[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: logistic tranformation, proportion variables

From   Marck Bulter <[email protected]>
To   [email protected]
Subject   Re: st: logistic tranformation, proportion variables
Date   Fri, 14 Dec 2007 01:08:14 +0100

Maarten buis wrote:
--- Marck Bulter <[email protected]> wrote:

Do you think that the conversion to 0.001 is appropriate?
The problem here is to choose an appropriately small value for zero. If
you choose a value that is too small, the logit transformation will
result in excessively negative values, leading to outliers. On the
other hand, you certainly don't want this value to be larger than the
smallest non-zero value in your data. So this will require some
experimatation and the end result will to some extend remain
arbitrairy. (For inspection of normality of the residuals I
wholehearedly recommend the -hangroot- package downloadable from ssc,
though I admit that I am not completely unbiased)

And more important, is it appropriate to use logistic transformed
variables both as dependent and independent variables?
There is no reason why did would be illegal. I would two things to make
my choice: 1) have a look at some scatter plots of logit(y) against x,
and logit(y) against logit(x), 2) estimate a model with logit(x) and a
model with x and inspect the residuals.
Hope this helps,

Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands

visiting address:
Buitenveldertselaan 3 (Metropolitan), room Z434

+31 20 5986715

Sent from Yahoo! Mail - a smarter inbox

* For searches and help try:
Dear Maarten,

In a previous email that I replied a moment ago, I tried the conversions but the results are everything but positive, if you have a minute have a look.

In short:

- y vs x, resid's are terrible
- logit y vs x , yields residuals that are appalling,
- logit y vs logit x, gives a situation with almost homoskedastic and almost normal resid's. Therefore I am some what confident to give this model a GO. But still the 0 values are problematic.

Thanks for suggestion for hangroot, looks good.

Bedankt Maarten

Marck Bulter
EUR, financial economics

* For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index