***** To join INSNA, visit http://www.insna.org ***** You could consider doing the quadratic assignment procedure (QAP) proposed by Krackhardt (1987).
You can do a cosine similarity on the dyad or triad census.
You can calculate Hamming or Euclidean distances.  Hamming 1951.

I am NOT saying that these are better than the nonparametric tests proposed by Snjiders and Borgotti.  These are just other tests that have been used in the literature to compare networks.

ORA does QAP, Hamming, Euclidean distances, and correlation between networks in the QAP/MRQAP report.  The different networks must be coded within the same meta-network for it to work.  ORA is free and available at www.casos.cs.cmu.edu.  ORA also has good over-time analysis features.

I have some code somewhere to calculate dyad and triad census, but I am sure there are more elegant programs than mine.  I think Stan Wasserman has a program that does this.

Best of luck.

Ian

Ian McCulloh, Ph.D.
Major, US Army
Assistant Professor
West Point, NY 10996

On Wed, Nov 25, 2009 at 1:48 PM, Nilesh Saraf wrote:
*****  To join INSNA, visit http://www.insna.org  *****

Dear Colleagues,

We are working with two small 17 members networks and need to run the following statistical tests.

1. Compare various network statistics of each network with itself over time (3-4 cross sections)
2. Compare the two networks with each other over time.

For this I am referring to Snijders and Borgatti note on Non-Parametric Standard Errors and Tests for Network Statistics in Connections. So my questions are:

Is there any other statistical literature that pertains to precisely such statistical tests?

Is there any software package to compute the network statistics (centrality, core/periphery, and many others), compute the (boostrap and/or jacknife) estimators of standard errors and spit out the t-test results?

If not, has anyone written code to do all this in one go? Preferably in SAS?

UCINET does compare densities but nothing else -- and the SPAN package by James Moody does not have any core-periphery measures that I can append with some matrix manipulations for computing the SEs myself.

Are there some features in existing software packages that can be indirectly used to do what we want?

thanks very much,

nilesh

Nilesh Saraf
Assistant Professor
Management Information Systems