Imagine that you have a table giving invoice values. You will want your spoof data to conform with the same ups and downs of the real data over time. You may be able to get the overall distribution the same as the real data, but the resulting data would be useless for seeing the effect of last years sales promotion. The invoice values will depend on your sales promotions if your marketing people have done their job properly.
By making your data the same distribution as your production data, you don’t necessarily get the same strategy chosen by the query analyser, but you dramatically increase the chances of getting it. SQL Server uses a complex paradigm to select amongst its alternative plans for a query. It maintains distribution statistics for every column and index that is used for selecting rows. These aren’t actually histograms in the classic sense, but they perform a similar function and are used by the SQL Server engine to predict the number of rows that will be returned.
The focus is on independent variables, though there is a little bit at the end about working with dependencies.