Re: st: marginsplot and transformed dependent variable

[Nick Cox <njcoxstata@gmail.com>]

As the Statalist FAQ advises, this is an interdisciplinary list. I
know CV as curriculum vitae and coefficient of variation, which both
sound wrong here. If I guess harder, you mean cardiovascular. It is
best to assume ignorance of your field and limited competence in Stata
in those likely to reply.

More importantly, this is difficult to comment on precisely. The
flavour is too much like "I tried something, but the results look
wrong" to which the answer is too much like "Well, it probably is
wrong". You show no exact code and no exact results.

Try a generalised linear model using -glm-. That way, you try to get
the best of both worlds by fitting on a transformed scale, but getting
predictions in terms of your response or outcome variable.

Nick
njcoxstata@gmail.com


On 25 April 2013 06:11, Tom Robinson <tomrobnz@gmail.com> wrote:

> Hi, I am wanting to look at the effect of body mass index on CV risk
> factors (HbA1c, BP, lipids) in people with T2 diabetes.
>
> I am using linear regression to control for demographic variables and
> have categorized BMI. I would like to use margins and marginsplot to
> present the results.
>
> Mt problem is that the models that include the dependent variables (the CV
> risk factors) untransformed have poor post-estimation diagnostics i.e.
> non-normal residuals, heteroskedasticity, and poor fit
>
> If I transform the dependent variables (log or inverse) the models are much
> better.  But then it is difficult to present the results to a general
> readership.  Is it possible to take the transformed margins estimates and
> un-transform them?  When I do it by hand they don't look right so maybe
> this doesn't make sense.
>
*
*   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/