```Date: Fri, 23 Sep 2005 13:32:24 -0500 Reply-To: "Peck, Jon" Sender: "SPSSX(r) Discussion" From: "Peck, Jon" Subject: Re: Regression Plot in SPSS Comments: To: Karl Koch Content-Type: text/plain; charset="iso-8859-1" There is no reason that the regression line would extend through the entire range of the dependent variable over the in-sample values of the regressors. But if your Y variable is in fact forced to lie within a bounded interval, then this violates the usual regression assumptions and you might need to use another method, because those bounds are incompatible with the normality assumption about the error term and with linearity of the regression. I can't tell whether in your case you are referring to the actual data range or a theoretical data range. This may or may not be a practical problem depending on the data, but it should be considered. Regards, Jon Peck SPSS, Inc. -----Original Message----- From: SPSSX(r) Discussion [mailto:SPSSX-L@LISTSERV.UGA.EDU] On Behalf Of Karl Koch Sent: Friday, September 23, 2005 12:48 PM To: SPSSX-L@LISTSERV.UGA.EDU Subject: Re: [SPSSX-L] Regresssion Plot in SPSS Yes, I know that. But why does is go lower then 1 even if it was never trained for that? I am just wondering and would like to know curiously-wise. Not becuase it is a major problem for me... Perpaps you have an idea? Or perhaps there is not general explaination and it depends on the model itself? Karl > --- Ursprüngliche Nachricht --- > Von: Hector Maletta > An: SPSSX-L@LISTSERV.UGA.EDU > Betreff: Re: Regresssion Plot in SPSS > Datum: Fri, 23 Sep 2005 13:52:52 -0300 > > Karl, > Since the coefficients of A and C are negative, the predicted value when A > and C are at their maximum is actually the MINIMUM predicted value, in > this > case 0.967 which is close to the minimum observed value of 1. When A and C > are at their minimum value of 0, the predicted value of Y would equal the > constant, 4.16, which is less than the maximum value of Y but within the > permissible range of 1 to 6. Other combinations would give predicted > values > between 0.967 and 4.16. > > Hector > > > > > -----Original Message----- > > From: Karl Koch [mailto:TheRanger@gmx.net] > > Sent: Friday, September 23, 2005 1:42 PM > > To: Hector Maletta; Mailinglist SPSS > > Subject: RE: Regresssion Plot in SPSS > > > > Hello all, > > > > I have now a regression model that does exclude the > > non-significant second factor (B) as it was suggested earlier > > on this discussion. To recall the earlier discussion, I had > > three factors A, B, and C. B, however turned out to be > > non-significant. The final regression (non-standarized) > > equation looks like now like this: > > > > Y = 4.16 - 0.778 A - 1.637 C > > > > The original scale of the dependent variable (Y) is > > contineous and but limited between 1.0 and 6.0. The following > > list is the range of values for A and C: > > > > A = {0,1,2} > > C = {0,1} > > > > However when using the equation with the most extreme values > > for A (=2) and C (=1) I get a result of 0.967. This value is > > lower than the minimum value of the orginal scale of the > > dependent variable. I am wondering why this is... The > > prediction should actually not go below the theoretical > > minimum of the scale especially since this was never learned > > from previous data. Is this the appearance of some sort of > > error (espsilon) ? > > > > Can somebody explain that? > > > > Kind Regards, > > Karl > > > > > > > > > --- Ursprüngliche Nachricht --- > > > Von: "Hector Maletta" > > > An: "'Karl Koch'" , > > > Betreff: RE: Regresssion Plot in SPSS > > > Datum: Wed, 21 Sep 2005 17:05:28 -0300 > > > > > > Karl, > > > As you can see from your own results, standardized > > coefficients do not > > > add up to 1. Each standardized coefficient (beta) is the expected > > > change in Y, measured in standard deviations, for each > > increase of one > > > std deviation in X. There is no reason for these > > coefficients to add up to 1. > > > The advantage of beta over b is simply that they are comparable to > > > each other, and across studies, because they are independent of the > > > unit of measurement and the distributional peculiarities of each > > > variable, since all are measured in each variable's own standard > > > deviation units. On the other hand, they are relative to the std > > > deviation and are less understandable for readers, who may > > understand > > > better the raw b coefficients (so much increase in Y for each unit > > > increase in X, both measured in their natural units). > > > > > > Hector > > > > > > > -----Original Message----- > > > > From: SPSSX(r) Discussion [mailto:SPSSX-L@LISTSERV.UGA.EDU] On > > > > Behalf Of Karl Koch > > > > Sent: Wednesday, September 21, 2005 4:40 PM > > > > To: SPSSX-L@LISTSERV.UGA.EDU > > > > Subject: Re: Regresssion Plot in SPSS > > > > > > > > The standardized coefficients would then be what I am > > looking for. I > > > > need a parameter that adds up to 1.0 which I can directly > > assign to > > > > the independent variables. > > > > > > > > Karl > > > > > > > > > --- Ursprüngliche Nachricht --- > > > > > Von: Vishal Dave > > > > > An: SPSSX-L@LISTSERV.UGA.EDU > > > > > Betreff: Re: Regresssion Plot in SPSS > > > > > Datum: Wed, 21 Sep 2005 13:53:29 -0500 > > > > > > > > > > Karl, > > > > > > > > > > Since your variables are not standardized, the coefficients > > > > won't add > > > > > up to 1.0. You have to covert the variables in > > > > standardized variables > > > > > and > > > > then > > > > > run the regression model. It will give you model with > > > > intercept = 0 > > > > > and > > > > > sum(coefficients) = 1. This will also take care of having > > > > different > > > > > non-standardized variables with different mean and SD. > > > > > > > > > > Vishal. > > > > > > > > > > > > > > > > > > > > -----Original Message----- > > > > > From: SPSSX(r) Discussion [mailto:SPSSX-L@LISTSERV.UGA.EDU] > > > > On Behalf > > > > > Of Karl Koch > > > > > Sent: Wednesday, September 21, 2005 12:48 PM > > > > > To: SPSSX-L@LISTSERV.UGA.EDU > > > > > Subject: Re: Regresssion Plot in SPSS > > > > > > > > > > So, what you are saying is I have to create first a new > > regression > > > > > model that only includes A and C. Then only use the > > unstandarized > > > > > regression coefficients. Did I understand you correctly? > > > > > > > > > > What is the difference to the standarized coefficients? Do the > > > > > standarized coefficients somehow add up to 1.0 so that > > they can be > > > > > seen as percentage of magnitiude? > > > > > > > > > > Kind Regards, > > > > > Karl > > > > > > > > > > > --- Ursprüngliche Nachricht --- > > > > > > Von: "Swank, Paul R" > > > > > > An: SPSSX-L@LISTSERV.UGA.EDU > > > > > > Betreff: Re: Regresssion Plot in SPSS > > > > > > Datum: Wed, 21 Sep 2005 12:35:19 -0500 > > > > > > > > > > > > No, you are using the unstandardized intercept but the > > > > standardized > > > > > > regression coefficients. And because B may bias the > > weights for > > > > > > A and C, > > > > > you need > > > > > > to run the regression again without B. The coefficients > > > > for Aand C > > > > > > will likely change. Then use the unstandardized > > > > coefficients for the > > > > > regression > > > > > > equation. > > > > > > > > > > > > > > > > > > Paul R. Swank, Ph.D. > > > > > > Professor, Developmental Pediatrics Director of > > Research, Center > > > > > > for Improving the Readiness > > > > of Children > > > > > > for Learning and Education (C.I.R.C.L.E.) Medical School > > > > UT Health > > > > > > Science Center at Houston > > > > > > > > > > > > -----Original Message----- > > > > > > From: Karl Koch [mailto:TheRanger@gmx.net] > > > > > > Sent: Wednesday, September 21, 2005 10:36 AM > > > > > > To: Swank, Paul R; Mailinglist SPSS > > > > > > Subject: RE: Regresssion Plot in SPSS > > > > > > > > > > > > Hello Paul, > > > > > > > > > > > > Yes, I agree. Given the fact that now I only have two > > > > paramters left > > > > > > (=being significant), is this the correct equation > > > > (keeping in mind > > > > > > the > > > > > table > > > > > > below)? > > > > > > > > > > > > Y = 4.195 - 0.319 * A - 0.442 * C + e > > > > > > > > > > > > What would be e in this case? I have standard erros for each > > > > > parameter... > > > > > > > > > > > > Kind Regards, > > > > > > Karl > > > > > > > > > > > > > > ---------------------------------------------------------------- > > > > > > Model Unstd. Coeff. Std. Coeff. t > > Sig. > > > > > > B Std. Error Beta > > > > > > > > ----------------------------------------------------------------- > > > > > > 1(Constant) 4.195 .070 > > 58.971 .000 > > > > > > A -.779 .042 -.319 > > -18.278 .000 > > > > > > B -.028 .046 -.010 > > -.618 .534 > > > > > > C -1.635 .064 -.442 > > -25.354 .000 > > > > > > > > ---------------------------------------------------------------- > > > > > > - > > > > > > > > > > > > > > > > > > > > > > > > > > Drop B from the model first as it could bias the > > > > parameters for A > > > > > > > and > > > > > C. > > > > > > > > > > > > > > > > > > > -- > > > > > > 5 GB Mailbox, 50 FreeSMS http://www.gmx.net/de/go/promail > > > > > > +++ GMX - die erste Adresse für Mail, Message, More +++ > > > > > > > > > > > > > > > > -- > > > > > 5 GB Mailbox, 50 FreeSMS http://www.gmx.net/de/go/promail > > > > > +++ GMX - die erste Adresse für Mail, Message, More +++ > > > > > > > > > > > > > -- > > > > 5 GB Mailbox, 50 FreeSMS http://www.gmx.net/de/go/promail > > > > +++ GMX - die erste Adresse für Mail, Message, More +++ > > > > > > > > -- > > > > Lust, ein paar Euro nebenbei zu verdienen? Ohne Kosten, > > ohne Risiko! > > > > Satte Provisionen für GMX Partner: > > http://www.gmx.net/de/go/partner > > > > > > > > __________ Información de NOD32 1.1228 (20050921) __________ > > > > > > > > Este mensaje ha sido analizado con NOD32 Antivirus System > > > > http://www.nod32.com > > > > > > > > > > > > > > > -- > > Lust, ein paar Euro nebenbei zu verdienen? Ohne Kosten, ohne Risiko! > > Satte Provisionen für GMX Partner: http://www.gmx.net/de/go/partner > > > > __________ Información de NOD32 1.1230 (20050922) __________ > > > > Este mensaje ha sido analizado con NOD32 Antivirus System > > http://www.nod32.com > > > > > -- 5 GB Mailbox, 50 FreeSMS http://www.gmx.net/de/go/promail +++ GMX - die erste Adresse für Mail, Message, More +++ -- GMX DSL = Maximale Leistung zum minimalen Preis! 2000 MB nur 2,99, Flatrate ab 4,99 Euro/Monat: http://www.gmx.net/de/go/dsl ```

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