GEE & ICC (intra-class correlation) Question

[Emil Coman <emilcoman@yahoo.com>]

Following Chris Barker?s suggestion (thanks!) I ran GEE using a command like:

> myGee.fit <- gee(formula = Y ~ X + Z, cluster = C, variance = "glm.scale", data = "" subset = A == 1)

and then asked for

> names(myGee.fit)

which listed

[1] "coefficients"  "correlation"   "scale"         "var.random"    "cor.work"      "iter"        

 [7] "fitted.values" "variance.y"    "variance"      "call"          "family"        "cor.design"  

[13] "control"       "n.clusters"    "n.obs"         "initial"       "residuals"     "contrasts" 

Now, if I look at:

> summary(myGee.fit$cor.work)

I get

  Min. 1st Qu. Median  Mean 3rd Qu.  Max.

 0.000 0.000   0.000  0.125 0.000   1.000

So, is this 0.125 the nonindependence or ICC (intra-class correlation)?

And if so, does this correlation represent the ICC in the outcome variable Y, or is it a measure of the group-level correlations between (ALL) predictors and the outcome? Kenny & LaVoie talk about this distinction between an individual-level correlation (correlation between group adjusted scores with group means substracted first) between 2 variables, and the group-level correlation (correlation between group means basically), so:

Do I see here this group-level association between X, Z and Y, or just the nonindependence in the Y outcome?

Thanks, emil

Kenny, David A.; Voie, Lawrence La  1985 Separating Individual and Group Effects  ; Journal of Personality and Social Psychology. Vol. 48(2) February. pp. 339-348.

    I could provide the Kenny & La Voie paper if needed.

Emil Coman, Statistician    
Institute for Community Research
2 Hartford Square West, Suite 100
Hartford, CT. U.S.A. 06106
Tel: 860-278-2044, ext. 257
www.incommunityresearch.org
Fax: 1-860-278-2141
http://sp.uconn.edu/~enc01003

The Institute for Community Research (ICR) is a non-profit organization in
Hartford, CT that conducts collaborative research with community partners
to promote justice and equity in a diverse, multiethnic world. We build
community-based research partnerships to promote
 positive changes in public
health and education, and foster cultural conservation and development.

SAVE THE DATE: Join ICR for the "Crossroads II: Transformations through
Community-Based Collaborative Research" Conference, June 7-9, 2007 in
Hartford, CT! Check www.incommunityresearch.org for updates on
registration, agenda, and related events.

---

Feel free to call! Free PC-to-PC calls. Low rates on PC-to-Phone.  Get Yahoo! Messenger with Voice