Re: [igraph] How to test if there is any relationship between attributes of edges and nodes

[Francesco]

If I run a regression analysis, will I obtain a probability distribution for 
the values of E(g)$weight? In other words, will I be able to tell that given 
V(g)$size of the two vertices, E(g)$weight is higher/lower than expected?

On 06/02/2014, at 10:31 AM, Tamás Nepusz <address@hidden> wrote:

>> I want to test two hypothesis:
>> 1) When V(g)$size increases for one or both nodes, E(g)$weight of the  
>> connecting edge also increases;
>> 2) When the difference between the two V(g)$time decreases (thus the  
>> event happens closer in time), E(g)$weight of the connecting edge  
>> increases.
>> 
>> I think I could attempt to use a regression analysis for the second  
>> hypothesis using the difference in time as the independent variable and  
>> E(g)$weight as the dependent variable. But what to do (statistically)  
>> and how to do it (in terms of code) with the first hypothesis?
> 
> Regression analysis works not only for a single independent variable; you 
> could have multiple independent variables as well; search for “multivariate 
> regression” and I’m sure that you’ll find plenty of resources on the 
> Internet. Basically, you would then use the sizes of the source and target 
> nodes as independent variables. In the simplest case, you could use linear 
> regression, where you would end up with three coefficients (two slopes and an 
> intercept).
> 
>> Also wouldn't be better to test both independent variables V(g)$size and
>> difference in V(g)$time in the same model?
> 
> Well, you could do that as well; just take a multivariate regression 
> technique and feed it with the sizes of the two endpoints and the time 
> difference - that would give you three independent variables. You can still 
> use the same technique as above.
> 
> T.
> 
> _______________________________________________
> igraph-help mailing list
> address@hidden
> https://lists.nongnu.org/mailman/listinfo/igraph-help