In Example 11.6 (Indonesian rice farm data) assess gain from introducing AR1 errors (in addition to unstructured errors) in both random and fixed effects bi models. Also find the posterior probabilities that farms 1 to 171 are the best – in terms of having highest bi after allowing for inputs. Which farm has the highest probability of being best?

Multiple comparison with the best To illustrate a multiple comparison model where both fixed and random effects approaches to the permanent subject effect may be relevant, consider data from Horrace and Schmidt (2000) applied to loglinear production functions. The observations are rice outputs for n = 171 Indonesian farms over T = 6 seasons with inputs being

1.       metric variables: seed in kg (KGS), urea (KGN) and trisodium phosphate (KGP), labourhours (LAB) and land in hectares (LAND).

2.       categorical variables: namely B P = 1 if pesticides used, 0 otherwise; VAR (1 if high-yield rice varieties planted, 2 if mixed varieties planted, 3 if traditional varieties planted); and BWS (1 for wet season).