Re: Sum of squared semi-partial correlations

Liesl wrote:
I was under the same impression as Shaun about squared semi-partial
correlations and their relationship to R squared.

Could someone please further clarify things here.

Thanks,
Liesl


SDW wrote:

Well - yes and no :)

From the text I primarily use (Tabachnick & Fiddel) I was under the
impression that the squared semi-partial correlation is the unique
variance in a DV attributable a single IV. So that the sum of all
squared semi-partial correlation PLUS the shared variance, equals R
square. Thus my concern over the results. From your response I take it
it isn't as simple as all that.

Should I be unconcerned about my results?


1. The squared multiple correlation of a set of variables with a
d.v. equals the sum of squares of the Pearson correlations of those
variables with the d.v. only if all the variables are mutually
unorrelated.

2. The semi-partial correlation of a predictor with the d.v. is the
ordinary Pearson correlation of the d.v. with the so-called "unique
part" of the predictor, where the "unique part" of a predictor is
defined as the residual when that predictor is regressed on the other
predictors. The unique part of a predictor is uncorrelated with all
the other predictors.

3. However, when the unique parts of all the predictors are
considered simultaneously, they are generally correlated with one
another. The only case in which the unique parts are all mutually
uncorrelated is when the original predictors are all mutually
uncorrelated. Hence, the squared multiple correlation equals the sum
of squared semi-partials only if all the predictors are mutually
uncorrelated.

.

Relevant Pages

  • Re: Ordinal ratings and relationship to continuous measurement variables
    ... image generator. ... A rater rates the severity of the lines/fuzziness over the square ... The depth of the fuzziness is then measured from the top of the ... is a "correlation". ...
    (sci.stat.math)
  • Re: Adjusted R2 (R Bar Squared)
    ... Correlation can be negative and can take the values anywhere from -1 ... it’s square also by definition cannot be greater ... Adjusted r-square and r are related. ... into account the strength of the correlation and the sample size. ...
    (sci.stat.math)
  • Re: Sum and difference of independent variables
    ... randomly spattered on a 45 degree tilted square. ... Y-X was independent of Y+X, just like X is independent of Y. ... I decided to run a correlation. ... result looked like white noise, though with about twice the RMS ...
    (sci.math)
  • Re: r-Squared Question
    ... correlation between observed and predicted, then it's a weakness. ... My point was that while, as you say, "For OLS fitted regression, R^2 is ... ALWAYS the [square of the] correlation between the observed Y and the ...
    (sci.stat.math)
  • Re: multiple regression question
    ... using the tolerance values, multicollinearity was not present. ... is totally uncorrelated with the other predictor can become ... the size of the error term is reduced. ... correlation with the original variable. ...
    (sci.stat.edu)