2
$\begingroup$

I'm using SAS in order to create a model for a cars datasets. The response variable y, is the price of the car. By the way I'm using the PROC MODEL statement in order to check heteroskedasticity. This is the output:

                               Test di eteroschedasticità
Equazione       Test               Statistica     DF    Pr > ChiQuadr    Variabili

price           Test di White           56.32     27           0.0008    Cross of all vars
                Breusch-Pagan           11.43      7           0.1208    1,
                                                                         normalized_losses,
                                                                         height, bore,
                                                                         stroke,
                                                                         compression_ratio,
                                                                         simbol, peak_rpm

Questions:

  1. Is there heteroskedasticity? Breusch-Pagan accepts the null hypothesis, while White doesn't.
  2. I couldn't put in the proc model statement all my variables, because some of them were qualitative. Could this be a (or maybe "THE") problem?
$\endgroup$
1
  • $\begingroup$ From Wikipedia's articles on the two tests I got an impression that Breusch-Pagan test is a special case of White test. (My impression may of course be wrong.) BP does not consider squares or cross products of regressors in the test equation, while White does. Hence, it seems that in your case these squares and cross products are the ones that yield the significance in the White test. In other words, heteroskedasticity is due to those higher order terms rather than levels of regressors. $\endgroup$ Apr 20, 2016 at 19:57

0