At a glance, it seems like we're saying "this is the player's stat line just against the opposing starter, broken down by that starter's handedness." In which case it would seem to be a bug when a guy has bigger totals against, say, lefty starters than against lefties in total.
Except that the definition I just wrote is not at all what that split is measuring. Instead, the "vs LH/RH Starter" split adds up all stats accumulated in games where the opposing starter was of a certain handedness, INCLUDING STATS ACCUMULATED LATER IN THE GAME WHEN THE STARTER IS PULLED, REGARDLESS OF THE RELIEF PITCHER(S)' HANDEDNESS.
Despite bolding, italicizing, and going all-caps, I still don't think I emphasized that enough. I realize the description of the split seems like it's talking only about stats accumulated against the starters, but it's really just counting up all stats in games where the opposing starter threw a certain way -- a BIG difference. If you want to know about performance against just starters of a given handedness... well, that's a double split, so we can't answer that right now. But we do hope to add the capacity for double splits in the future.
Anyway, now you might be saying, OK, well what's the point of that split, then? It seems pretty useless. And in some ways it is, for recent seasons at least -- but it's the only way we can approximate platoon splits for pre-PBP era players (1940s and before). Even going back to 1916, we at least know who started each game and their handedness, which usually is good enough to get platoon data on the majority of PAs (especially before managers started aggressively using relievers).
So that's why that split is there, even if it's not especially useful for 2013 players. And now you know what that split means, even if it seems like it should mean that other thing (which it DOESN'T).
Added 2-pointers made/attempted & 2P% to stat tables. - Now you can see that Shaqdid in fact shoot much better on 2-pointers (58.3%) than 1-pointers (52.7%) in his career.
Added Free Throw Rate to advanced stats. - Borrowing from my old Blog-tionary... "FTr = (FTA / FGA). Free throw rate is the ratio of free throws attempted per field goal attempted. It's useful to assess where the bulk of a player's shots are coming from -- low ratios usually indicate a lot of jump shots (fouls are less prevalent the further away from the basket you get), while high rates are the province of inside players who often take strong shots in traffic down low. A variation on FTr is Bob Bellotti's old Rice-Scott Index (RSI), which is just FGA / (FGA + FTA) (so named because when the stat was developed, long-range bombers Glen Rice and Dennis Scott had extremely high RSIs)."
Added 3-Point Attempt Rate to advanced stats. - 3PAr = (3PA / FGA), and is a measure of what % of a player's shots come from long-distance, another good gauge of how they're utilized offensively. Note that 3PAr and FTr also appear on team & league pages (so you can see what the league averages were).
Added Plus/Minus (and re-added Game Score) to player gamelog pages. - Plus/Minus represents the team's scoring differential while the player was on the floor, and is available going back to the 2000-01 season. Game Score is a John Hollinger method of evaluating single-game performances.
Added Advanced Stats to player splits pages. - Splits pages now feature the player's True Shooting %, Usage %, Dean Oliver individual Offensive/Defensive Ratings, and Plus/Minus per 100 possessions in each split.
Added franchise summaries under player stat tables. - Underneath each player's stat tables, you'll now see career totals & averages for every franchise and league they played for/in (when applicable).
A few more tweaks are forthcoming before the season, and as always, email us any additional ideas/suggestions/requests you might have, and we'll see what we can do.
After the introduction of SV%+ to goalie stats a little over a week ago, there was some discussion as to whether or not it was the best way to represent goals allowed relative to league average. After reviewing some poll results, we've decided to change things up a bit and switch the representation to Goals Allowed %- (the minus because lower is better). The formula is now 100*((1-player save %)/(1-league average save %)) -- this means that Semyon Varlamov, who currently has the best save percentage in the league at .965 (vs. league average of about .913), gets a 40 GA%- &emdash; on a per-shot basis, he allows goals at 40% of the league average rate. Conversely, Martin Biron has allowed 9 goals on 38 shots, which gives him a GA%- of 273 (yikes). This works out better for us because goalies who have not allowed a goal will get a 0 GA%- instead of getting an undefined value as they would have in the old system.
"This query should return this game [...] Elvis Grbac threw for 504 yards, and the Raiders scored 49 points. That game was not returned by the query[.]"
In both cases, it's a question of individual passing totals seemingly not matching up to team totals. And that's because the NFL doesn't define "passing yards" in the same way for teams as it does for individuals.
Confusingly, sack yards do not count against the "passing yards" you see listed next to individual players, but they do count against "passing yards" at the team level. To put it another way, individual passing yards are always presented as GROSS yards (with sack yards not subtracted out), while team passing yards are always presented as NET yards (meaning sack yards are subtracted out).
