Tests of heteroscedasticity:

It is natural  for us to  consider whether appropriate statistical procedures  can be found to test for heteroscedasticity. We wish  to find a test statistic for testing the null hypothesis of homoscedasticity,  that is, where x  is  the number of observations. The specific  alternative hypothesis against which  the null hypothesis  is  to be tested depends  on  the estimation procedure that is considered  to yield the most desirable correction  for heteroseedasticity.

While  there are a number of specific tests  for heteroscedasticity, a useful  first procedure  is  the  informal one of examining  the  pattern of  the  residuals  to see whether estimated variances differ from observation  to observation. To do  this, we suggest calculating the squares of the residuals,and plot these against one or several variables  in the model.