I'm analysing slightly unbalanced panel data in r (n = 136, T = 9-213, N = 24894) and planning on using fixed or random effects (Hausman test indicates random effects, but high likelihood of omitted variable bias in the model due to lack of available data so unsure whether that means that random should be ruled out in this case?).
This is an example of the model:
rm2 <- plm(Assault_rate ~ Private + Sex + Youth,
index = c("Institution", "Month_yr"),
model = "random", data = panel1)
The dependent variable is rate of assaults, and the primary explanatory variable is a binary dummy for privately run institution (12% fall into this category). Other independent variables include male/female institution, youth/juvenile institution. The likely confounding unobserved variables would be staffing levels, funding/expenditure, activities/programs (no data possible for these).
When I try to run tests for cross-sectional dependence it returns NA and the below warning message:
Breusch-Pagan LM test for cross-sectional dependence in panels
data: AA_rate ~ Private + Sex + YOI
chisq = NA, df = 9045, p-value = NA
alternative hypothesis: cross-sectional dependence
Warning messages:
1: In cor(wideres, use = "pairwise.complete.obs") :
the standard deviation is zero
2: In pcdres(tres = tres, n = n, w = w, form = paste(deparse(x$formula)), :
Some pairs of individuals (2.9 percent) do not have any or just one time period in common and have been omitted from calculation
Does this mean that there is something wrong with my data or model? Is there something fundamentally wrong with this approach that I'm missing?
I am new to this so apologies if I've formatted this incorrectly or missed any important information. Thanks in advance for any advice
I have tried re-specifying the model with different combinations of independent variables, different types of model (pooled, fixed) but get the same warning and NA result.
