_hypothesis_test(
  data,
  method,
  split_index = NULL,
  lag_max = NULL,
  use_log10 = FALSE
)

Arguments

data

a numeric vector of observations, or a corehydro_data object from analysis_data() (only its exact series is read).

method

one of "jarque_bera" (normality), "ljung_box" (autocorrelation), "equal_variance_t" / "unequal_variance_t" (difference in means, Student's / Welch's), "f" (difference in variances), "linear_trend" (trend), "wald_wolfowitz" (runs test for independence), "mann_whitney" (homogeneity / jump), "mann_kendall" (homogeneity / trend), "unimodality" (a Gaussian-mixture likelihood-ratio test, where a SMALL p-value is evidence AGAINST unimodality), or "summary_hypothesis" (all ten at once – see Details).

split_index

the record index to split the sample at; required by "equal_variance_t", "unequal_variance_t", "f", and "mann_whitney", ignored otherwise.

lag_max

the maximum lag for "ljung_box"; NULL (the default) uses the library's own default rule. Ignored by every other method.

use_log10

logical; test the log10-transformed values instead of the real-space values.

Value

A named numeric vector of length 1, named method: the 2-sided p-value. For method = "summary_hypothesis", a named numeric vector of length 10 carrying every test's p-value, named by the library's own descriptive test names and in its own order.

Details

method = "summary_hypothesis" is the library's own ten-test summary, and it behaves differently from calling the ten tests individually in three ways worth knowing, all inherited from upstream:

• split_index is OPTIONAL. Left NULL (or given a value outside the record's index range) it selects the midpoint split rather than erroring.

• Any single test that fails turns the WHOLE result to NaN, rather than reporting the nine that worked. A record shorter than the 20 observations "mann_whitney" needs will therefore come back all-NaN.

• "ljung_box" is run at the library's default lag, ignoring lag_max.

See also

analysis_data_statistics() for the summary-statistics facades, analysis_data_summary() for plotting positions.

Examples

peaks <- c(122, 244, 214, 173, 229, 156, 212, 263, 146, 183, 161, 205)
analysis_data_hypothesis_test(peaks, "mann_kendall")
#> mann_kendall 
#>    0.8370115 
analysis_data_hypothesis_test(peaks, "equal_variance_t", split_index = 6)
#> equal_variance_t 
#>        0.8396897