Jan 9-20, 2023: High-Trying Offense
High attempt values, an open question on scoring, and highlights of the last two weeks of champions
Something in the Water
As anti-promised in the update two weeks ago, I was very much not committed to a weekly update last week. I was in Boston for the 2023 MIT Mystery Hunt, which is a good time for me personally, but not for this very nonprofessional Jeopardy! analytics endeavor. This two weeks was a good set to push together into one update, though, as this period has seen a surge in attempts that calls out for some investigation.
Since the start of the box score era, it hasn’t been unusual for the leader of a game to come in over 40 attempts. However, it has been unusual for second place (or even third place in some instances!) to reach that plateau. Not this month. Patrick Curran, Max Davison, Andrew Whatley, Kyle Daly, Kristina Mosley, Michael Cavaliere, Yogesh Raut, Katie Palumbo, Erin Portman, and Eric Kerr-Heraly all reached that number in losses. Yogesh Raut and Troy Meyer each posted at least 50 attempts in each of their games.
With just the data on hand, I don’t think it’s particularly possible to pinpoint why attempts are up this month, but some possibilities I’ve thought of include:
More knowledgeable contestants
More inherently aggressive contestants (perhaps knowing that attempts are tracked)
Contestants becoming more aggressive in response to the presence of 50+ attempters like Yogesh and Troy
More incorrect answers leading to more rebound attempts
Easier material
I mostly lean towards 1 through 3, as I don’t see evidence of changing incorrects, and there’s no reason to believe the calibration of material has suddenly changed. (I listed it mostly out of completeness.)
Diminishing Returns
The question bleeds into another one I’ve been considering after the last two weeks. Why isn’t anyone in the true upper right of this graph?
The upper edge of converting attempt value into score tracks parallel to the line of 100% conversion (where a player would be if they won every attempt and were correct) up through Amy Schneider’s best performances, but then trails off horizontally and even sloping down on more recent performances by Cris Pannullo, Troy Meyer, Ray Lalonde, Luigi de Guzman, and Yogesh Raut.
Two immediate concerns: First, there is not a lot of gameplay at this end of the attempt spectrum, so maybe this emptiness just doesn’t exist in actuality. (It would be great to have regular play attempt data on James Holzhauer and Matt Amodio.) Second, I remain concerned that my handling of re-attempts on incorrect responses is leading to an overestimate of attempt value, though a similar but less defined empty space does appear when using actual attempts directly.
It remains an interesting open question to me. Is Jeopardy! scoring inherently limited? Perhaps high-attempt play induces its own buzzer competition, and that buzzer competition allows competitors who would have been less sure and less likely to ring in to take chances and get in good guesses. Perhaps high-attempt play is built from experience in other quizzing culture where guessing is unpenalized and encouraged, and doesn’t have as strong an inhibitor on accuracy for when to buzz. (I would definitely buy something like this in the TOC, where answer accuracy was a strong indicator of performance.) Perhaps it’s just not physically possible to be consistently “on” with timing for fifty-seven clues, or to get the right interpretation and pull from that many J!-style-written clues. Perhaps it’s nothing.
Whatever it is, I’m interested in learning more if the high-attempt environment persists.
Champion Highlights
Connor Sears
Connor got in a runaway game and crush game before getting crushed in return by the steamrollers of Yogesh Raut and Andrew Whatley. Connor comes in near the bottom of Season 39 champions for most statistics, but that’s mostly because of that last game.
Yogesh Raut
Yogesh easily tops the attempt-based statistics but drops into “normal champion” numbers for buzzes and accuracy, earning three wins along the way. No one else has played quite like him so far in the dataset, so it’s not possible to tell what his streak could have been. Yogesh played a string of very strong games and yet never pulled off a runaway, and I take this as a reminder that there’s a ton of variance built in to Jeopardy! and long streaks take luck plus mastery of multiple skills that are not all related to knowledge.
To my eye, he had a perhaps difficult time with the buzzer, as he visibly used several different stances and techniques over his games. However, the timing metrics turned out in line with other champions, while it was the solo numbers that were lower than I expected. Looking over the individual games, the timing numbers actually vary quite a bit, which I am coming to interpret as “dependent on opponent skill” and therefore something of “median timing”. The solo numbers are part of why I started looking into the whole section of diminishing returns above, and I have to assume that’s a level of competition effect.
Katie Palumbo
Katie used excellent timing, competitive attempting, and zero incorrects to build a slight lead over Yogesh. Then she was able to seal the win with a good read on the wagering situation for FJ and a non-standard-cover bet that still would have resulted in wins for her on both triple get and triple stumper (as happened).
The second game was a little more down to earth, though I thought her DJ DD wager in her second game was a great move to try. Missing it pushed her into a different FJ situation where strategic betting did less for her, though, and she locked herself out of contention by betting small-ish when Vince also bet small.
Vince Bacani
Vince worked strong solo scores in both of his games, combining that with a strong timing game in his first to beat Katie. His second saw an exciting comeback in DJ after a missed true DD, including a successful near-true DD for the last one of the game, closing the gap back into a competitive position. It wasn’t quite enough, with both players getting FJ, but it got him where he needed to be in the moment.
Vince tops Season 39 champions on BuzV%, reflecting his strong solo scores, especially in DJ.
Erin Portman
Erin’s game with Vince was statistically interesting in that she also had a strong solo score, but hers was concentrated in the J round while Vince’s was in DJ. Two players with strong solos often reflect a battle of differing knowledge bases, particularly when they contrast so strongly between rounds. A good battle that worked out in Erin’s direction overall as the net effect of the DDs was small, though each individual DD loomed large.
Erin continued to put up champion numbers in her second game but found herself just locked out by Troy.
Troy Meyer
Troy is a high-attempt player, averaging over 50 per game, but although his Buz% is average among this season’s champions, his accuracy is high. In fact, it’s the second-highest AccV% after Martha Bath, which at the number and value of buzzes he’s getting is translating into a lot of scoring. This puts him at the top of average Buz$ for the season, but he’s only two games in.
Programming Notes
Somewhere in the middle of all this, Jeopardy! also announced Masters. Like Margaret Shelton said, it’s exciting to see my friends play each other again (and James, who I also only met briefly, and it was actually at the same time as Margaret did), and we’ll see what comes out of the show statistically. It is making me think more seriously about how to separate and display different forms of play from each other, especially due to the different levels of competition.
Celebrity is also continuing, with the third semifinal coming this week. Data will go onto the site for this as available. (I’m going to go put in Game 10 now.)