Bonjour Dear SOCNET Members,
Would you be kind to advise how to treat "missing values" in longitudinal ego networks, when "missing" represent as having no alliance of ego with anyone in a given year. Can we impute zeros (Huisman, 2009) in such cases to balance the panel data. As a matter of fact, I imputed zeros for such cases, however has been facing issue of non normal distributions of residuals when I analyze panel data (in STATA, using random or fixed effects) due to large number of low values or zeros. Can you suggest solution please.
Moreover can we transform to normalize ego network variables (efficiency, degree centrality etc.)? Due some outliers in my network, the relationships between independent and dependent
variables becomes non linear. Do we need to treat outliers in models where objective is to test network effect on performance?
Highly appreciate if someone could extend helping hand.