Print

Print


*****  To join INSNA, visit http://www.insna.org  *****

Hi Peyina,

The 'goodness of fit' is not a bad way to interpret Q.

You can fit any grouping of nodes to any network. You can arbitrarily
assign nodes to an arbitrary number of groups. For any arbitrary grouping
you can calculate a value of Q. Of course, as the size of a network
increases attempting all possible combinations of groupings becomes
intractable. So you have the Girvan-Newman method (and a great, huge host
of other methods) which operate on good assumptions about network
structure. The Girvan-Newman method assumes that edges in the network with
high betweenness are edges tying communities together.

The Q (modularity) value is based on a null model of network connectivity.
 In a random network you would expect the likelihood of two nodes   to be
linked to be based on the degree of each of the nodes.

Pab = (degree(a) * degree(b)) / (2 * total edges in network)

Community detection algorithms, generally, attempt to find a way of
grouping nodes together such that they are more likely to connect to other
members of the group than would be expected based on this null model. Most
community detection values attempt to find the *highest* value of Q
possible (although a value greater than 0.7 would probably not be
considered 'significant', but most algorithms won't approach that value).
You would get a value of Q=0 if there was one community and all nodes
belonged to it and a value of Q=1 if there were as many groups as there are
nodes and there was a single node in each group. Both Q=0 and Q=1 are
not significant. The best values for Q are between Q=0.3 and Q=0.8, but
there could be considerable discussion concerning the 'best' values.

Check out Chris's recommendation. It will provide a good survey of methods
and uses of modularity in networks. I hope this helps.

-Jesse Fagan


On Wed, Feb 8, 2012 at 4:26 PM, P. Lin <[log in to unmask]> wrote:

> ***** To join INSNA, visit http://www.insna.org *****
>
> Hi,****
>
> ** **
>
> Iím a PhD student at the University of Washington Information School
> studying high school social groups and the multiplexity of technology
> interactions.****
>
> ** **
>
> Iím intrigued by the *Q-value in Netdrawís Girvan-Newman subgroup
> function.* According to Hannemanís online book on social network methods,
> this Q-value is a goodness of fit.****
>
> *Does anyone know how to interpret the Q value (or goodness of fit) in
> Netdraw's Girvan-Newman subgroup function?*
>
> *What is considered a good fit?*
>
>  ****
>
> I got no responses on the UCINET group (posted 2 days ago), and emailed
> Dr. Hanneman, who kindly suggested I email this listserv.****
>
> ** **
>
> To elaborate, Netdraw has an Analysis>Subgroup function that :****
>
> ** **
>
> a) visualizes the subgroups partitioned by coloring the subgroups in
> different colors.****
>
> b) produces partition information, including a Q value.****
>
> ** **
>
> I looked through the Girvan and Newman (2002) article cited for this
> algorithm, and couldn't see an explanation of the goodness of fit either.*
> ***
>
> ** **
>
> UCINET does not produce this goodness of fit measurement either, which
> leads me to think that the goodness of fit is specific to Netdraw?****
>
> ** **
>
> The Netdraw output makes sense to me (in the way I'm interpreting my
> data), and I'm just looking for a way to validate the clusters produced.
> So, perhaps the Q goodness of fit measure produced could help, though I
> need some help in knowing *how that Q value is produced and what it means.
> *
>
> ** **
>
> Thank you!****
>
> ** **
>
> ** **
>
> Peyina****
>
> -****
>
> Peyina Lin****
>
> PhD Candidate****
>
> The Information School****
>
> University of Washington****
>
> http://students.washington.edu/pl3****
>
> @peyinalin****
>
> ** **
> _____________________________________________________________________
> SOCNET is a service of INSNA, the professional association for social
> network researchers (http://www.insna.org). To unsubscribe, send an email
> message to [log in to unmask] containing the line UNSUBSCRIBE SOCNET
> in the body of the message.



On Wed, Feb 8, 2012 at 4:44 PM, Weare, Christopher <[log in to unmask]>wrote:

> ***** To join INSNA, visit http://www.insna.org *****
>
> This may have more information than you want, but this paper covers the
> modularity index very well.****
>
> ** **
>
> Porter, M. A., J.-P. Onnela, et al. (2009). "Communities in Networks." *Notices
> of the American Mathematical Society, Vol. 56, No. 9, 2009*.****
>
> ** **
>
> Chris****
>
> ** **
>
> *From:* Social Networks Discussion Forum [mailto:[log in to unmask]] *On
> Behalf Of *P. Lin
> *Sent:* Wednesday, February 08, 2012 3:26 PM
> *To:* [log in to unmask]
> *Subject:* [SOCNET] Girvan-Newman's Q value in Netdraw****
>
> ** **
>
> ***** To join INSNA, visit http://www.insna.org ***** ****
>
> Hi, ****
>
> ** **
>
> Iím a PhD student at the University of Washington Information School
> studying high school social groups and the multiplexity of technology
> interactions. ****
>
> ** **
>
> Iím intrigued by the *Q-value in Netdrawís Girvan-Newman subgroup
> function.* According to Hannemanís online book on social network methods,
> this Q-value is a goodness of fit. ****
>
> *Does anyone know how to interpret the Q value (or goodness of fit) in
> Netdraw's Girvan-Newman subgroup function?*
>
> *What is considered a good fit? *
>
>  ****
>
> I got no responses on the UCINET group (posted 2 days ago), and emailed
> Dr. Hanneman, who kindly suggested I email this listserv. ****
>
> ** **
>
> To elaborate, Netdraw has an Analysis>Subgroup function that : ****
>
> ** **
>
> a) visualizes the subgroups partitioned by coloring the subgroups in
> different colors.****
>
> b) produces partition information, including a Q value. ** **
>
> ** **
>
> I looked through the Girvan and Newman (2002) article cited for this
> algorithm, and couldn't see an explanation of the goodness of fit either.
> ****
>
> ** **
>
> UCINET does not produce this goodness of fit measurement either, which
> leads me to think that the goodness of fit is specific to Netdraw?****
>
> ** **
>
> The Netdraw output makes sense to me (in the way I'm interpreting my
> data), and I'm just looking for a way to validate the clusters produced.
> So, perhaps the Q goodness of fit measure produced could help, though I
> need some help in knowing *how that Q value is produced and what it means.
> *
>
> ** **
>
> Thank you!****
>
> ** **
>
> ** **
>
> Peyina****
>
> -****
>
> Peyina Lin****
>
> PhD Candidate****
>
> The Information School****
>
> University of Washington****
>
> http://students.washington.edu/pl3****
>
> @peyinalin****
>
> ** **
>
> _____________________________________________________________________
> SOCNET is a service of INSNA, the professional association for social
> network researchers (http://www.insna.org). To unsubscribe, send an email
> message to [log in to unmask] containing the line UNSUBSCRIBE SOCNET
> in the body of the message. ****
>  _____________________________________________________________________
> SOCNET is a service of INSNA, the professional association for social
> network researchers (http://www.insna.org). To unsubscribe, send an email
> message to [log in to unmask] containing the line UNSUBSCRIBE SOCNET
> in the body of the message.
>

_____________________________________________________________________
SOCNET is a service of INSNA, the professional association for social
network researchers (http://www.insna.org). To unsubscribe, send
an email message to [log in to unmask] containing the line
UNSUBSCRIBE SOCNET in the body of the message.