Gold? [🇷 for BE/BA]

posted by Helmut Homepage – Vienna, Austria, 2008-10-29 14:03 (6076 d 00:55 ago) – Posting: # 2602
Views: 33,492

Dear all!

I haven’t performed analyses of incomplete data of a 2×2×2 study until now – and haven’t seen one in my work as a consultant performed by anybody else.

Another quote* (referring to their example 3.2 in 49 subjects):

Before we can continue to fit a linear model to the (log-transformed) data, we must decide what to do with the data from those subjects who did not provide a value in both periods. Such a comparison is not possible for those subjects with only a single value. If however, there are two such subjects and has a value only on T and the other has a value only on R, then a between-subject com­parison of T and R is possible by taking the difference of these two single values. However, the precision of such a comparison will be low because the between-subject variation […] is much higher than the within-subject vari­abi­lity. Because we have assumed the subject effects are random variables, these between-subject comparisons can be recovered in the analysis if we fit what is known as a mixed model. […] However, the recovery of between-subject infor­ma­tion on the comparison of T and R is unlikely to make much difference to the results, and so nothing of significance will be lost by ignoring the data on those subjects that provided only a single value. To justify this assertion, we will also report the results of fitting the mixed model to the complete data set […]

Personally I think it’s a little bit ambiguous to perform an analysis of the complete data set (simple model) as primary and ‘justify’ it by running a second one on the full (mixed) model. For our example the width of the confidence interval is 0.198015 (simple model) and 0.189859 (full model). As DLabes pointed out in this post the mixed model is scientifically valid, but I would be wary to present it as the only one – since without suitable software it’s not possible to reproduce the results. If done the other way ’round (from Patterson/Jones), I would expect a lot of discussions (imagine if model #1 shows BE, and #2 does not)…

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

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

Complete thread:

UA Flag
Activity
 Admin contact
23,424 posts in 4,927 threads, 1,671 registered users;
31 visitors (0 registered, 31 guests [including 21 identified bots]).
Forum time: 15:59 CEST (Europe/Vienna)

Truth and clarity are complementary.    Niels Bohr

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5