Test for outliers


New Member
I have a set of concentrations of lead in soil from samples taken at random (spatially) across a 20m x 20m area. The data fit a normal distribution and on constructing a Quantile-Quantile plot there appear to be three extreme(ly high) values. Grubbs' test shows that the maximum value may be an outlier at p=0.10. I then remove the supposed outlier and run Grubbs's test again and the maximum value is again shown to be an outlier. I do the same again and get the same result. I carry on until the test accepts the Ho (i.e. no outliers). I end up with three supposed outliers.

Is this a reliable application of Grubbs' test?

Another related question: Are there statistical tests for detecting seperate populations in a data set, with known confidence?

Any help or suggestions would be much appreciated.