LISTSERV at the University of Georgia
Menubar Imagemap
Home Browse Manage Request Manuals Register
Previous messageNext messagePrevious in topicNext in topicPrevious by same authorNext by same authorPrevious page (January 2005)Back to main SPSSX-L pageJoin or leave SPSSX-L (or change settings)ReplyPost a new messageSearchProportional fontNon-proportional font
Date:         Mon, 31 Jan 2005 11:05:18 -0500
Reply-To:     Art@DrKendall.org
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Art Kendall <Art@DRKENDALL.ORG>
Organization: Social Research Consultants
Subject:      Re: linear regression
Comments: To: Josh Johnson <joshjohnson88990@yahoo.co.uk>
In-Reply-To:  <20050131143720.46861.qmail@web86908.mail.ukl.yahoo.com>
Content-Type: text/plain; charset=windows-1252; format=flowed

A more specific discussion would depend on what kinds of information the analysis is designed to produce, who is going to use the information, how the set of cases was gathered, whether there are sufficient cases to do more complex explorations, how many aspects of satisfaction there are on the explained/criterion dependent side of the analysis, etc.

The zero order correlations/regressions are certainly part of a complete exploration. This is called "ignoring" other effects. What about controlling for ("eliminating") other effects. If there is sufficient data, and if it is of sufficient quality, more complex explorations (e.g., multiple regressions, clustering, etc.) can help find out if there are interactions of predictors with themselves (bends in the regression line) or with other variables (non-parallel regression lines). A hammer is not the only tool in the tool box. If you have the data, why not do a more thorough exploration? I presume it cost something to gather the data, why not take a more complete look at it using other tools?

Why would one think satisfaction only has simple relations to these aspects? The whole field of conjoint analysis exists because preference (a pretty close concept to satisfaction) is based on trade off of price, qualities, etc.

Simple explanations (models, schemata) are to be preferred when there are small differences in how well the explanations work, but explanations can be oversimplified. Occham's razor can shave too closely. You will never know if more complex explanations add information if you do not look. Are there subgroups of cases? E.g., is one model explanatory for men and another for women?

Although the internal part of the analysis may be complex, the presentation of the conclusions is not necessarily complex.

Hope this helps.

Art Art@DrKendall.org Social Research Consultants University Park, MD USA (301) 864-5570

Josh Johnson wrote:

>Hi all, > > > >I’d like some advice on a non-SPSS (but still statistical) issue. > >Lately I have encountered a recurring difficulty of convincing colleagues to use multiple regression rather than separate simple linear regression runs. > >The background is fairly simple- they are researching what drives satisfaction of a certain product based on different (around 18) aspects such as service, price etc… > >Their main reasoning is in simplicity- for example no predictors missing from the final model that on one hand are significantly correlated to the dependent variable but nevertheless are missing due to another ‘stronger’ predictor (very difficult explaining to a client). > > > >If our objective is not perfectly modelling and forecasting but rather finding the main drivers of satisfaction amongst a group of 18, is there any justification to run multiple regression? > > > >Many thanks, > >JJ > > > > >--------------------------------- > ALL-NEW Yahoo! Messenger - all new features - even more fun! > > > >


Back to: Top of message | Previous page | Main SPSSX-L page