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 2000)Back to main SPSSX-L pageJoin or leave SPSSX-L (or change settings)ReplyPost a new messageSearchProportional fontNon-proportional font
Date:         Mon, 31 Jan 2000 08:37:51 -0600
Reply-To:     Ian.Straus@VIAINFO.NET
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Ian Straus <Ian.Straus@VIAINFO.NET>
Subject:      Re: logistic regression and levels of measurement
Comments: To: lianli@NET1.HKBU.EDU.HK
Content-Type: text/plain

You might try graphing / scatter plotting your ages vs. the residuals from a model before age is entered / a model without age. Or even, for a first cut, graphing age vs. your dependent variable. This often clarifies regression projects.

It's possible that for your dependent variable, whatever that may be, the relationship with age is not linear but is still significantly dependent on age: For instance, bicycle use is low at age 4, high at age 12, very low at age 70. That would yield a regression line with a beta but a bad fit, unless you got into curve fitting and added age^2 as a dependent variable. So both your models may be telling you a piece of the truth.

Ian Straus VIA Metropolitan Transit San Antonio, Texas, USA > -----Original Message----- > From: Lianjiang Li [SMTP:lianli@NET1.HKBU.EDU.HK] > Sent: Sunday, January 30, 2000 6:38 AM > To: SPSSX-L@LISTSERV.UGA.EDU > Subject: Fw: logistic regression and levels of measurement > > When I entered Age as a continuous variable into a multivariate > logisitic > regression model (response was dichotomous), the result showed it had > very > small beta and was insignificant (p = .35). But when I recoded Age > into a > five-point ordinal variable and included it in the same model, it had > a much > larger beta and was highly significant (p = .002). I assume that the > change > of betas was due to the change of unit of change of the predictor, but > I'd > like to be reassured that the significance level change was due to the > same > thing. > > More puzzling is that when I did the same thing with ordinal logit > analysis > with GOLDMineR, I observed similar change of betas, but the > significance > levels of the predictor Age remained largely unchanged regardless of > its > levels of measurement. > > Any advice would be greatly appreciated. > > Lianjiang Li > Hong Kong Baptist University


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