logistic regression when dependent variable is a ratio

[jimjohn]

can someone plz help me with this:

if my dependent variable is a ratio (ie the ratio of events that have a specific property, so it will only take on values between 0 and 1, then should i be using logistic regression? I know that logistic regression is to be used when your dependent variable can take on a discrete number of possible values, but in this case my dependent variable can take on any value in the continuous interval from 0 to 1. any ideas if logistic regression or just normal linear regression would work here? thanks so much!

[Hector Maletta]

Logistic regression applies with dependent variables that are discrete
EVENTS (binary or otherwise) which happen to have a logistic probability
distribution. Log Regression predicts the odds (and implicitly the
probability) of such events.
Now, your dependent variable is not a discrete event, but a continuous
variable with a restricted range (it varies continuously between 0 and 1).
It may vary within that range as a function of your predictors, and the
predictor function may be linear, logistic or whatever else your theory
dictates. The problem is, a linear function (or other functions ultimately
tending to infinity) will not restrict predicted values to the range (0, 1).
They may predict values outside that range, which may not make sense.
You may consider the possibility of regarding your dependent variable as a
probability. You may moreover adopt the hypothesis that such probability
follows a logistic curve, i.e. that it is near zero at low values of the
predictor function, and then increases towards one in the shape of an
elongated S as the predictor function increases.
Hope this helps.

Hector



-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
jimjohn
Sent: 17 June 2008 00:32
To: [hidden email]
Subject: logistic regression when dependent variable is a ratio

can someone plz help me with this:

if my dependent variable is a ratio (ie the ratio of events that have a
specific property, so it will only take on values between 0 and 1, then
should i be using logistic regression? I know that logistic regression is to
be used when your dependent variable can take on a discrete number of
possible values, but in this case my dependent variable can take on any
value in the continuous interval from 0 to 1. any ideas if logistic
regression or just normal linear regression would work here? thanks so much!
--
View this message in context:
http://www.nabble.com/logistic-regression-when-dependent-variable-is-a-ratio
-tp17877437p17877437.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[David Hitchin]

Quoting jimjohn <[hidden email]>:

> can someone plz help me with this:
>
> if my dependent variable is a ratio (ie the ratio of events that have
> a specific property, so it will only take on values between 0 and 1,
> then should i be using logistic regression?

If you know the underlying sample sizes, then you might transform your
data, e.g. suppose that for a particular combination of independent
variables the ratio 0.7 has been obtained from 100 cases, then replace
it with two new observations, one of them with 1 as the dependent and
the other with 0 as the dependent. Give the first one a weight of 70 and
the other a weight of 30.

David Hitchin

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[willers]

Good Morning,
if you can't apply David's solution, an ANOVA is the second choice, but you
can't uses this tool in any case:

a) Ratios between 0.3 and 0.7:
No problem. The variance of the ratios should be nearly constant.

b) ratio near zero (<0.1):
A logarithmic transformation stabilises the variance and avoids meaningless
values.

c) Ratios near one:
Apply he transformation y=1-x. You have case b.

d) Other: ???
Change cut off values??
Arcus-Sinus-Transformation???
Transform the ratios in a binary variable (low high)??

Good Luck!
Reinhart Willers

> Betreff: logistic regression when dependent variable is a ratio
>
> can someone plz help me with this:
>
> if my dependent variable is a ratio (ie the ratio of events that have a
> specific property, so it will only take on values between 0 and 1, then
> should i be using logistic regression? I know that logistic regression is
> to
> be used when your dependent variable can take on a discrete number of
> possible values, but in this case my dependent variable can take on any
> value in the continuous interval from 0 to 1. any ideas if logistic
> regression or just normal linear regression would work here? thanks so
> much!
> --
> View this message in context: http://www.nabble.com/logistic-regression-
> when-dependent-variable-is-a-ratio-tp17877437p17877437.html
> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Peck, Jon]

The easiest way to deal with this problem is, since you have 0 < p < 1, do the same transformation that the logistic model is doing.  Calculate y = p/(1-p).  That gives you a variable with an unrestricted range.  Then you can run ordinary regression on it.

Logistic regression is a technique that substitutes for the inability to do this transformation when you only observe 0/1 outcomes.

If you do have some 0 or 1 outcomes, you would need a little fudge factor in the transformation.

