Skip to content
Prev 2512 / 5636 Next

[R-meta] Dear Wolfgang

Hi JU,

Responses to your follow-up questions below.

James
On Thu, Dec 10, 2020 at 4:44 PM Ju Lee <juhyung2 at stanford.edu> wrote:

            
As a statistician who doesn't know anything about the subject-matter, I'm
afraid I don't really feel qualified to answer this question. This requires
making judgements about the relevance for your research questions of the
different types of sites, species, measures, time-points, etc. included in
the agency data and in the peer-reviewed data. I would not recommend using
a different protocol for agency data than for peer reviewed data just for
the sake of shrinking down the dataset. I think the thing to do is focus on
which data are relevant and suitable for answering the questions you have.
It doesn't seem unreasonable to me to run your analyses with and without
the agency data, and reporting both sets of results. The problem with this
approach would be figuring out how to interpret everything and draw
bottom-line conclusions if the results aren't consistent. That's why I
suggested running the analyses separately for the peer-reviewed data and
the agency data. It think that would let you get more of a purchase on
what's going on.
If you model the agency data on its own, I don't think it would make sense
to include study as a random effect. The results would be conditional on
the available agency studies, so you'd have to interpret them accordingly.
But, with so much available data, there would still be many sources of
variation that could be investigated and modeled (as Wolfgang noted in his
reply).
Message-ID: <CAFUVuJwT_xmReYBBV98pgjxMTYVq8QKOqmUrg5hAR2v7F8q1BQ@mail.gmail.com>
In-Reply-To: <BYAPR02MB55597FB62E2CCD533D83ED84F7CB0@BYAPR02MB5559.namprd02.prod.outlook.com>