Grubbs' test is based on the assumption of normality.

For Grubbs' test, all variables need to be log-transformed.

Grubbs' test is not a statistical test. T

here is no strict rule to decide, but some literature

suggests that points for which Grubbs’s value is higher than 1 should be

considered as outliers. Other literature provides a threshold of 4/n or

4/(n−k−1).

Since the Grubbs' test detects influential observations in the model, the model must be estimated first.

Grubbs' test detects one outlier at a time.