Network matrix

[Haja Ham]

Hello everyone,

I’d like to build a network matrix G, where G_ij = n_i^-1, in which n_i is the size of the group of individuals (excluding the individual himself). The information about the size of the group is given by the variable “tamanho”. We consider students belonging to the same group only those who are in the same class. The classes are identified by a code given by the variable “turma”. Moreover, we have 5 distincts periods, identified by the variable “onda”. Thus, I want to build a matrix G for each one of the periods. The individuals are identified by the variable “idaluno”. The shape of the database is the following:


(id student) (id class) (period) (class size)
idaluno turma onda tamanho
1 100 1 2
1 101 2 2
1 700 3 3
1 900 4 1
1 540 5 1
2 100 1 2
2 101 2 2
2 700 3 4
2 870 4 3
2 450 5 3
3 200 1 2
3 201 2 2
3 700 3 4
3 870 4 3
3 450 5 3
4 200 1 2
4 201 2 2
4 700 3 4
4 870 4 3
4 450 5 3

Thank you for any help!

[Haja Ham]

In other words, I want a code that if i (idaluno) and j (other idaluno) are in the same class (turma), defines g_ij = 1/tamanho and if i and j are not in the same class,g_ij = 0 .

[Sergio Correia]

So you want something like a weighted adjacency matrix?

1. Maybe this set of tools for network analysis can help (it has utilities that create adj matrices, etc.): https://nwcommands.wordpress.com/getting-started/
2. How big is your problem? you can probably brute force it with Mata and asarray() if it's not huge

[Haja Ham]

Thank you Sergio!

Yes, I want a wighted adjacency matrix.
I've tried some nwcommands but it did not work until now.
My dataset have more than 100,000 observations and the dimension of each matrix will be approximately 20,000 x 20,000. I was trying to create these matrices with brute force with Mata, but I need some help with it.

thank you again.

[Sergio Correia]

What will you use these matrices for? There is an example in the help for -asarray- of using it to create a sparse matriz, which might be what you want given the size of your problem.

[Haja Ham]

I'm studying the peer effects over academic achievement. So, I want to use these matrices for the estimation of the reduced form of equation: (I-G)Y=β(I-G)GY+γ(I-G)X+δ(I-G)GX+(I-G)ε, under the hypothesis that students interact through a network. I will try -asarray- command.

Thanks!!

[Benjamin Chartock]

Perhaps consider netsis, by Miura (2012) http://econpapers.repec.org/article/...p_3a94-129.htm