Regresssion Question - Basic

[cbjs]

Hello, List.

I have customer satisfaction survey data from 96 different entities, all in the same industry, all using exactly the same survey form.  There is one overall satisfaction question (DV) and 23 satisfaction factors (IVs).  I need to create a separate report for each entity with a very basic description of the "key satisfaction drivers" using regression analysis.  The report I'm trying to replicate shows an R Squared and an R Square Change for each factor and uses the R Square Change to select the "key drivers."  It seems to me, however, that I should use multiple regression analysis and base the "key drivers" decision on the Sig T.  That is, R Squared and an R Square Change tell me how significant the model is, but does not tell me the individual "key drivers."

In case you can’t tell, I don't have a lot of experience with regression analysis.  Any thoughts/suggestions would be greatly appreciated.

Steve

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

[Stephen Hampshire]

Hi Steve,

I assume from your desciption that the "R square change" is essentially a part or semipartial correlation, in other words the change in R square you get by adding that IV to a multiple regression? That is a fair, and stringent, test of an IV's impact, though the standardised regression coefficients are more often used.

However, I would be very anxious about using multiple regression to pick key drivers in this way. Customer satisfaction surveys are notoriously prone to heavy multicollinearity, and the result is often very unstable regression estimates.

There is a whole literature on relative importance analysis, much of it relating to customer satisfaction measurement. Perhaps the easiest of these measures to use in practice is the product of the zero-order correlation and the standardised regression coefficient for each IV.

However, a much more robust method would be to first look at factor analysis or some other data reduction technique to clear up some of the collinearity, before moving to regression analysis.

Hope that helps,

Stephen

> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]]On Behalf Of
> [hidden email]
> Sent: 21 January 2008 23:40
> To: [hidden email]
> Subject: Regresssion Question - Basic
>
>
> Hello, List.
>
> I have customer satisfaction survey data from 96 different
> entities, all in the same industry, all using exactly the same
> survey form.  There is one overall satisfaction question (DV) and
> 23 satisfaction factors (IVs).  I need to create a separate
> report for each entity with a very basic description of the "key
> satisfaction drivers" using regression analysis.  The report I'm
> trying to replicate shows an R Squared and an R Square Change for
> each factor and uses the R Square Change to select the "key
> drivers."  It seems to me, however, that I should use multiple
> regression analysis and base the "key drivers" decision on the
> Sig T.  That is, R Squared and an R Square Change tell me how
> significant the model is, but does not tell me the individual
> "key drivers."
>
> In case you can’t tell, I don't have a lot of experience with
> regression analysis.  Any thoughts/suggestions would be greatly
> appreciated.
>
> Steve
>
> =====================
> 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