Thanks for the scenarios ¡ª I follow what you are outlining ¡ª and thank you. I do not have directed networks. The dependent variable is time spent as association between individual lizards. Independent variables are all continuous, such as how close retreats are from each other or difference in mass or personality measures (cm flight initiation distance).?
It seems that you could covert all to rank values and then matrify¡ª but again, not sure when/why you¡¯d choose to look at absolute difference vs. other metrics.?
Thanks for your patience ¡ª just trying to understand!?
--AE Nash
(Sent from an iPhone: please excuse lizards, typos, and brevity. )
toggle quoted message
Show quoted text
On Dec 1, 2018, at 4:34 PM, 'Steve Borgatti'
steve.borgatti@... [ucinet] <
ucinet@...> wrote:
?
That¡¯s a very broad question. Here are a couple a couple of illustrative responses.
?
Suppose you have a categorical attribute such as gender, coded 1=female, 0=male.
?
In Attribute to matrix, if you choose exact matches, you will create a matrix X such that Xij = 1 if the I and J are the same gender, and Xij = 0 if I and J are different genders. Now suppose you correlate this with a matrix F in which Fij = 1 if I is friends with J, and zero otherwise. If the correlation is positive, that means that big numbers Xij (¡°big¡± meaning 1) tend to occur when Fij is big (i.e., 1) and when Xij is small (0), Fij tends to be small (0). In other words, a positive correlation would indicate that friendships tend to be between people of the same gender.
?
If you choose sender effects, you create a matrix such that Xij = gender of I. If you then correlated this with friendship, a positive correlation would indicate that when Xij is large (1), friendship tends to be large (1) and when Xij is small (0), friendship tends to be small (0). That means a friendship tie from I to J is more likely when I is female. In short, women have more outgoing ties than men.
?
Suppose you have a continuous attribute such as age. And a network such as seeks advice from, as in Aij = 1 if I seeks advice from J and Aij = 0 if I does not seek advice from J.
?
If you choose difference, you create a matrix X in which Xij = Age(i) ¨C Age(j). So if you correlate with Advice, a positive correlation would occur if, when Xij is large (I older than J), advice tends to be large (1), and when Xij is negative (J older than I), advice tends to be small (0). A positive correlation would mean that older people tend to ask advice of younger people (e.g., which mobile phone should I buy?)
?
steve
?
?
?
When converting attribute to matrix data, can anyone explain/point me to resource about when you would use each similarity metric and why? Thanks in advance,
?
AE Nash
