[Nevermark]

You are right.

Actual average returns are simply total cap at one time divided by total cap at a previous time: TC(t2)/TC(t1)

That probably is easy enough to do.

But an estimate of that would be an average of returns for randomly chosen companies, but with either:

1) straight random selection of companies, with the individual returns weighted by their company size, or ...

2) ... random selection of companies, with probabilities of each company weighted by size, then the unweighted average of those returns.

So you are right, one way or another a subsample of companies needs to weight companies by size to account for the greater levels of investment in large companies.