APBRmetrics

Let's assume we have a team with an average starting lineup and an average bench. Now let's say they played against each other. Ballpark, what would the betting spread be?

I'd guess 6-8 pts would be a decent estimate of average quality difference, as measured by performance against other team lineups with adjustments for average opponent lineup quality. But that isn't the same as direct competition.

Betting spread? I dunno. Maybe higher.

Somebody might have looked at something related to or useful for this before but I am not searching for it. Maybe something from Crumpled Jumper or at bballindex or pbpstats.com would help.

At pbpstats, look at performance data for several actual near average starting lineups vs 0-2 opponent starters to get some values to average.

I looked at median starter / bench unit results as measured at hoopsstats.com and adjusted each by 1.5 to 2pts to get about an 8 pt difference. And then player data on DRIP (finding values for 75th and 225th best players then assuming median starter would be somewhat worse than that and median bench somewhat better. Difference ×5 players gets you in same ballpark). IF by average bench you meant 6th thru 10th man.

If you meant average bench including some use of deep bench players, the gap would be notably bigger.

5 on 5 ?

Yes. I was thinking starting 5 vs. 6-10.

Intuitively, I first started thinking in terms the worst teams in the NBA and whether their starters are better than the average bench. That made me think the spread could be quite large because even an average team will often be favored by 6-8 points over a terrible team.

Let's guesstimate: the average team has an average per-minute plus minus rating of 0.

This is made up of about 80 minutes of bench and 160 minutes of starters, so two thirds starters and one third bench.

So:

starter_pm*2/3 + bench_pm*1/3 = 0

giving:

starter_pm = -bench_pm/2

starter_pm - bench_pm = -1.5*bench_pm

So, now we just need to estimate what the rough quality of a bench is on the average team. Let's guess it's probably about the same as a very bad whole team in any given year, so around a -7 or something close ish to that. That gets us:

starter_pm - bench_pm = -1.5*-7 = 10.5

So, assuming an average game length is played, I would think something like 10.5 is about right.

Your calculation is straightforward and agreeable as is, though it involved the whole bench.

He said "I was thinking starting 5 vs. 6-10."
So trying to adjust it to vs. 6-10, your initial calculation would probably come close to mine and give a close enough consensus average estimate, to use as a background insight.

"At pbpstats, look at performance data for several actual near average starting lineups vs 0-2 opponent starters to get some values to average."

I circled back to my first suggestion.

(5 starters vs 2 for 3 median teams had wildly different actuals. The average of the 3 was actually a big loss. But tiny samples. I didn't check situation leverage / "motivation", something discussed elsewhere that might also help explain the worse than expected results for the 3 median and the only slight win in this situation across 30 teams / about 800 minutes. But this was only mentioned to try to help indirectly estimate results for bench lineups composed strictly of 6-10th men, so in the end mostly or entirely ignore this paragraph, except to note I did follow up on my own suggestion.)

Now, closer to the question, 5 on 0 was very rare, only 111 minutes for whole league. But the average margin (looking at results for 5 on 0 - 0 on 5) was 7.5 points. This involves some degree of deep bench. Against strictly 6 - 10th guys it logically might be closer to 6 on average, the lower end of my previous estimate. But motivation is still an issue.

If this wasn't game data but rather a practice challenge, maybe the starters would play harder. Or maybe they would care about the same as in real games or less. Ultimately there isn't a perfect answer to the original hypothetical question. But a convergence of estimates from different approaches.

In expansion drafts, 8 players are typically protected; with some younger players protected over older ones, it means expansion teams should basically be a team of all 8th men - which is around the average of 6-10. Maybe slightly worse, as they dont get any 6th men, but should be pretty close.

Last 7 expansion teams averaged -7.73 SRS,; if we assume that is average bench, then starters playing 2/3 of minutes versus this bench solves for 3.86 for the starters.

So maybe a 12 point spread?
And if average bench 6-10 is mimus 8, then a real good bench would be better than the worst teams.

An interesting note:

The differences between good teams and bad teams are almost 100% their starters.

End of bench players (replacement level players) are generally similar between good teams and bad teams.

If you plot MPG vs. player all-in-one metric, the curves for good teams and bad teams stay pretty similar up to starter level minutes. Then on the top teams the stars are far above a typical MPG vs. Player Metric relationship.

Thanks for the insights.

The answers were better than I hoped for.