So despite Morton & Starr's combined GROSS pass yards, there were also 114 combined sack yards in that game, leading to a record -11 total combined NET passing yards in the game. And despite Grbac's 504 GROSS pass yards, he was sacked 4 times for 30 yards, meaning KC only had 474 NET passing yards in the game.
This probably doesn't make a great deal of sense in 2013, but it does come in handy for years before QB sack data was tracked (a.k.a. the history of football up until 1969). Because of this practice, we know how many gross yards a quarterback -- and, therefore, a team -- had, as well as how many net yards they had, since sack totals were recorded for teams (but not players) before 1969. It would make for even more confusion if, starting in 1969, we all of a sudden began defining yardage for individuals differently than it had been defined in the past (though as we saw yesterday, this didn't stop the NBA from abruptly changing how team rebound totals were defined).
So it's a minor inconvenience now, but probably a necessary one from an historical perspective. And hopefully this post will reduce some of the confusion going forward (though I doubt it).
Recently, a BBR user posed a question we get on occasion:
"[I] noticed that the team rebound totals for every team appear to be incorrect from 1956-57 to 1967-68. They are correct in 1968-69. For example in 1956-57 it shows the Celtics led the league in rebounds with 4963. When you add up the rebounds for every player on their roster the total comes to 4578. The points and assists totals match exactly. [...] It is only the rebound total that is off, and it is off for every team by a significant margin."
He's right -- you can see this by adding up the team rebound totals for each league-season and comparing it to the sum of individual player rebounds... It changes around 1968:
| year_id | lg_id | team_trb | player_trb |
| 2013 | NBA | 103575 | 103575 |
| 2012 | NBA | 83513 | 83513 |
| 2011 | NBA | 101816 | 101816 |
| 2010 | NBA | 102640 | 102640 |
| 2009 | NBA | 101586 | 101586 |
| 2008 | NBA | 103271 | 103271 |
| 2007 | NBA | 100994 | 100994 |
| 2006 | NBA | 100754 | 100754 |
| 2005 | NBA | 102970 | 102970 |
| 2004 | NBA | 100361 | 100361 |
| 2003 | NBA | 100604 | 100604 |
| 2002 | NBA | 100829 | 100829 |
| 2001 | NBA | 100988 | 100988 |
| 2000 | NBA | 102062 | 102062 |
| 1999 | NBA | 60395 | 60395 |
| 1998 | NBA | 98798 | 98798 |
| 1997 | NBA | 97703 | 97703 |
| 1996 | NBA | 98099 | 98099 |
| 1995 | NBA | 92006 | 92006 |
| 1994 | NBA | 95192 | 95192 |
| 1993 | NBA | 95504 | 95504 |
| 1992 | NBA | 96680 | 96680 |
| 1991 | NBA | 95776 | 95776 |
| 1990 | NBA | 95518 | 95518 |
| 1989 | NBA | 90031 | 90031 |
| 1988 | NBA | 81828 | 81828 |
| 1987 | NBA | 83020 | 83020 |
| 1986 | NBA | 82161 | 82161 |
| 1985 | NBA | 82008 | 82008 |
| 1984 | NBA | 81150 | 81150 |
| 1983 | NBA | 83853 | 83853 |
| 1982 | NBA | 81987 | 81987 |
| 1981 | NBA | 82010 | 82010 |
| 1980 | NBA | 81065 | 81065 |
| 1979 | NBA | 81564 | 81564 |
| 1978 | NBA | 84984 | 84984 |
| 1977 | NBA | 84886 | 84886 |
| 1976 | NBA | 69960 | 69960 |
| 1976 | ABA | 30815 | 30815 |
| 1975 | NBA | 69468 | 69468 |
| 1975 | ABA | 40239 | 40239 |
| 1974 | NBA | 67230 | 67230 |
| 1974 | ABA | 40705 | 40705 |
| 1973 | NBA | 70555 | 70555 |
| 1973 | ABA | 41062 | 41062 |
| 1972 | NBA | 71299 | 71299 |
| 1972 | ABA | 48429 | 48429 |
| 1971 | NBA | 74059 | 74059 |
| 1971 | ABA | 50234 | 50234 |
| 1970 | NBA | 60697 | 60697 |
| 1970 | ABA | 49991 | 49991 |
| 1969 | NBA | 65324 | 65324 |
| 1969 | ABA | 46674 | 46674 |
| 1968 | NBA | 65166 | 56999 |
| 1968 | ABA | 46957 | 46957 |
| 1967 | NBA | 54536 | 48095 |
| 1966 | NBA | 49118 | 43250 |
| 1965 | NBA | 48433 | 42563 |
| 1964 | NBA | 47423 | 41894 |
| 1963 | NBA | 48044 | 41418 |
| 1962 | NBA | 51415 | 43183 |
| 1961 | NBA | 46314 | 40732 |
| 1960 | NBA | 44104 | 37401 |
| 1959 | NBA | 40343 | 34794 |
| 1958 | NBA | 41279 | 35814 |
| 1957 | NBA | 35948 | 32136 |
| 1956 | NBA | 34616 | 31395 |
| 1955 | NBA | 32292 | 28777 |
| 1954 | NBA | 32987 | 29808 |
| 1953 | NBA | 36168 | 32314 |
| 1952 | NBA | 35977 | 31209 |
| 1951 | NBA | 35019 | 30621 |
So what's the deal? Is this a bug in our database?
