The obvious filter would have deleted what I was measuring

I was asking whether transfer spending concentrates in fewer clubs, and whether young players carry a growing premium. The club table in my source is incomplete: some deals name both clubs, some only the buyer or only the seller, some neither.

The obvious move is to keep the deals where both clubs are identified. That is 93.0% of the money, which makes it read as cheap hygiene — a rounding error traded for a clean join.

Before dropping anything, I counted what was in each block:

Block	Deals	18–23 share	Median fee
Both clubs identified	5,977	38.8%	2.2M€
Only the buyer identified	1,579	60.2%	0.5M€
Only the seller identified	406	31.8%	0.6M€
Neither	518	51.7%	0.3M€

The block I was about to discard is 60.2% players aged 18–23, at roughly a quarter of the median price. Those are the cheap purchases of young talent from clubs outside the covered competitions — which is the exact thing I had set out to measure. Dropping them would not have cleaned the data. It would have deleted the subject and left me with a confident answer about a different population, one that skews old and expensive by construction.

So the analysis runs on two declared universes instead of one: 6,716 deals for the price and age metrics, and the 5,109 with both clubs named for the metrics that have to attribute spend to a club by name. Every published figure states which one it came from. Nothing is thrown away; the scope narrows only where the metric itself demands it.

Missing data is harmless only when it is missing at random, and that is a claim you test, not one you inherit from how tidy the filter looks. Four counts and a median per block is the whole test. It cost one query.