1

First question on the site. I have been following Princeton's plm package guide and have everything done (without errors). However, I am stuck in the last step of estimating a "random" effects model. I think my data fits a random model well, so this is a big issue. I cannot understand why the fixed specification works but random does not. Here is the error I get: 

Error in solve.default(M[therows, therows], quad[therows]) : Lapack routine dgesv: system is exactly singular: U[2,2] = 0

Any help on this would be really useful!

Here is my code: 

random <- plm(Reserves ~ Corruption +
                          RuleofLaw + 
                          BankingRisk + 
                          GDPPC + Electricity + 
                          Mobile + ATM + SchoolEnrollment + Savings + 
                          Population + NumberPOS + POSper, 
                          data=paneldata,
                          index=c("Country", "Year"), model="random")
summary (random)

All of my variables are general format in csv.

Henry Cyranka
  • 2,970
  • 1
  • 16
  • 21
PBhGU123
  • 11
  • 2
  • I cannot provide a thorough answer, because I'm still learning mixed/random models myself. The error means that `plm` cannot invert a matrix which is designed to depict the covariances between or within your random effects. You'll need to dive into the theory of random models to understand why. In my practice, if this error occurs it may mean that your model is overspecified, so your data is not enough to estimate all the effects you wrote in your formula. For example, what if two of your observables, e.g. `Electricity` and `ATM` are always the same? The model can't decide between those then. – akraf Nov 12 '18 at 21:36
  • I actually just figured out the solution to it. In my case, all the independent variables are unique, so it was tougher. The default for plm under model = random is that random.model is always "swar". If you are to replace that with "walhus" it fixes the issue. Walhus is more tolerate of smaller values (I have a lot of variables that are second of third digit of zeros) – PBhGU123 Nov 12 '18 at 22:05
  • You can post this as an answer to your own question and accept your own answer after some time, this marks this case as done on the site. – akraf Nov 13 '18 at 09:57
  • Can you make the data available to check why the Swamy/Arora model fails? – Helix123 Nov 15 '18 at 12:39

0 Answers0