3
$\begingroup$

How do i test/verify if i can analyze my panel like dataset by simply pooling the individual series?

I have a dataset structured as a panel. Now I am wondering if i can simply pool the individual series and estimate it via OLS or if I have to use another estimation technique.

(Any R hints and references are highly welcomed.)

Thanks!

$\endgroup$
4
$\begingroup$

You can fit a hierarchical bayesian (HB) model without pooling and do an ordinary OLS by pooling the data and compare the models in terms of model fit, hold-out predictions etc to evaluate whether pooling outperforms the HB model. The model very briefly will look like so:

Model

$y_i \sim N(X\ \beta_i,\sigma^2\ I)$

$\beta_i \sim N(\bar{\beta},\Sigma)$

Priors

$\bar{\beta} \sim N(\bar{\bar{\beta}},\Sigma_0)$

$\Sigma \sim IW(R,d)$

$\sigma^2 \sim IG(sp,sc)$

While I do not use R, I do know that there are packages that will do the above for you. Someone more knowledgeable about R can perhaps help you out.

| cite | improve this answer | |
$\endgroup$
3
$\begingroup$

Srikant is right. The book you want is "Data Analysis Using Regression and Multilevel/Hierarchical Models" by Gelman and Hill, all the R code from the book, and the associated arm package in R.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ that book looks pretty nice! i just installed the 'arm' package, too. thanks shane! $\endgroup$ – user283 Aug 18 '10 at 13:52
  • $\begingroup$ @mropa: That is my favorite statistics text: enjoy it! $\endgroup$ – Shane Aug 18 '10 at 13:54
3
$\begingroup$

The only further comment I would make is that the approach need not be Bayesian and the model need not be a mixed or random effects model.

In the simplest case if you had two series in x the mean model may be:

y = b01 + I*b02 + b11*x + I.b12*x

Where I indicates a sample from the 2nd series. An omnibus F-test can be used to determine whether the additional parameters are required to maintain distinct series (Ho: b02 = b12 = 0). http://en.wikipedia.org/wiki/F_test This can be extended to more series, but it soon becomes more efficient to use a mixed or random effects model.

| cite | improve this answer | |
$\endgroup$
0
$\begingroup$

The plm package provides a a function for the poolability test in just three steps:

# 1. Run a normal OLS model with fixed effects (model="within")
plm_model<- plm(y ~ x, data= dataset, model= "within"

# 2. Run a variable coefficients model with fixed effects (model="within")
pvcm_model<- pvcm(y ~ x, data= dataset, model= "within"

# Run the poolability test
pooltest(plm_model, pvcm_model)

The null hypothesis is that the dataset is poolable (i.e. individuals have the same slope coefficients), so if p<0.05 you reject the null and you need a variable coefficients model.

More info here

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.