Underneath the Box Score Hood – the Hometown Assist
By Seth Partnow
Nov 18, 2013; Los Angeles, CA, USA; Los Angeles Clippers point guard Chris Paul (3) drives the lane on Memphis Grizzlies point guard Mike Conley (11) during the sec on half at Staples Center. Mandatory Credit: Robert Hanashiro-USA TODAY Sports
A topic to which I return with some frequency is the degree to which statistical analysis of basketball is impacted by imperfect input data. As the expression goes, “garbage in, garbage out.” No matter how much care is put into building a model, flawed inputs lead to misleading outputs. Especially in terms of box score stats, traditional numbers can be pretty good, but don’t capture what happens on the floor with anything resembling perfect accuracy.
We know this already, and have had many practical demonstrations: all rebounds aren’t created equal; shot blocking and rim protection aren’t totally synonymous; differentiating between turnovers and steals is important, and so on. Even on such topic of plus/minus-based one-number metrics, the input data is messy. The degree of randomness in made or missed jump shots by teammates or opponents can easily skew the results. It is hoped with sufficient sample size, this evens out. And on aggregate, it probably does. But in evaluating players or specific teams, we aren’t dealing in aggregates. None of the above invalidates the whole process, but it should operate as a reminder that for all the numeric exactitude of the result, that result remains an estimate.
Almost nowhere is this problem on better display than the assist stat. One of the things to come out of last week’s looks at individual shot creation[1. also here and here] was just how variable and often generous the award of an assist can be:
Over half of shots made after the shooter possessed the ball for two seconds and took three dribbles saw an assist awarded during the 2014/15 regular season[2. Per SportVU data, though even this “objective” data can be imperfect when the system’s algorithms very occasionally struggle to accurately measure touch time and dribbles during quick passing movement into and out of traffic.]. That seems…generous. By comparison, the public SportVU data only awards “assist opportunities” for one dribble or less and two seconds or less of touch time prior to the shot. Given the frequency of assist awards outside of those parameters, this introduces a certain bias to analysis of playmaking efficiency, as certain shots only “count” if made, much like how it only counts as a shot attempt if a fouled player makes the shot.
Still, if official scorers are universally generous, what’s the big deal? Well, they aren’t. For example, while he was in New Orleans, Chris Paul was notoriously awarded some rather generous assists at home. This is somewhat generalizable, there exists an apparent bias towards the home team on more questionable assists. Looking at percentage of made field goals just by touch time[3. As the chart shows, touch time appears at first pass to be much more of a determinant that number of dribbles.]:
The effect isn’t huge, and there are many plausible alternative explanations than just hometown scorekeeper bias[4. Part of team’s “playing better” at home could be crisper execution of sets leading to better ball movement and thus slightly more playmade shots, leading to both
and slightly more assisted shots.], but the home team, or more accurately the home team’s players, get the benefit of more assist awards. Sill, since teams play 41 games on the home and road, no big deal. It evens out, right? Not so much:
By comparing expected assists based on touch time with actual assists awarded both home and away[5. Which should control for variations in team offensive styles to a degree.], very different patterns emerge for each team. I don’t want to toss charges of intentional bias around, as this is very 10,000 foot level analysis, as there could easily be categorical differences in the plays leading up to made baskets home or on the road. For example important injuries could easily fall during an especially travel-heavy portion of the season, or trades could have happened where a team had played more road games already.
Besides, it’s not that big a deal in the real world. A few extra assists here or there matter far more to the chattering classes than it does in the actual outcome of games. Further, if a team is systematically overrating its own players with generous scoring decisions, they are only hurting themselves. If management buys into a player being better because of inflated totals, the team is set up for disappointment, while if the league as a whole buys in, the free agent price tag goes up, forcing the team to either overpay or let the player walk.
However, for the purposes of statistical analysis, it does matter. If Paul’s assists are worth less than Mike Conley’s because of the relative generousness of the awards, that impacts a comparison between the two. At the very minimum it’s just another reminder that there remains a great deal of uncertainty in terms of measuring basketball impacts.