skewed and zero-inflated data

[Nabaneeta Saha]

Hi listers,

I have a dataset with one binary dependable variable and 9 independent
variables which are highly skewed and zero-inflated. Could anyone suggest
me a right regression method (or any other method) to analyze the data?

Thanks,

Nabaneeta

=====================
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

[Maguin, Eugene]

Nabaneeta,

What matters most is whether your dependent is highly skewed and
zero-inflated. Is it? Also, please quantify both characterizations so that
we understand the variable in the same way that you do.

Gene Maguin



-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Nabaneeta Saha
Sent: Wednesday, November 03, 2010 8:30 AM
To: [hidden email]
Subject: skewed and zero-inflated data

Hi listers,

I have a dataset with one binary dependable variable and 9 independent
variables which are highly skewed and zero-inflated. Could anyone suggest
me a right regression method (or any other method) to analyze the data?

Thanks,

Nabaneeta

=====================
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

[Bruce Weaver]

Nabaneeta Saha wrote

Hi listers,

I have a dataset with one binary dependable variable and 9 independent
variables which are highly skewed and zero-inflated. Could anyone suggest
me a right regression method (or any other method) to analyze the data?

Thanks,

Nabaneeta

With a binary dependent variable, the most common choice of model would be binary logistic regression.  That model has no distributional assumptions for explanatory variables.  With 9 explanatory variables, I think your biggest concern is whether your sample size  is large enough.  For binary logistic regression, you need at least 10 events per model parameter, and 15 or 20 events per parameter would be better.  What I mean by "event" is the less frequent of the two possible values for the dependent variable.  For more on this, see Mike Babyak's nice article, or Frank Harrell's book on regression models.

   http://www.class.uidaho.edu/psy586/Course%20Readings/Babyak_04.pdf

HTH.