HTH,
Jon Peck

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Hector Maletta
Sent: Monday, June 16, 2008 9:59 PM
To: [hidden email]
Subject: Re: [SPSSX-L] logistic regression when dependent variable is a ratio

Logistic regression applies with dependent variables that are discrete
EVENTS (binary or otherwise) which happen to have a logistic probability
distribution. Log Regression predicts the odds (and implicitly the
probability) of such events.
Now, your dependent variable is not a discrete event, but a continuous
variable with a restricted range (it varies continuously between 0 and 1).
It may vary within that range as a function of your predictors, and the
predictor function may be linear, logistic or whatever else your theory
dictates. The problem is, a linear function (or other functions ultimately
tending to infinity) will not restrict predicted values to the range (0, 1).
They may predict values outside that range, which may not make sense.
You may consider the possibility of regarding your dependent variable as a
probability. You may moreover adopt the hypothesis that such probability
follows a logistic curve, i.e. that it is near zero at low values of the
predictor function, and then increases towards one in the shape of an
elongated S as the predictor function increases.
Hope this helps.

Hector



-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
jimjohn
Sent: 17 June 2008 00:32
To: [hidden email]
Subject: logistic regression when dependent variable is a ratio

can someone plz help me with this:

if my dependent variable is a ratio (ie the ratio of events that have a
specific property, so it will only take on values between 0 and 1, then
should i be using logistic regression? I know that logistic regression is to
be used when your dependent variable can take on a discrete number of
possible values, but in this case my dependent variable can take on any
value in the continuous interval from 0 to 1. any ideas if logistic
regression or just normal linear regression would work here? thanks so much!
--
View this message in context:
http://www.nabble.com/logistic-regression-when-dependent-variable-is-a-ratio
-tp17877437p17877437.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[willers]

Hi,
easy does not mean good. To treat small and great ratios in an equal way you
should use log-odd-transformation y=log(x/(1-x)).
In the 'normal' range 0.3-0.7 this transformation is nearly no
transformation. Near to 0 and 1 the function explodes. Regression and ANOVA
is sensitive to inhomogeneous variances
Regards
Reinhart Willers

> -----Ursprüngliche Nachricht-----
> Von: SPSSX(r) Discussion [mailto:[hidden email]] Im Auftrag von
> Peck, Jon
> Gesendet: Dienstag, 17. Juni 2008 15:14
> An: [hidden email]
> Betreff: Re: logistic regression when dependent variable is a ratio
>
> The easiest way to deal with this problem is, since you have 0 < p < 1, do
> the same transformation that the logistic model is doing.  Calculate y =
> p/(1-p).  That gives you a variable with an unrestricted range.  Then you
> can run ordinary regression on it.
>
> Logistic regression is a technique that substitutes for the inability to
> do this transformation when you only observe 0/1 outcomes.
>
> If you do have some 0 or 1 outcomes, you would need a little fudge factor
> in the transformation.
>
> HTH,
> Jon Peck
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> Hector Maletta
> Sent: Monday, June 16, 2008 9:59 PM
> To: [hidden email]
> Subject: Re: [SPSSX-L] logistic regression when dependent variable is a
> ratio
>
> Logistic regression applies with dependent variables that are discrete
> EVENTS (binary or otherwise) which happen to have a logistic probability
> distribution. Log Regression predicts the odds (and implicitly the
> probability) of such events.
> Now, your dependent variable is not a discrete event, but a continuous
> variable with a restricted range (it varies continuously between 0 and 1).
> It may vary within that range as a function of your predictors, and the
> predictor function may be linear, logistic or whatever else your theory
> dictates. The problem is, a linear function (or other functions ultimately
> tending to infinity) will not restrict predicted values to the range (0,
> 1).
> They may predict values outside that range, which may not make sense.
> You may consider the possibility of regarding your dependent variable as a
> probability. You may moreover adopt the hypothesis that such probability
> follows a logistic curve, i.e. that it is near zero at low values of the
> predictor function, and then increases towards one in the shape of an
> elongated S as the predictor function increases.
> Hope this helps.
>
> Hector
>
>
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> jimjohn
> Sent: 17 June 2008 00:32
> To: [hidden email]
> Subject: logistic regression when dependent variable is a ratio
>
> can someone plz help me with this:
>
> if my dependent variable is a ratio (ie the ratio of events that have a
> specific property, so it will only take on values between 0 and 1, then
> should i be using logistic regression? I know that logistic regression is
> to
> be used when your dependent variable can take on a discrete number of
> possible values, but in this case my dependent variable can take on any
> value in the continuous interval from 0 to 1. any ideas if logistic
> regression or just normal linear regression would work here? thanks so
> much!
> --
> View this message in context:
> http://www.nabble.com/logistic-regression-when-dependent-variable-is-a-
> ratio
> -tp17877437p17877437.html
> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Peck, Jon]

Right,  I meant to put a log in front of that transformation.

