Quantifying Variables in model

[mawais31]

Dear All,

I have created a model for prediction, I have seen that most of the variables are significant, my model contains counts for categorical variables.

What my question is that is there any methods to find that how much important particular variables is? e.x. say 19% weather effect, 5% mileage, etc.

Thanks

Regards
Awais
 

[Maguin, Eugene]

Awais,
Did you create the model for a continuous DV using regression or for a dichotomous DV using logistic regression or for an ordinal DV using Plum or Genlin or for a nominal DV using Genlin or Nomreg?

Gene Maguin

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of mawais31
Sent: Friday, August 31, 2012 5:12 AM
To: [hidden email]
Subject: Quantifying Variables in model

Dear All,

I have created a model for prediction, I have seen that most of the variables are significant, my model contains counts for categorical variables.

What my question is that is there any methods to find that how much important particular variables is? e.x. say 19% weather effect, 5% mileage, etc.

Thanks

Regards
Awais




--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Quantifying-Variables-in-model-tp5714917.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

[mawais31]

I have created a model and my response variable is number of counts, so I am using Poisson regression with lasso penalty.

By creating a model will only specify the variables which are significant i.e. looking at P value. But how shall I find that say?
Weather has 20% effect
mileage has 10% for example

From where in the output of a model I interpret these terms...

[David Marso]

I have no idea what this means!
"Weather has 20% effect
mileage has 10% for example "...

mawais31 wrote

I have created a model and my response variable is number of counts, so I am using Poisson regression with lasso penalty.

By creating a model will only specify the variables which are significant i.e. looking at P value. But how shall I find that say?
Weather has 20% effect
mileage has 10% for example

From where in the output of a model I interpret these terms...

[mawais31]

I mean to say that in response variable, out of 100. weather variables has a high effect and I shall find how much effect it has in changing response variable, say mileage variables has 10% effect on output variable, and so on...

and I can find other variables effects also which are mentioned in model..

[Albert-Jan Roskam]

 Maybe the proportion of explained variance (R2) by the predictors weather, mileage? Curious what the dependent variable is. Number of accidents?

Regards,
Albert-Jan

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
All right, but apart from the sanitation, the medicine, education, wine, public order, irrigation, roads, a
fresh water system, and public health, what have the Romans ever done for us?
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 

---

From: David Marso <[hidden email]>
To: [hidden email]
Sent: Saturday, September 1, 2012 11:07 AM
Subject: Re: [SPSSX-L] Quantifying Variables in model

I have no idea what this means!
"Weather has 20% effect
mileage has 10% for example "...


mawais31 wrote

>
> I have created a model and my response variable is number of counts, so I
> am using Poisson regression with lasso penalty.
>
> By creating a model will only specify the variables which are significant
> i.e. looking at P value. But how shall I find that say?
> Weather has 20% effect
> mileage has 10% for example
>
> From where in the output of a model I interpret these terms...
>

-----
Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email me.
--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Quantifying-Variables-in-model-tp5714917p5714934.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

[mawais31]

yes, Albert-Jan you are right my response variable is number of accidents per day,

Thanks

[Rich Ulrich]

I think you will be better served if you forget the stuff
about percent of the variance.  That is really an awkward
and ineffective version of "effect size" in most cases, and
especially (I think) for this.

Take a baseline for days with everything favorable.  Call that
100, or use the actual numbers.
Then report the increased rate observed during inclement weather,
etc., alone and in combinations.  If you have several predictors, you
can report the results of the fitted equation by plugging in particular
values for the predictors.

You sometimes see lines using percentage-of-variance, like saying that
"personality is one-third genetic and two-thirds environment".  These
statements are presumptive in considering *some* particular aspect
of personality...  which the reader is expected to intuit...  and the
results only apply well to samples that have the *same* distribution
of environmental exposures and genetic loadings.  Your data on
accidents will have similar limits -- not that "accident" is so fuzzy, but
there are differences between types of roads, regional variations in
how rain typically falls, and so on. 

Your description should pay attention to the actual units that *can* be
measured, and don't increase the imprecision by glossing with "percent
of variation" attributed to an abstract label.

--
Rich Ulrich

> Date: Sat, 1 Sep 2012 12:35:12 -0700
> From: [hidden email]
> Subject: Re: Quantifying Variables in model
> To: [hidden email]
>
> yes, Albert-Jan you are right my response variable is number of accidents per
> day,
>
> Thanks
> ...