Every level can be sorted by a different statistic or label. In this sort the first level is "Label", it does not require setting more parameters. It could have been any statistics type calculated for this variable in a given spreadsheet.
The second level will be unweighted count. There is only one Unweighted count in the spreadsheet, but we still have to choose within which of the models the order of sort will be calculated. It is possible to use the sample total.
The third level will be calculated for "%of column, sample base". There is more than one column with this statistic type so it is necessary to determine where the calculations of order sort should take place.
In some cases sort is more complicated. Please see an example spreadsheet below.
If you want to sort Make of new phone in this query by "% column, sample base" you have to define within which of the columns the calculation should take place.
It might be even more complicated with nested questions (see example below):
