I dont have any links, but have been making this same claim based on the available evidence. That the media value of games goes up exponentially (not linearly) with viewership.One thing that isn't really clear how media value works is how number of games x viewers compares. In other words, I doubt 8 CBB games at 500k viewers each is the same value as one CFB game at 4 million viewers. I don't know if there's been some good explanations of this, but I suspect the latter has way more unique vs. repeat viewers.
So I don't think it's quite that the viewership can be 1/3 or so of a CFB game to have the same relative value.
But I think your point remains, and is what I'm after. When people say that basketball doesn't matter, they are wrong. Of course it is not as valuable as CFB, and a basketball brand alone isn't going to provide enough value to justify addition by the Big 10 or SEC, that are getting a ton of $/team because of the likes of Ohio State, Michigan, Alabama, Georgia, OU and UT. But for a league like the Big 12 at the current payout, it's conceivable that a low-value football brand but high-value basketball brand like UConn or to a less extreme extent Arizona MAY make sense. I'm not saying it is the case, but I think it is possible. We'll know a lot more about the relative value of MBB vs. FB if the Big 12 splits them out.
The evidence is that the bigger games are paid for way higher. I dont have the stats on my phone, but from memory- CCGs w appx 5-10m viewers bring in $10-20m. Big bowl games similar, NY6/CFP averages 10-20m viewers per game and was bringing in $470m(?) for 6 games. Compare these to average games (take the avg value of all big12 fb games at ~$2m on the new contract).
The Super Bowl would be the ultimate top end example, hundreds of millions of viewers, and a Billion$ value for one game.
On yhe other end, a low end game drawing 250k viewers... you could rerun Gilligans Island or Professional Basket Weaving and get the same viewership, and it would cost peanuts for the rights.
I could prob come up w a formula, but with limited data points, and changing values over time, it would be pretty low r^2.