-Jon

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Reinhart Willers
Sent: Tuesday, June 17, 2008 7:56 AM
To: [hidden email]
Subject: [SPSSX-L] AW: logistic regression when dependent variable is a ratio

Hi,
easy does not mean good. To treat small and great ratios in an equal way you
should use log-odd-transformation y=log(x/(1-x)).
In the 'normal' range 0.3-0.7 this transformation is nearly no
transformation. Near to 0 and 1 the function explodes. Regression and ANOVA
is sensitive to inhomogeneous variances
Regards
Reinhart Willers

> -----Ursprüngliche Nachricht-----
> Von: SPSSX(r) Discussion [mailto:[hidden email]] Im Auftrag von
> Peck, Jon
> Gesendet: Dienstag, 17. Juni 2008 15:14
> An: [hidden email]
> Betreff: Re: logistic regression when dependent variable is a ratio
>
> The easiest way to deal with this problem is, since you have 0 < p < 1, do
> the same transformation that the logistic model is doing.  Calculate y =
> p/(1-p).  That gives you a variable with an unrestricted range.  Then you
> can run ordinary regression on it.
>
> Logistic regression is a technique that substitutes for the inability to
> do this transformation when you only observe 0/1 outcomes.
>
> If you do have some 0 or 1 outcomes, you would need a little fudge factor
> in the transformation.
>
> HTH,
> Jon Peck
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> Hector Maletta
> Sent: Monday, June 16, 2008 9:59 PM
> To: [hidden email]
> Subject: Re: [SPSSX-L] logistic regression when dependent variable is a
> ratio
>
> Logistic regression applies with dependent variables that are discrete
> EVENTS (binary or otherwise) which happen to have a logistic probability
> distribution. Log Regression predicts the odds (and implicitly the
> probability) of such events.
> Now, your dependent variable is not a discrete event, but a continuous
> variable with a restricted range (it varies continuously between 0 and 1).
> It may vary within that range as a function of your predictors, and the
> predictor function may be linear, logistic or whatever else your theory
> dictates. The problem is, a linear function (or other functions ultimately
> tending to infinity) will not restrict predicted values to the range (0,
> 1).
> They may predict values outside that range, which may not make sense.
> You may consider the possibility of regarding your dependent variable as a
> probability. You may moreover adopt the hypothesis that such probability
> follows a logistic curve, i.e. that it is near zero at low values of the
> predictor function, and then increases towards one in the shape of an
> elongated S as the predictor function increases.
> Hope this helps.
>
> Hector
>
>
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> jimjohn
> Sent: 17 June 2008 00:32
> To: [hidden email]
> Subject: logistic regression when dependent variable is a ratio
>
> can someone plz help me with this:
>
> if my dependent variable is a ratio (ie the ratio of events that have a
> specific property, so it will only take on values between 0 and 1, then
> should i be using logistic regression? I know that logistic regression is
> to
> be used when your dependent variable can take on a discrete number of
> possible values, but in this case my dependent variable can take on any
> value in the continuous interval from 0 to 1. any ideas if logistic
> regression or just normal linear regression would work here? thanks so
> much!
> --
> View this message in context:
> http://www.nabble.com/logistic-regression-when-dependent-variable-is-a-
> ratio
> -tp17877437p17877437.html
> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Bryan Tec]

Hi,

To piggyback on the ongoing logistic regression discussion I am curious if
there is any way to obtain standardized coefficients for logistic regression
so that the influence of variables can be directly compared.  Specifically I
am curious how to do this in SPSS, but if there's not a button to click or
an option to select (I haven't seen one), I would welcome any information on
how to do it manually.

Thanks,
Bryan

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Peck, Jon]

Remember that the size of the effect depends on where you evaluate it.  In the logistic model,
dp/dx = p(1-p) * beta.  Typically you would evaluate this at p-bar, where it is at its max.  But at least all the coefficients will move together, since p(1-p) captures all the interaction.

