Data-snooping

In the Swedish market, it is common to use a method based on data-snooping.

The Dutch geodesist Baarda (1968) developed this formal statistical test to search for gross measurement errors.

Abstract

In data-snooping, we test standardized residuals (\(W_i\)) to identify potential gross measurement errors in the observations.

Process for data-snooping

Data-snooping should be performed as an iterative process where a maximum of one measurement is removed in each iteration:

1. Find all standardized residuals with sigma greater than rejection level
2. Remove the observation with the largest sigma value
3. Run gross error detection again
4. Continue steps 1 to 3 until all statistically detected gross errors are removed
5. If more than one observation is removed in steps 1 to 4, start by reintroducing the observations one by one. Run a new analysis and check if it is still marked as a gross error. If so, remove it from the observation material.

If the gross errors are not too numerous and/or if they are in different parts of the network, data-snooping works quite well – especially if the k-value is reasonable.

Warning

This form of gross error detection is sensitive to errors in observation weighting. It is critical to use weights that reflect the actual observation errors.