Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: ordered logistic regression with endogenous variable
From
"JVerkuilen (Gmail)" <[email protected]>
To
[email protected]
Subject
Re: st: ordered logistic regression with endogenous variable
Date
Thu, 11 Oct 2012 11:11:57 -0400
On Thu, Oct 11, 2012 at 9:48 AM, Justina Fischer <[email protected]> wrote:
> we both do not know what type of ordered variable Anat is using.
No we don't but I guess what I'm saying is that the article cited has
quite limited generality.
(When I was in grad school one of the big names in happiness research
was in my department. When papers had a lot of methodological content
he'd corral either my advisor or me, but it's been a bit since I read
anything in that area.)
> As you might have noticed, we both give the same advise, for obviously the same reason(s).
I think we happen to agree, but not for the same reasons. Using a
least squares estimator will work OK in certain circumstances, namely
when there's minimal skew and the number of categories is relatively
large and hence Gaussian error is reasonable, which is typical for 10
point satisfaction with life scales but won't be for a very coarse
scale like Disagree/Neither Agree nor Disagree/Agree where most people
Agree.
> Whether or not effect size is meaningful depends on the research question; I am sure your supervisor will agree with me on that statement.
>
> In many of my publications on happiness (LFS), observing an effect or not was the most important question, whereas magnitude played only a minor role for the referees.
It might play a minor role for referees but that seems to me to be
sloppy on their part, though I've certainly encountered the attitude
myself so I know where you're coming from. Referees say the darnedst
things sometimes, e.g., "why don't you use listwise deletion and then
run repeated measures ANOVA?"
Statistical significance can be quite beside the point and is
frequently a function of sample size as much as anything else. In a
huge survey with thousands of respondents, nearly anything will be
statistically significant and almost all goodness of fit statistics
will tell you that your model is rejected even so!
--
JVVerkuilen, PhD
*
* 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/