Print

Print


 Dear Steve and others

 I have another question about the QAP. How does the QAP in the UCINET take the size variable into account? 
In other words, values of each cell in the first matrix are much smaller (or larger or binary) than those of the other matrix. 

 Thanks in advance, Han.

 
 you have taken the size variable fully into account in your analysis, for example as an explanation for the QAP correlations.

Dr. Han Woo Park, Research Associate
Networked Research & Digital Information(Nerdi)
Royal Netherlands Academy of Arts & Sciences(NIWI-KNAW)
Joan Muyskenweg 25, PO Box 95110
1090 HC Amsterdam, The Netherlands
T 3120 4628737, F 3120 6658013
http://www.niwi.knaw.nl/nerdi
http://www.pscw.uva.nl/ascor/
http://www.wmw.utwente.nl/wtmc/

>>> Steve Borgatti <[log in to unmask]> 08/07/02 08:40PM >>>
Christina, a couple of responses:

1. Assuming you hypothesized positive autocorrelation (i.e., like goes with
like and Geary coefficient SMALLER than 1), then you should regard the
proportion as small as your p-value.

2. Your comparison of geary with qapping adjacency against difference matrix
is sensible, though I think Geary uses squared differences.

3. I'm sure you realize (but is worth confirming) that the two p-values
being exactly identical (.008) is something of a coincidence, in the sense
that these are constructed by generating 'random' distributions on the fly.
So even within one method, such as qap, you can get a different p-value the
next time you run it. The variation decreases with the number of
permutations used to construct the distribution.

steve.

----- Original Message -----
From: "Christina Prell" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Tuesday, August 06, 2002 7:12 AM
Subject: QAP test differs from Geary Autocorrelation, huh?


> Dear SocNet,
> I received two different results using first QAP (matrix x matrix
> correlation)
>  and then Autocorrelation (matrix x vector -- interval/ratio data --
> correlation). The same data was used for both tests. Please read on:
> 1. Autocorrelation:
> I used the normalized degree centrality score (vector data), with a
> square matrix.
> I performed  an autocorrelation on this data using Geary's C statistic
> as the default model,
> I got the following results:
>   Autocorrelation:       0.629
>              P as Large :       0.992
>        P as Small:  0.008
> According to my understanding of the Geary statistic (just based on the
> UCINET help),
> this means I don't have a significant correlation by a long shot.
> 2. QAP:
>
> However, when I took that same normalized degree score, transformed it
> into a square matrix
> through the UCINET command Data --> Attribute (chose absolute
> difference as method), and
> then ran a QAP with the same square matrix data, I got significant
> results:
>
> Pearson Correlation:      r = 0.058     p = 0.008
> Notice that P as Small for Geary is the same p value for Pearson.
> However, my understanding
> re: the Geary is that one should read the P as Large value.
>  So what's up? Thanks in advance, Christina
>