Closed tommorris168 closed 4 months ago
There are no current tests on the content of the matrix because we only have one example where we know what the expected output is, and that uses contrast-level data rather than arm-level data. Also it only covers one case. I will use it where possible in a test, and then may have to do some very small examples and calculate the expected output by hand. It's not ideal but it might be all we can do.
Closing, and making the fix on a new branch where tests will be added as well.
@tommorris168 Can this branch be deleted?
Weinblatt1999 now has a contribution when
weight_or_contribution = "contribution"
.If you look at the row for Strand2006 and compare
weight_or_contribution = "contribution"
toweight_or_contribution = "weight"
you see that this study now has zeros whenweight_or_contribution = "contribution"
. In this case it is correct, because both arms have the same outcome (R = 5, N = 40), so the treatment effect is zero and the arms cancel each other out.