Date: Wed, 1 Apr 1998 11:18:53 -0300
Sender: "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From: "Hector E. Maletta" <hmaletta@OVERNET.COM.AR>
Subject: Re: [help] the meaning of R square?
Content-Type: text/plain; charset=us-ascii
Pedro Sousa wrote:
> Sang-Gyun Lee,
> Despite my experience in social sciences is null, in biological
> experimental designs is usually difficult to obtain high values of R2.
> However, once R2 states the proportion of the total variation which is
> explained by the model fitted, it does not seem reasonable to me to accept
> any model with R2 below, say, 0.5 (half of the variation).
> Note that this does not have to do with significance of the regression
> (F test) or the significance of each variable included in the model (t
> tests). They may be significant, as seems to be the case. The significance
> depends substancially on the sample size.
> However, when you ask if some variables have to be added, there are tests
> which compare the original model (with all variables included) with the
> sub-model (with only the significant variables of the original model
> included) through a modified F-test. I think this is referenced in Draper
> and Smith (I may be mistaked). If you are interested with references
> contact me later.
> Pedro Sousa --------------------------------------------------------------
> Universidade do Algarve, U.C.T.R.A. Campus de Gambelas, 8000 FARO PORTUGAL
Adding my bit to Pedro's suggestions:
In some cases, the model is only able to explain a small portion of
total variance, but this, if it has statistical significance, may be of
value. Suppose you prove that 5% of the variation in some medical
condition is due to your selected factors: this may be useful to
introduce a new treatment for those 5% of cases, even though there are
95% caused by other factors.
However, the values mentioned by Sang-Gyun Lee seem exceedingly low.
Universidad del Salvador
Buenos Aires, Argentina