calc_HomogeneityTest() ... wrong calculation(?)

Question

calc_HomogeneityTest() ... wrong calculation(?)

Closed this issue 7 years ago · 3 comments

Expected behaviour

The test should test against the null hypothesis that the values including the standard errors
are resulting from a common distribution. If P is small enough the null hypothesis should be rejected.

Observed behaviour

P = 1 is returned even for datasets which are a mix of two distributions.

Running mini example

library(Luminescence)
calc_HomogeneityTest(do.call(what = rbind, ExampleData.DeValues))

Reason

Please check code line 101, the lower.tail = FALSE appears to be the wrong argument setting.

Answer 1 · 2017-12-15T11:11:39.000Z

Ok, this was probably not the reason, but the log transformation is wrong. The standard error is not just simply the logarithm if the dose log-transformed. The relationship is that the log-dose is 'effectively the relative standard error of the dose' (Galbraith, 2003, see also Galbraith et al., 1999).

Answer 2 · 2017-12-15T11:30:00.000Z

@tzerk Please double check. Now it should be ok, however, please check again.

Answer 3 · 2017-12-15T15:38:15.000Z

You are absolutely correct. But I have no idea how this error was introduced, as I did also cross-check with the worked example in Galbraith (2003). The proposed fix produces the correct results and thanks for including the test.