Actually, no. It's just a matter of accounting by the official scorer, because technically speaking, for every missed shot there has to be a rebound.
Starting in 1968-69 (1967-68 in the ABA), so-called "team" rebounds (rebounds where there was no clear individual who should receive credit) were no longer counted toward a team's overall rebounding total. Before that, "team rebounds" were credited at the team level, but they (obviously) didn't make it into individual rebounding totals. This is what caused the massive discrepancies between teams' rebounds and the sum of their individual players' rebounds.
Nowadays, those "team rebounds" are thrown into their own bucket, neither allocated to individuals nor to teams for the purpose of overall stat totals.
With the League Championship Series starting tonight, it's time to check out the Baseball-Reference Play Index if you haven't already signed up. What's the Play Index? It's a set of research tools that allow you to create customizable queries on our database, save the results, and share them with others. Using the PI, you can:
Search full-season or multi-year totals to find your own custom leaderboards - Look at the entire history of baseball from 1871-2012 with every year, team, and position available, or filter the results in a vast number of ways: by specific years, by age, by first six seasons or last ten seasons, by American League only, by Cubs only, by switch-hitters, by catchers, by outfielder or infielder, by year of debut, but active or retired, by Hall of Famer, by height and weight, by living or deceased, or by a range of common statistical categories. Then sort the results by any common statistic, by the teams with the most players matching that category, by players with the most seasons matching that category, or by most recent, youngest, oldest, final year, or year of debut, and others.
Search the records of a specific player - Output a detailed summary and play-by-play list of all events of a specific type from a single year or an entire career. For example, you can see all of Harmon Killebrew's triples or even his outs to the second baseman.
Search Batter vs. Pitcher Matchups - This tool presents a complete sortable list of batter or pitcher with totals for every opponent they faced by career or by year. Clicking on the player's name will lead you to a detailed output of their head-to-head plate appearances.
Personal Subscriptions to the Play Index still cost just $36 for a year, $6 for a month, or $2 for 24 hours. Subscriptions may only be used by a single user, and there are discounts for users sponsoring at least $35 in pages.
Organizational Subscriptions can be set up for either an unlimited number of users ($600/year), or for up to five users ($125/year).
There are Two Steps to Subscribe to the Play Index:
In case you missed it yesterday, we rolled out a couple of new goaltending stats for the 2014 season -- Save Percentage+ and Goals Saved Above Average. Overnight we got some feedback about SV%+ in particular, and the structure of having 1 minus the goalie's own SV% in the denominator of the formula. Right now, we are measuring the rate at which the league allowed more goals than the player, where 100 = average. So in the case of a 133 SV%+, the league's rate of goals allowed per shot would be 33% higher than the player's rate.
There are other alternatives. If we re-arrange the formula so that 1 minus the league SV% is in the denominator, we would be measuring the rate at which the player allowed fewer goals than the league. (Because of the nature of division, this is an important distinction from the definition in the previous paragraph.) If we tweak the formula in this direction, there are 2 options: we can compute a "minus" type of stat in which lower numbers are better, or we can stick with the "plus" style familiar to all from baseball's OPS+ and ERA+. The former would represent a goalie who allows 33% fewer goals than the league as a 67 (remember, lower is better); the latter would represent that as a 133.
As is the case sometimes in this business, there is no "right answer" here, but rather a matter of preference. For instance, we've long assumed users would find it counterintuitive to have a rate stat where lower numbers = better performance, but maybe that's not true. I'm interested in opinions on this, so let your voice be heard in the following poll -- which format do you prefer? Or do you care at all? I can't promise the poll-winning format will be what we eventually stick with, but I'd like to know what people think.