HTH,
Jon Peck

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Bryan Tec
Sent: Tuesday, June 17, 2008 8:22 AM
To: [hidden email]
Subject: [SPSSX-L] Standardized coefficients in logistic regression

Hi,

To piggyback on the ongoing logistic regression discussion I am curious if
there is any way to obtain standardized coefficients for logistic regression
so that the influence of variables can be directly compared.  Specifically I
am curious how to do this in SPSS, but if there's not a button to click or
an option to select (I haven't seen one), I would welcome any information on
how to do it manually.

Thanks,
Bryan

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[SR Millis-3]

Scott Long has written some programs in Stata to do this.

Scott R Millis, PhD, MEd, ABPP (CN,CL,RP), CStat
Professor & Director of Research
Dept of Physical Medicine & Rehabilitation
Wayne State University School of Medicine
261 Mack Blvd
Detroit, MI 48201
Email:  [hidden email]
Tel: [phone]
Fax: [phone]


--- On Tue, 6/17/08, Bryan Tec <[hidden email]> wrote:

> From: Bryan Tec <[hidden email]>
> Subject: Standardized coefficients in logistic regression
> To: [hidden email]
> Date: Tuesday, June 17, 2008, 10:21 AM
> Hi,
>
> To piggyback on the ongoing logistic regression discussion
> I am curious if
> there is any way to obtain standardized coefficients for
> logistic regression
> so that the influence of variables can be directly
> compared.  Specifically I
> am curious how to do this in SPSS, but if there's not a
> button to click or
> an option to select (I haven't seen one), I would
> welcome any information on
> how to do it manually.
>
> Thanks,
> Bryan
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body
> text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the
> command
> INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Gary Rosin]

At 10:31 PM 6/16/2008, jimjohn wrote:
>if my dependent variable is a ratio (ie the ratio of events that have a
>specific property, so it will only take on values between 0 and 1, then
>should i be using logistic regression? I know that logistic regression is to
>be used when your dependent variable can take on a discrete number of
>possible values, but in this case my dependent variable can take on any
>value in the continuous interval from 0 to 1. any ideas if logistic
>regression or just normal linear regression would work here? thanks so much!

I tried using the logit transformation that Jon Peck spoke of, together with
a linear regression.  The problem is that the residuals are not normally
distributed.  Using a stepwise regression, with the variables weighted by
1/variance (1/(p)(1-p)) helped.  You use the sample proportions as the
initial weights, calculate the predicted proportions, re-weight and run another
regression, and another, until the model converges.  Somewhere, I have a
macro or Python script that does that.

If you have v. 15 or v. 16, and the Advance Regression module, the
Generalized Linear Model, with a binary distribution, a logit transformation,
and an events/trials variables works well.  You can even adjust for
overdispersion.

That about exhausts my limited resources on this.

Gary


     ---

Prof. Gary S. Rosin              Internet:  [hidden email]
South Texas College of Law
1303 San Jacinto                   Voice:  [phone]
Houston, TX  77002-7000           Fax:  [phone]

SSRN:  http://ssrn.com/Author=15115

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[jimjohn]

thanks a lot for the replies guys! just wondering, lets say I just conducted a normal linear regression with my dependent variable being this ratio between0 and 1. I know that the main problem with doing this is that the regression equation could sometimes provide results not in this interval. but these would only be extreme cases where I plug in values of the independent variable that really shouldn't happen, right? If I'm just using this regression to forecast into the future, my regression equation should still provide valid results for my ratio dependent variable. do you guys agree? just wanted to see what you think. thanks!

jimjohn wrote

can someone plz help me with this:

if my dependent variable is a ratio (ie the ratio of events that have a specific property, so it will only take on values between 0 and 1, then should i be using logistic regression? I know that logistic regression is to be used when your dependent variable can take on a discrete number of possible values, but in this case my dependent variable can take on any value in the continuous interval from 0 to 1. any ideas if logistic regression or just normal linear regression would work here? thanks so much!

[Hector Maletta]

Not necessarily. The relationship of predictors to a probability is
frequently non linear (passing from 0.4 to 0.5 may be "easier", i.e. it may
require lower increments in predictors, than passing from 0.89 to 0.99).
In certain cases it is perfectly possible that "admissible" values of
predictors produce "impossible" values in the outcome.
The whole idea of log regression is that predictors are not linearly related
to the probability of the outcome. Supposing only one predictor positively
related to a binary outcome, the probability of the outcome would start at
zero for a low value of the predictor, would remain very low as the
predictor grows, up to a certain point when the probability starts growing
faster, then finally the probability gradually flats out at rather high
values of the predictor.

