I have data like this:

date  item sold nusers

d1     i1     5     1

d1     i2     3     1

d1     i3     1     1

d2     i1     2     2

d2     i4     4     2

Here d1, d2... are dates,  i1, i2... are names of items, sold is the sales for items of the particular type, and nusers is the number of users of the product on that day. By definition, nusers is the same for all the rows for a particular date.

I would like to compute the average items sold per user over a time period: i.e. (5+3+1 + 2+4) / (1 + 2) = 5

Or, in another word, I'd like to be able to first average the nusers column for each date, and then sum up over the different dates.

If I only have data for one day, I could get the average items sold per day as SUM(sold)/AVG(nusers), or equivalently in this case, SUM(sold)*SUM(nusers)/COUNTD(item)/COUNTD(date).

However, with data for several days, where the items sold may change from day to day, this computation will break. Is there a better way to create such a computation (or to alter my input data) without padding the data with all items in the database for each day of data?