India has 29 States, each of which is further sub-divided into smaller administrative Districts of varying size and population. I need to download micro (firm-level) data from each state, but I do not need to choose all the districts.
How do I choose the Districts?
The conditions are the following:
- Since I want data that is representative of India, I was thinking that if I obtain 30-40 per cent of all data from India with a suitable sampling method, this should be more than sufficient.
- I need data for all the firms in each district.
- I need data from all states.
- The number of firms in the dataset from each State should be proportionate to the proportion of firms from that State in the whole of India. Since I want 30-40% of the whole population of firms in India, would it be appropriate to simply select 30-40% of those in each State?
My proposed solution - Since I need data from all States,I can treat each State as a population. I then treat each District within a State as a cluster, and select Districts at random (without replacement), with the probability of selection being proportionate to the proportion of all firms in that State that come from that District. I keep selecting entire Districts until I have say, 35% of all the firms in that State.
Would this work? Is their a better alternative given the conditions that have to be met?
EDIT: added "(without replacement)"