Hector


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
jimjohn
Sent: 20 June 2008 11:45
To: [hidden email]
Subject: Re: logistic regression when dependent variable is a ratio

thanks a lot for the replies guys! just wondering, lets say I just conducted
a normal linear regression with my dependent variable being this ratio
between0 and 1. I know that the main problem with doing this is that the
regression equation could sometimes provide results not in this interval.
but these would only be extreme cases where I plug in values of the
independent variable that really shouldn't happen, right? If I'm just using
this regression to forecast into the future, my regression equation should
still provide valid results for my ratio dependent variable. do you guys
agree? just wanted to see what you think. thanks!



jimjohn wrote:

>
> can someone plz help me with this:
>
> if my dependent variable is a ratio (ie the ratio of events that have a
> specific property, so it will only take on values between 0 and 1, then
> should i be using logistic regression? I know that logistic regression is
> to be used when your dependent variable can take on a discrete number of
> possible values, but in this case my dependent variable can take on any
> value in the continuous interval from 0 to 1. any ideas if logistic
> regression or just normal linear regression would work here? thanks so
> much!
>

--
View this message in context:
http://www.nabble.com/logistic-regression-when-dependent-variable-is-a-ratio
-tp17877437p18031087.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[jimjohn]

thanks again for all the answers guys! i just have a follow up. so my variable p is in the range from 0 to 1.
So, as suggested, I plan to do the following transformation:
Y = ln ((p / (1-p))
now I will run normal linear regression with Y as my dependent variable, and thus the range of my dependent variable will not just be from 0 to 1 but the range will not include all possible values.
Now, once I have done this, I would have to convert Y back to P
the equation I will get from regression is Y = b0 + b1X ..., so once i get the values of b0 and b1, I can solve for p by:
p = [exp (b0 + b1X)]  /  [ 1 + exp (b0 + b1X)]
is this correct?

Can I still interpret the Adjusted R^2 and other regression outputs the same way as normal? ie if my linear regression on Y has an Adjusted R^2 of .6, I can still say that 60% of the variance in p is explained by this model?
thanks a lot, plz let me know if theres anything im misunderstanding here.

Peck, Jon wrote

Remember that the size of the effect depends on where you evaluate it.  In the logistic model,
dp/dx = p(1-p) * beta.  Typically you would evaluate this at p-bar, where it is at its max.  But at least all the coefficients will move together, since p(1-p) captures all the interaction.

HTH,
Jon Peck

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[e-mail]] On Behalf Of Bryan Tec
Sent: Tuesday, June 17, 2008 8:22 AM
To: [e-mail]
Subject: [SPSSX-L] Standardized coefficients in logistic regression

Hi,

To piggyback on the ongoing logistic regression discussion I am curious if
there is any way to obtain standardized coefficients for logistic regression
so that the influence of variables can be directly compared.  Specifically I
am curious how to do this in SPSS, but if there's not a button to click or
an option to select (I haven't seen one), I would welcome any information on
how to do it manually.

Thanks,
Bryan

=====================
To manage your subscription to SPSSX-L, send a message to
[e-mail] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[e-mail] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[jimjohn]

sorry i meant to put my last msg under the "Logistic Regression when dependent variable is a ratio" post.
-------
thanks again for all the answers guys! i just have a follow up. so my variable p is in the range from 0 to 1.
So, as suggested, I plan to do the following transformation:
Y = ln ((p / (1-p))
now I will run normal linear regression with Y as my dependent variable, and thus the range of my dependent variable will not just be from 0 to 1 but the range will not include all possible values.
Now, once I have done this, I would have to convert Y back to P
the equation I will get from regression is Y = b0 + b1X ..., so once i get the values of b0 and b1, I can solve for p by:
p = [exp (b0 + b1X)]  /  [ 1 + exp (b0 + b1X)]
is this correct?

Can I still interpret the Adjusted R^2 and other regression outputs the same way as normal? ie if my linear regression on Y has an Adjusted R^2 of .6, I can still say that 60% of the variance in p is explained by this model?
thanks a lot, plz let me know if theres anything im misunderstanding here.

jimjohn wrote

can someone plz help me with this:

if my dependent variable is a ratio (ie the ratio of events that have a specific property, so it will only take on values between 0 and 1, then should i be using logistic regression? I know that logistic regression is to be used when your dependent variable can take on a discrete number of possible values, but in this case my dependent variable can take on any value in the continuous interval from 0 to 1. any ideas if logistic regression or just normal linear regression would work here? thanks so much!