## complicate the assessor's life [GxP / QC / QA]

Hi Mittyri,

» well, 'smart' analytical labs with intentions to correct the results based on the intermittent analysis will change the order of subjects in analysis.

Agree. Nevertheless, the analytical report should contain information at which dates which subject was analyzed.

» The unveiling is still possible, but will require some boring process of order extraction from analytical docs

If one suspects that the report is fabricated, one has to resort to the raw data.

Since the runs test was not useful, I tried the Durbin-Watson test for autocorrelated errors (available in the car package). A dead end as well, even for my extremely manipulated third data set. However, testing the slope of a simple regression looks promising. Here for my second data set:

Deviating from the paper I decided to use residual = predicted – observed.
That’s more common.

For the original (real world) data set none of the slopes was significant (0.7252, 0.8362, and 0.8850). For my third data set I got p-values of 5.233·10–4, 1.161·10–4, and 1.699·10–5, respectively. We know that regression is sensitive to extreme values (see this post for a famous example). Here it is a desirable property.

This one is telling:

Dif-tor heh smusma 🖖
Helmut Schütz

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes