 |
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: linear probability model (LPM)
My experience is that even the most
sensible & intelligent & non-statistician
is likely to be defeated by odds ratios.
Time & again, I have known respected colleagues
present them as though they are risk ratios,
& they seem quite surprised when I point out the
differences.
The confusion is not helped by the fact that when
odds ratios are taught, somehow horse racing tends
to come up. This is not only irrelevant and a complete turn-off
for most audiences, but actually wrong. "Odds" in horse
racing are not based on the expected rate of success & failure,
but on the money to be paid out & received depending on the result.
If "risks" were given instead, it would be obvious (just by adding them
up) that the bookmaker hopes & plans to pay out a good deal less
than he gets in - and that would not be good for business.
The best explanation I ever heard from a non-statistician
was that odds ratios are "optimistic" risk ratios. This is far
more useful. It draws attention to the similarity but also to
the difference & in particular to the direction of the difference
(away from "no effect of the exposure").
Date: Sun, 13 Jan 2008 18:04:55 +0000 (GMT)
From: Maarten buis <[email protected]>
Subject: Re: st: linear probability model (LPM)
- --- Steven Samuels <[email protected]> wrote:
> A linear probability model is desirable because effects are risk
> differences, which are much easier to interpret than odds ratios.
I disagree, both measures are perfectly understandable. With odds
ratios you are representing chance by at odds instead of probability.
Both are easy to understand: odds give you the expected number of
successes for every failure, while probability gives you the expected
proportion of successes. With odds ratios groups are compared by
calculating the ratio: e.g. the odds of success for men is 10% higher
than the odds of succes for women. With risk difference you are
comparing the groups by computing differences.
- -- Maarten
- -----------------------------------------
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
http://home.fsw.vu.nl/m.buis/
- -----------------------------------------
__________________________________________________________
Sent from Yahoo! Mail - a smarter inbox http://uk.mail.yahoo.com
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
------------------------------
Date: Fri, 11 Jan 2008 22:50:15 -0800
From: Nils Kok <[email protected]>
Subject: st: color scheme cfd graph
I use the lean2 scheme for my graphs, but this does not work when plotting a
cdf graph (it appears in colors and with a different font). How can I change
the color scheme?
Nils Kok.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
------------------------------
Date: Sun, 13 Jan 2008 09:04:40 +0100
From: "roland andersson" <[email protected]>
Subject: Re: st: to compare regression coefficient across sub-samples.
I am not a statistician but as noone has answered yet I can give my 2
cents just to activate the question. What you want to do is to look
for the interaction term between gender and x1 and so forth. Maybe you
can include gender with interaction terms in the model instead of
using by?
Greetings
Roland Andersson
2008/1/11, Jinkuk Hong <[email protected]>:
> Hi,
> fist time posting (I'm a newbie in Stata in the early stage of
learning curve).
> I have a question about comparing regression coefficients across
subsamples.
> In specific, I've run multilevel regression using xtmixed for men and women
> separately, as follows:
> by gender, sort: xtmixed y x1 x2 x3 || id: x1 x2, cov (unstr) ||
day: cov (unstr)
> and wanted to test if coefficients for x1, x2, and x3 are
different between
> the two groups.
> Is there any way to do this in Stata? I've checked "test" and "estimates"
> commands, but couldn't find any commands for that.
> Any help or advice will be greatly appreciated.
> Have a nice weekend,
>
> Jin
>
> --
> *
> * For searches and help try:
> * http://www.stata.com/support/faqs/res/findit.html
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
------------------------------
Date: Sat, 12 Jan 2008 15:06:51 -0600
From: "Suryadipta Roy" <[email protected]>
Subject: Re: st: Sample selection?
Maarten,
Your comments are very helpful indeed!
Thanks,
Suryadipta.
On 1/12/08, Maarten buis <[email protected]> wrote:
> --- Suryadipta Roy <[email protected]> wrote:
> > My dependent variable is continuous taking negatitve to positive
> > values. In order to convert it into logarithms, I used a positive
> > transformation and then ran the regression with the logarithm value
> > as the dependent varible. I am finding that the values of the
> > coefficients of my explanatory variables are different from the one
> > when the original value was used as the dependent variable. I was
> > wondering if I need to address sample selection issues when I am
> > using the logarithm and thereby constraining the variable to be
> > positive only, and why the values of the coefficients differ in the
> > two cases.
>
> No, say your original depedent variable is in euros, than your
> coefficient tells you the expected increase or decrease in euros for a
> unit increase in your dependent variable. In your log transformed model
> you get the expected increase or decrease in log(euros) for a unit
> increase in your depedent variable. So the coefficients are different
> because they mean something different.
>
> > Also, my main explanatory variable is categorical in nature ( 0 -
> > 10). Is it alright to treat this as a continuous variable in the
> > regression?
>
> Depends, if it is truely categorical (e.g christian, muslim, hindu,
> etc.) than you can add it as a continuous variable (in the sense that
> Stata won't complain) but the outcome is complete rubbish. If it is
> oridinal than you are assuming that the distance between the categories
> are the same. I you think (and can convince others) that that is ok,
> than you can do it.
>
> Hope this helps,
> Maarten
>
> -----------------------------------------
> 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
>
> http://home.fsw.vu.nl/m.buis/
> -----------------------------------------
>
>
> ___________________________________________________________
> Yahoo! Answers - Got a question? Someone out there knows the answer. Try it
> now.
> http://uk.answers.yahoo.com/
> *
> * For searches and help try:
> * http://www.stata.com/support/faqs/res/findit.html
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
------------------------------
Date: Sun, 13 Jan 2008 19:39:59 +0000 (GMT)
From: Maarten buis <[email protected]>
Subject: Re: st: color scheme cfd graph
What exact command are you using to create a cdf graph?
- --- Nils Kok <[email protected]> wrote:
> I use the lean2 scheme for my graphs, but this does not work when
> plotting a cdf graph (it appears in colors and with a different
> font). How can I change the color scheme?
- -----------------------------------------
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
http://home.fsw.vu.nl/m.buis/
- -----------------------------------------
___________________________________________________________
Yahoo! Answers - Got a question? Someone out there knows the answer. Try it
now.
http://uk.answers.yahoo.com/
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
------------------------------
Date: Sun, 13 Jan 2008 18:05:02 -0500
From: "Tom Trikalinos" <[email protected]>
Subject: Re: st: linear probability model (LPM)
Maarten,
I kind of disagree back - from a specific angle... The odds ratio is
something I personally -and others- have difficulty grasping, in terms
of its natural meaning. Risk ratios/ratios of probabilities on the
contrary are considerably easier to get... Under known circumstances
odds ratios and risk ratios are numerically very similar ... all too
often they are not... and then this misinterpretation bites... I
think that this is what Steven meant...
tom
On Jan 13, 2008 1:04 PM, Maarten buis <[email protected]> wrote:
> --- Steven Samuels <[email protected]> wrote:
> > A linear probability model is desirable because effects are risk
> > differences, which are much easier to interpret than odds ratios.
>
> I disagree, both measures are perfectly understandable. With odds
> ratios you are representing chance by at odds instead of probability.
> Both are easy to understand: odds give you the expected number of
> successes for every failure, while probability gives you the expected
> proportion of successes. With odds ratios groups are compared by
> calculating the ratio: e.g. the odds of success for men is 10% higher
> than the odds of succes for women. With risk difference you are
> comparing the groups by computing differences.
>
> -- Maarten
>
> -----------------------------------------
> 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
>
> http://home.fsw.vu.nl/m.buis/
> -----------------------------------------
>
>
> __________________________________________________________
> Sent from Yahoo! Mail - a smarter inbox http://uk.mail.yahoo.com
>
>
> *
> * For searches and help try:
> * http://www.stata.com/support/faqs/res/findit.html
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
------------------------------
End of statalist-digest V4 #2941
********************************
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
==========================
Paul T Seed MSc CStat
Senior Lecturer in Medical Statistics
King's College London
Division of Reproduction and Endocrinology
St Thomas' Hospital,
Lambeth Palace Road,
London SE1 7EH
tel (+44) (0) 20 7188 3642
fax (+44) (0) 20 7620 1227
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/