Re: st: random quadratic effect without random linear effect?

[Maarten Buis <maartenlbuis@gmail.com>]

On Thu, Jun 20, 2013 at 12:07 AM, Jordan Silberman wrote:
> For a model run in xtmixed, we all know that if you include a fixed
> quadratic effect of predictor A, then you typically should include the
> lower order fixed linear effect of A as well.
>
> Does an analogous rule apply to random effects? I think that it may
> make theoretical sense to include only a random quadratic effect,
> without the corresponding random linear effect. Can anyone tell me
> whether or not this would be "kosher?"

That obviously depends on your argument. As long as your can give a
reasonable argument for your choices everything is allowed. However,
what you propose is unusual, so you'd better have a good argument
because when you try to publish this someone is guaranteed to ask you
about it. Moreover, I find it hard to imagine an argument that could
support such a model, so I suspect that there is an error in your
argument. But this is just a guess based on very limited information.
This is all we can say, since you did not give us your argument.

-- Maarten

---------------------------------
Maarten L. Buis
WZB
Reichpietschufer 50
10785 Berlin
Germany

http://www.maartenbuis.nl
---------------------------------
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/