Compilation of Michael Jordan’s on-off data with the Bulls

[lessthanjake]

I have just updated the OP to include Squared’s sample for the 1991-1992 regular season. This basically just adds an extra 17 games that Squared tracked for that season above and beyond the Dipper tracking that I’d previously been using in this thread. We now have data for 70 of Jordan’s games in that season, instead of 53 games. The overall numbers with this change remain very similar. The 1991-1992 regular season on-off is a bit lower, but the “on” value for that season is now higher, and the total numbers across all data now has a slightly higher “on,” slightly better “off,” and very slightly better on-off.

Great stuff!

I don't want to give you more work but I really think you should update the playoff numbers we got on the spreadsheet and update with the new regular season numbers from Squared too. For 1996, you can keep using the Pollack data I guess since both sets are complete.

Playoffs

All runs from 1988-1996 have slightly different tallies.



1986-1987 Regular Season (40 games)
ON: +71 1622 minutes
OFF: -80 318 minutes

1992-1993 Regular Season (79 games)
ON: +548 3003 minutes
OFF: -98 819 minutes

1995-1996 Regular Season (82 games)
ON: +988 3090 minutes
OFF: +16 856 minutes

Another thing that comes to mind that I think you mentioned...

Should we just add the two games MJ missed (one in 1992 and one in 1993) to the OFF samples for those seasons? I haven't done it yet but it makes sense. We know that 48 minutes for those games are part of the OFF sample so I see no reason not to.

I may be blanking on something I should know the answer to, but where do those slightly different playoff tallies come from? My OP has a lot of explanation of exactly where all the data comes from, so if I were to make changes, I’d want to be able to explain exactly where the new tallies came from. And is there anything I can link to for the Squared regular season data you list above? I could potentially just link your post, but am curious if there’s something directly from Squared that could be linked to for that.

As for the stray missed games, yeah they could easily be included. I’ve basically just included games if they’re directly in the data source I’m using. It’s a somewhat arbitrary decision, though, since I don’t need someone to have tracked a game Jordan missed to know the relevant data for purposes of this thread. That said, the treatment of missed games is a bit of a tricky issue, because I’m not including 1995 games before he came back (nor do I think I would include the games in 1986 that he didn’t play in, if we got his regular season plus-minus data for the games he did play that year). Those seem materially different than random missed games (1995 even moreso, because he wasn’t even in the NBA), but arguably if I’m not including those then I shouldn’t be including any missed games at all and the current inclusion of one missed game in the 1992 regular season is perhaps a bit sloppy if that should be the rule. I don’t really feel strongly about any of it. A significant issue with 1995 (and 1986 if we actually had regular season data for Jordan for that year too) is it’d just get way outsized weight in the off sample of the overall data, making the off sample not actually representative. Of course, I think some would like that and would see not including 1995 non-Jordan games in the off sample as being biased in favor of Jordan, so it’s definitely a loaded issue.

[Djoker]

I have just updated the OP to include Squared’s sample for the 1991-1992 regular season. This basically just adds an extra 17 games that Squared tracked for that season above and beyond the Dipper tracking that I’d previously been using in this thread. We now have data for 70 of Jordan’s games in that season, instead of 53 games. The overall numbers with this change remain very similar. The 1991-1992 regular season on-off is a bit lower, but the “on” value for that season is now higher, and the total numbers across all data now has a slightly higher “on,” slightly better “off,” and very slightly better on-off.

Great stuff!

I don't want to give you more work but I really think you should update the playoff numbers we got on the spreadsheet and update with the new regular season numbers from Squared too. For 1996, you can keep using the Pollack data I guess since both sets are complete.

Playoffs

All runs from 1988-1996 have slightly different tallies.



1986-1987 Regular Season (40 games)
ON: +71 1622 minutes
OFF: -80 318 minutes

1992-1993 Regular Season (79 games)
ON: +548 3003 minutes
OFF: -98 819 minutes

1995-1996 Regular Season (82 games)
ON: +988 3090 minutes
OFF: +16 856 minutes

Another thing that comes to mind that I think you mentioned...

Should we just add the two games MJ missed (one in 1992 and one in 1993) to the OFF samples for those seasons? I haven't done it yet but it makes sense. We know that 48 minutes for those games are part of the OFF sample so I see no reason not to.

I may be blanking on something I should know the answer to, but where do those slightly different playoff tallies come from? My OP has a lot of explanation of exactly where all the data comes from, so if I were to make changes, I’d want to be able to explain exactly where the new tallies came from. And is there anything I can link to for the Squared regular season data you list above? I could potentially just link your post, but am curious if there’s something directly from Squared that could be linked to for that.

As for the stray missed games, yeah they could easily be included. I’ve basically just included games if they’re directly in the data source I’m using. It’s a somewhat arbitrary decision, though, since I don’t need someone to have tracked a game Jordan missed to know the relevant data for purposes of this thread. That said, the treatment of missed games is a bit of a tricky issue, because I’m not including 1995 games before he came back (nor do I think I would include the games in 1986 that he didn’t play in, if we got his regular season plus-minus data for the games he did play that year). Those seem materially different than random missed games (1995 even moreso, because he wasn’t even in the NBA), but arguably if I’m not including those then I shouldn’t be including any missed games at all and the current inclusion of one missed game in the 1992 regular season is perhaps a bit sloppy if that should be the rule. I don’t really feel strongly about any of it. A significant issue with 1995 (and 1986 if we actually had regular season data for Jordan for that year too) is it’d just get way outsized weight in the off sample of the overall data, making the off sample not actually representative. Of course, I think some would like that and would see not including 1995 non-Jordan games in the off sample as being biased in favor of Jordan, so it’s definitely a loaded issue.

