Survey analysis: relationship between one discrete outcome variable and multiple categorical and ordered categorical variables I have a data set that is the result of a survey. The survey asks the respondents to name 5 people in their community whom they turn to for advice. It then goes to these 5 people and asks the same. I have calculated in-degree for each person (the number of people who have chosen them) and I would like to understand how the person's various features affect their in-degree.
The survey asked each person their age range, gender, location, religion, how often they read/watch news/go online..etc. (daily, weekly, monthly), job and other similar questions considering the person. I have information missing about 44% of the people who were mentioned, since not everybody who was mentioned was surveyed.
They also asked questions about the relationship between the person and the person they turn to for advice - how long thay have known each other (5-10 years, less than 5 years, 10-20 years, over 20 years), how often they meet and other similar things. I have relationship information for every connection.
I would like to understand which predictors best affect a person's chance to have a high in-degree as well as which relationship attributes influence a person's decision to choose somebody as the person they turn to for advice.
Which statistical methods should I be looking into? Do I need to transform all my categorical variables? What would be the best way to do that? Also, in-degree distribution, so the outcome variable, is not normally distributed, it is extremely left-skewed.
 A: In general, it's best to not use degree as a dependent variable when both the sender and receiver are included in the data. It's a problem of interdependent observations--an actor's indegree depends both upon their propensity to receive (or not receive) ties as well as the propensity of all other actors to send (or not send) ties, in addition to triadic and other higher order effects.
Rather then posing the question as an actor ability to have high (or low) indegree, it's best to pose the question as a matter of tie formation for the entire network. Some of the variables will be measured at the individual level and these measurements can correspond to an actor's propensity to receive ties.
The method to look into is referred to as p* or "exponential random graph models." The literature on this method is large and there are a number of resources and tutorials readily available.
Lastly, it's unusual that the degree distribution is left-skewed. Degree distributions are usually right-skewed.
