Skip to content
Prev 19361 / 20628 Next

Estimation of variance components in random- and mixed-effects models

Thank you very much for your valuable insights and suggestions, Thierry
and Dr. Bolker! They have brought me to relearning and expanding my
statistical knowledge.

Treating the predictor "cohort" as an ordered factor is actually most
suitable for my original experimental design. I study cohort effects with
the levels "early, middle and late". The in-situ temperatures (T) I
measured by each level were in a nearly linear sequence "by chance": T2 =
T1 + delta, T3 = T1 + 1.7 * delta.

Because of this, linking cohort effect to temperature would probably be
more difficult. With the new model
    offspring trait ~ cohort +female length + (1 | female/clutch),
the only significant fixed predictor is cohort.L.

It suggests the levels "early, middle and late" are best interpreted as
having nearly equal-spaced intervals (similar to the temperatures). But it
does not support directly that the variance among cohorts is due to
temperature. So the best I could do is to show the correlation between
cohorts and temperatures and propose the link (?)

(Thank you again for the warning about model selection, I am carefully
revising and relearning the process.)

(I tried to use a subset of the data to compare the variance estimates by
classical ANOVA and mixed-effects model, but it is unfortunately too much
for my brain right now..)

Best regards,
Amy Huang

Am Sa., 3. Juli 2021 um 17:00 Uhr schrieb kalman.toth via
R-sig-mixed-models <r-sig-mixed-models at r-project.org>:

  
  
Message-ID: <CAK2iko=NKQM_zu7J0wYYPkj+gYhxfrskQBcyZXt5VvZcd036Vg@mail.gmail.com>
In-Reply-To: <boE5pTz1kRk1UEaDwrCx4OwQFHa_SshZzBoqKjWRLraYlQ6BIoIdQHf5Bnm-C1Z7Ivy-sGeIj1rP4RD59ONVIvXSVkGRz-T9xoiVgqpSXmE=@protonmail.com>