Date: Mon, 29 Jun 1998 10:40:01 -0500
Reply-To: "Greer, Gary" <GreerG@ZEUS.DT.UH.EDU>
Sender: "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From: "Greer, Gary" <GreerG@ZEUS.DT.UH.EDU>
Subject: goodman's gamma
Content-Type: text/plain
Thanks Rich for providing information about Goodman & Kruskal's gamma;
their article is referenced in SPSS Users Guide : (Goodman & Kruskal,
1954, Measures of association for cross-classification, J. of the Am.
Statistical Assn. 49, 732-764 ).
Marascuilo, 1977, p. 207 describes (another?) Goodman's gamma as a log
transform of Phi (cf. Fisher's z transform of r). The transforms permits
CIs, etc. Marascuilo's reference is Goodman, Leo J., 1964, J. of the
Royal Statistical Society, Series B, 26, 86-102.
Thanks for your frequent contributions and clarifications.
Gary
U. of Houston Downtown
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++
> : SPSS generates EXACT tests when _________ % of
cell expected
> values
> : are less than 5. What is the percentage that SPSS uses as a
> minimum?
>
> - check your documentation?
>
My documentation says that Fisher's Exact test is generated if ANY
EXPECTED CELL VALUE IN A 2 X 2 table is less than 5. So, then Fisher's
Exact test is ONLY for 2 x 2 tables?
>
> : Regarding gamma; is this Goodman's gamma? The help menu seems
> to
> define
> : gamma as Kendall's tau.
>
> - See the SPSS Statistical Algorithms manual - I have Release
> 8.0 -
>
> It defines some statistics in terms of P and Q, which are the
> counts of "concordant" and "discordant" "pairs" (check
> closely,
> to understand what those "pairs" are). SPSS uses the (usual)
> gamma
> that is described, say, in Agresti's text, Categorical Data
> Analysis.
> Gamma is (P-Q)/(P+Q) . Kendall's tau-b has (P-Q) in the
> numerator,
> and a more complicated denominator.
>
>
> : We assert that gamma is the log transform of Pearson's phi.
> Gamma's
> << snip rest, including some 'gamma' citation that looks
> irrelevant;
> I will try to check >>
>
> ??? I don't know why you would *assert* that, unless you
> were
> intending to follow with a clever proof of identity. But in
> that
> case, you should not be asking, "What's what?"
>
> I don't even guess what you MEAN by "log transform of Pearson's
> phi"
> but I don't think it is any one-to-one transform of phi.... if
> that may help your thinking.
>
>
> --
> Rich Ulrich, biostatistician wpilib+@pitt.edu
> http://www.pitt.edu/~wpilib/index.html Univ. of Pittsburgh
>
>
>
| 
