I am currently conducting research on the influence of budget revisions on budget absorption in government units. The estimation method I am using is the system GMM, with approximately 15,000 observations across 10 periods of data (5 years, semi-annual). Among the various models tested, the analysis for all observations (around 15,000 units) still does not yield a valid Hansen J Test (Hansen p-value = 0.00). However, when I conduct decomposition or subsample analysis, I find that the model is valid for certain groups of observations, such as observations in Island A, Sector B.

The questions are:

  • Is it common practice to conduct subsample analysis when the analysis at the general level is not valid?
  • How do we interpret modeling that is valid only at specific subsample levels? Can we still conclude that the instruments used are valid partially?
  • Are there any references I can use to perform subsample analysis as an empirical strategy?
  • Are there similar studies that have found similar results that can be used as a reference?
  • Similar questions and discussions