Smoothbutta wrote:Thanks I then presume that as long as you are consistent then using at either one is just as reputable
Hey Smoothbutta! Both are
reasonably reputable, and indeed consistency is probably the most important choice. That said, pbpstats is certainly slightly more reputable. Along with nba.com (which has terrible user interface and isn't very searchable for things like on/off etc.), pbpstats is probably the most reputable source for plus minus data.
BR:
Like Colbinii said, basketball reference uses an estimate of possessions (rather than true possessions) for certain stats, whereas pbpstats uses true possessions. BR's possession estimates (calculated using box stats) are reasonably accurate (they're fine if you keep the larger error bars in mind), but you do get occasional sticklers here who reject it if it's not 100% accurate like pbpstats.
One area I'm not sure of that might change things: I know when calculating Basketball Reference's VORP, for example, it assumes the team pace does not change when a player is on the court vs when they're off the court. It assumes the team plays at a constant pace. This is reasonably accurate, but does introduce some error in something like VORP. I'm not sure if BR does the same thing for on-off, but it wouldn't shock me. This might make a difference for someone like Curry; the Warriors frequently slowed the pace when Curry wasn't on, so his 'off' numbers would actually look worse if you used true possessions instead of assuming a constant pace.
pbpstats:
Like Colbinii said, there are some edge cases that make it hard to 100% tell what a possession is. Different websites have different definitions. You can find the article on pbpstats talking about its possession definition here:
https://darrylblackport.com/posts/2019-04-03-why-pbpstats-possession-counts-lower/One thing to be aware of: you might get slight differences between true 'on-off' and 'On - Off Net RTG'. They're very close, but a slight difference in units while you do the calculation can change the numbers slightly.
True on-off: Plus-Minus (ON) - OFF = (point differential when player is on)*(100 possessions)/(total possessions when on) - (point differential when player is off)*(100 possessions)/(total possessions when off).
-To find this on pbpstats, you can go to pbpstats.com -> On/off in the header -> 'Wowy' (with or without you; note Wowy is also used to refer to a different stat.) -> search for your players' Plus minus ('P/M) when they're ON and when they're OFF.
-Then use the formula above:(ON P/M)*100/(total ON possessions) - (OFF P/M)*100(total OFF possessions). This will give you a player's On-off exactly.
On - Off Net RTG: (ORTG - DRTG when ON) - (ORTG - DRTG when OFF). The difference here from above is that when calculating ORTG and DRTG, we normalize the pace on both sides of the court first, so we assume the offense and defense has the same pace. This is close to reality, but there are times when one team gets a few more offensive possessions than the other, so the true pace of the offense and the defense isn't always exactly the same. Consider if one team gets lots of offensive rebounds that are considered different possessions, or if one team gets lots of steals before the other team's possession starts, or if a player gets substituted on the court more frequently during offensive or defensive possessions, etc. True on-off doesn't make this assumption. So True on-off and on-off NetRTG are similar -- they just make measure slightly different quantities in how they deal with team pace.
-To find this on pbpstats, you can go to pbpstats.com -> On/off in the header -> 'Wowy Combinations' -> look up your player.
Then use the formula (NetRTG when ON) - (NetRTG when OFF). If you'd 'like to look at a player across multiple years, you can also:
go to pbpstats.com -> search your player in the search bar (top right) -> go to on-off in the Table Data dropdown menu -> look at the 'On-Off NetRTG'
...
On the topic of your thread, thanks for the clarification! It looks like the other players are reasonably close to true, although the chart value still does differ slightly. Take Dirk for instance: his Basketball Reference on-off during the timespan should be +10.0 for Ages 20-36, whereas the new chart shows him around +9.5. They're reasonably close like you say, enough to get a sense for how the players are, but the value on the chart may differ slightly from BR's value, or pbpstats' true value.
Anyways, cool chart! It's interesting to see everyone plotted all together.
