Re: st: qladder

[Nick Cox <njcoxstata@gmail.com>]

Say rather that an -ovtest- without rejection merely fails to find
problems. There may well be misspecification problems not detected by
-ovtest-. Try variations on -regress mpg weight- and -ovtest- in the
auto data. Here a plain linear regression looks good but you'd still
be better off working with the reciprocal of -mpg-, and even better
off with a -glm, link(power -1)-.

This is, I suggest, true of all health checks.

Nick

On Thu, Sep 22, 2011 at 2:00 PM, Carlo Lazzaro <carlo.lazzaro@tin.it> wrote:
>
> Dear Ozgur,
> I do not understand your concern.
> If the -ovtest- does not reject the null hypothesis of no omitted variables,
> your regression model has no misspecification problems (ie, all the relevant
> explanatory variables are included in the RHS).
>
>
> Kindest Regards,
> Carlo
>
> -----Messaggio originale-----
> Da: owner-statalist@hsphsun2.harvard.edu
> [mailto:owner-statalist@hsphsun2.harvard.edu] Per conto di Ozgur Ozdemir
> Inviato: giovedì 22 settembre 2011 13.57
> A: Stata
> Oggetto: RE: st: qladder
>
>
>
> Thanks Nic,
> you are right, in my case, residuals I think should be normally distributed
> and i am happy with it but when I run the ovtest, it does not reject the
> null hypothesis, that means, something is wrong with the transformations or
> interactions within the model, not sure if i really need to pass the ovtest.
> I did not see lots of papers reported ovtest results anyway.
>
>
>
>
>
> kind regards
> Ozgur
>
>
> ----------------------------------------
>> Date: Thu, 22 Sep 2011 12:52:31 +0100
>> Subject: Re: st: qladder
>> From: njcoxstata@gmail.com
>> To: statalist@hsphsun2.harvard.edu
>>
>> I don't think I can suggest a complete list for you. You need to look
>> at good publications in your field and talk to your
>> advisors/supervisors/committee about what is expected. The idea of a
>> link function is perhaps best covered in books on generalised linear
>> models which, to be frank, are from what you say likely to be a tough
>> read for you.
>>
>> That said, the purpose of -qladder- is to suggest transformations that
>> make data more nearly normal. However, it is not an assumption of
>> regression that data are normally distributed. Consider x = 1(1)10, y
>> = 2 + 3x. Here regression makes perfect sense and it is not a problem
>> that neither y nor x is normally distributed.
>>
>> Nick
>>
>> On Thu, Sep 22, 2011 at 12:42 PM, Ozgur Ozdemir
>> <ozdemirozgur@hotmail.com> wrote:
>> >
>> >
>> > Hi Nick,
>> > that is great, thanks, what kind of link functions? meanwhile, I am very
> new to Stata and just collected my data for my phd however still struggling
> with regression analysis. it seems, it will be difficult than what I
> expected. How can i find a complete list of activities to be done during a
> regression / step by step journey ? it seems,  i am finding something new
> everyday. thanks in advance.
>> >
>> >
>> >
>> >
>> >
>> >
>> >
>> > kind regards
>> > Ozgur Ozdemir
>> > T: +44 (0) 75 0332 9865
>> > E: ozdemirozgur@hotmail.com<mailto:ozdemirozgur@hotmail.com>
>> > Skype : ozgurozdemir2005
>> >
>> >
>> > ----------------------------------------
>> >> Date: Thu, 22 Sep 2011 12:38:44 +0100
>> >> Subject: Re: st: qladder
>> >> From: njcoxstata@gmail.com
>> >> To: statalist@hsphsun2.harvard.edu
>> >>
>> >> It's enough that some are real zeros for the transformations concerned
>> >> to be invalid. That doesn't rule out using link functions that indulge
>> >> zeros.
>> >>
>> >> Nick
>> >>
>> >> On Thu, Sep 22, 2011 at 12:35 PM, Ozgur Ozdemir
>> >> <ozdemirozgur@hotmail.com> wrote:
>> >> >
>> >> > Hi Nic,
>> >> > It seems you are right, i have too many zeros but some of them are
> real zeros, and some of them are missing values. i am not sure how I can
> handle it.
>> >> >
>> >> >
>> >> >
>> >> >
>> >> >> From: njcoxstata@gmail.com
>> >> >> To: statalist@hsphsun2.harvard.edu
>> >> >>
>> >> >> The implication is that you have zeros in your data so that the
>> >> >> transformations not plotted are not computable for all values. It is
>> >> >> thus not clear that you are missing anything that could be useful.
>> >> >>
>> >> >> Nick
>> >> >>
>> >> >> On Thu, Sep 22, 2011 at 12:18 PM, Ozgur Ozdemir
>> >> >> <ozdemirozgur@hotmail.com> wrote:
>> >> >>
>> >> >> > when I use the qladder command, it does show only four graphs
> including cubic, square, identity and square root.   but does not show i.e
> inverse, 1/square and some others. Is there any way that I can get others
> graphed?
>> >> >> >
>>
>> *
>> * For searches and help try:
>> * http://www.stata.com/help.cgi?search
>> * http://www.stata.com/support/statalist/faq
>> * http://www.ats.ucla.edu/stat/stata/
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>

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