You will have to make two decisions here: first, choose a total
number of administrative divisions (clusters) to include in
your sample, and then distribute the clusters between strata.
While budgetary considerations largely dictate the number
of administrative divisions you can sample, every stratum
must be represented. The fewer you choose, the fewer
places your researchers have to visit to conduct interviews,
which translates into lower expenses. However, as a general
rule, the fewer clusters you select, the less representative
your sample of the general population will be.
Generally, aim to select 30-50% of divisions. In countries
with many administrative divisions, the proportion you
select is likely to be smaller. For example, if a country is
divided into 50 divisions, you may only be able to select 15
(30%), whereas in a country with only 10 divisions, you
should easily be able to select five (50%).