sampling and stats by classification

[neptunhiker]

Hi everyone,

I have a problem with settiung up a sample.

Let's assume I have a dated panelstructure with data for 7 years for 200 companies (revenue), which are located in 10 different countries. So there are about (not exactly) 20 companies located in each of the 10 countries. Now I am trying to look at statistics sorted by classification, i.e. sorted by country. How that works, I know. But, I would like to sample my datastructure in a way, that only the top five companies for each country are to be included in the statistics sorted by classification.

I seem not to be able to find the solution for this probably very simple problem. I'll be very grateful for help or advice. Thanks a lot in advance.

[EViews Gareth]

Not easy. We'll need a bit more information. How is your workfile currently structured (i.e. what are the identifiers?). How are you defining "top 5 companies"?

[neptunhiker]

Hi Gareth,

the identifier series could be the company names. The top 5 companies should be ranked by revenue, i.e. the 5 companies which have the highest revenue in each country are the top 5 for each country. To clarify things, I have attached a workfile with some sample data. I hope that helps to specify my question.

[EViews Gareth]

Not the shortest, but the easiest to see what is going on:

Code: Select all

series first=0 series second=0 series third=0 series fourth=0 series fifth=0 smpl if revenue = @maxsby(revenue,country) first = 1 smpl if revenue = @maxsby(revenue,country, "if first=0") second = 1 smpl if revenue = @maxsby(revenue,country, "if first=0 and second=0") third=1 smpl if revenue = @maxsby(revenue,country, "if first=0 and second=0 and third=0") fourth=1 smpl if revenue = @maxsby(revenue,country, "if first=0 and second=0 and third=0 and fourth=0") fifth=1 smpl if first=1 or second=1 or third=1 or fourth=1 or fifth=1

[neptunhiker]

Thanks Gareth. Like always: quick and working answer. Maybe really not the shortest, but I can at least follow what's going on. If you had a solution which is more difficult to understand, but maybe also easier to apply to a bigger and more complex data set, then I'd be happy to see it as well. Thanks a lot.