Answer Happy

Suppose that the individuals are divided into groups j = 1,..., J each with n, observations respectively, and we only observe the reported group means y, and a,. The model becomes ÿj = Bæj+uj, (2) with error terms uj = 1, where u, indicates error term u, of individual i belonging to group
3(g) Test the null-hypothesis that Ho: E=o² for j = 1,..., J, against the alternative that the variance is a smooth unknown function of ;. Explicitly state which regression(s) you use, the null and the alternative, and the test statistic with its distribution under the null. (5 marks) 3(h) Suppose that we observe the group size n, for j = 1,..., J. Regress √n, on √n,. Show that the error terms of this regression are homoskedastic. (4 marks) 3(i) Apart from the problem of heteroskedasticity in the errors, what would be another reason to prefer the regression with individual data over the regression with grouped data? (3 marks)