I have been asked to forward this to the list.
Please reply to the details given and not to me.
I would appreciate it if you would post the following enquiry on your
ALLSTAT statistical forum:
All necessary information cannot be provided here, but I would like to
discuss my statistical conundrum (via email) with one or more advisers who
may have an explanation for the following problem:
a. An 8-item variable is computed using the mean of items (responses range
from 1-2);
b. The composite variable is exponentially transformed to make its
distribution MORE normal;
c. Intraclass Correlation Coefficients are computed for both the raw and
transformed variables;
d. The raw variable shows ICC = .30;
e. The transformed one shows ICC = .08;
f. Data profile: N = 468; N of groups = 44;
g. Variables profile:
COMPOSITE VARIABLE: RAW TRANSFORMED
Mean 1.7626 5.9709
Std. Error of Mean .01061 .05668
Median 1.8125 6.1257
Std. Deviation .22956 1.22618
Skewness -1.010 -.627
Std. Error of Skewness .113 .113
Kurtosis .244 -.599
Std. Error of Kurtosis .225 .225
QUESTION: Why does the ICC statistic change so radically with the linear
transformation?
Please reply directly to me, Adam Long, at <[log in to unmask]>
Thank you.
|