The playoff revision comes from my own tracking. I basically tracked all Jordan playoff games from 1986-1996 at some point. That's how the spreadsheet got completed.

Here is the 1993 regular season updated from Squared2020's tweet - 79 games:

Read on Twitter

The 3 games that aren't logged are:

Read on Twitter

Here is the 1987 regular season updated from another tweet - 40 games:

Read on Twitter

And the 42 games that aren't logged are:



Those are the sources..

As for 1986 and 1995, it makes sense to do it either way. As long as people understand the issue of uneven samples and don't just combine the off sample from 1995 for instance (which is huge!) to other off samples in neighbouring seasons. That would be problematic but otherwise it's ok.

[lessthanjake]

Great stuff!

I don't want to give you more work but I really think you should update the playoff numbers we got on the spreadsheet and update with the new regular season numbers from Squared too. For 1996, you can keep using the Pollack data I guess since both sets are complete.

Playoffs

All runs from 1988-1996 have slightly different tallies.



1986-1987 Regular Season (40 games)
ON: +71 1622 minutes
OFF: -80 318 minutes

1992-1993 Regular Season (79 games)
ON: +548 3003 minutes
OFF: -98 819 minutes

1995-1996 Regular Season (82 games)
ON: +988 3090 minutes
OFF: +16 856 minutes

Another thing that comes to mind that I think you mentioned...

Should we just add the two games MJ missed (one in 1992 and one in 1993) to the OFF samples for those seasons? I haven't done it yet but it makes sense. We know that 48 minutes for those games are part of the OFF sample so I see no reason not to.

I may be blanking on something I should know the answer to, but where do those slightly different playoff tallies come from? My OP has a lot of explanation of exactly where all the data comes from, so if I were to make changes, I’d want to be able to explain exactly where the new tallies came from. And is there anything I can link to for the Squared regular season data you list above? I could potentially just link your post, but am curious if there’s something directly from Squared that could be linked to for that.

As for the stray missed games, yeah they could easily be included. I’ve basically just included games if they’re directly in the data source I’m using. It’s a somewhat arbitrary decision, though, since I don’t need someone to have tracked a game Jordan missed to know the relevant data for purposes of this thread. That said, the treatment of missed games is a bit of a tricky issue, because I’m not including 1995 games before he came back (nor do I think I would include the games in 1986 that he didn’t play in, if we got his regular season plus-minus data for the games he did play that year). Those seem materially different than random missed games (1995 even moreso, because he wasn’t even in the NBA), but arguably if I’m not including those then I shouldn’t be including any missed games at all and the current inclusion of one missed game in the 1992 regular season is perhaps a bit sloppy if that should be the rule. I don’t really feel strongly about any of it. A significant issue with 1995 (and 1986 if we actually had regular season data for Jordan for that year too) is it’d just get way outsized weight in the off sample of the overall data, making the off sample not actually representative. Of course, I think some would like that and would see not including 1995 non-Jordan games in the off sample as being biased in favor of Jordan, so it’s definitely a loaded issue.

The playoff revision comes from my own tracking. I basically tracked all Jordan playoff games from 1986-1996 at some point. That's how the spreadsheet got completed.

Here is the 1993 regular season updated from Squared2020's tweet - 79 games:

Read on Twitter

The 3 games that aren't logged are:

Read on Twitter

Here is the 1987 regular season updated from another tweet - 40 games:

Read on Twitter

And the 42 games that aren't logged are:



Those are the sources..

As for 1986 and 1995, it makes sense to do it either way. As long as people understand the issue of uneven samples and don't just combine the off sample from 1995 for instance (which is huge!) to other off samples in neighbouring seasons. That would be problematic but otherwise it's ok.

Okay, thanks! That all makes sense. Do you have any link to where Squared has all 82 games for 1996 (the above stuff you linked from Squared does include some 1996 stuff, but it looks like he only had 78 games at the time he posted those)? I’d probably prefer to use Squared’s version for that than the Pollack data that’s there now, because I believe the Pollack data is just contained in a random spreadsheet someone made at some point, rather than actually being pulled from the original data source. The numbers are very similar so it doesn’t really matter much, but they’re not exactly the same and if I can avoid using that spreadsheet for something, I’d prefer that.

[Djoker]

I may be blanking on something I should know the answer to, but where do those slightly different playoff tallies come from? My OP has a lot of explanation of exactly where all the data comes from, so if I were to make changes, I’d want to be able to explain exactly where the new tallies came from. And is there anything I can link to for the Squared regular season data you list above? I could potentially just link your post, but am curious if there’s something directly from Squared that could be linked to for that.

As for the stray missed games, yeah they could easily be included. I’ve basically just included games if they’re directly in the data source I’m using. It’s a somewhat arbitrary decision, though, since I don’t need someone to have tracked a game Jordan missed to know the relevant data for purposes of this thread. That said, the treatment of missed games is a bit of a tricky issue, because I’m not including 1995 games before he came back (nor do I think I would include the games in 1986 that he didn’t play in, if we got his regular season plus-minus data for the games he did play that year). Those seem materially different than random missed games (1995 even moreso, because he wasn’t even in the NBA), but arguably if I’m not including those then I shouldn’t be including any missed games at all and the current inclusion of one missed game in the 1992 regular season is perhaps a bit sloppy if that should be the rule. I don’t really feel strongly about any of it. A significant issue with 1995 (and 1986 if we actually had regular season data for Jordan for that year too) is it’d just get way outsized weight in the off sample of the overall data, making the off sample not actually representative. Of course, I think some would like that and would see not including 1995 non-Jordan games in the off sample as being biased in favor of Jordan, so it’s definitely a loaded issue.

The playoff revision comes from my own tracking. I basically tracked all Jordan playoff games from 1986-1996 at some point. That's how the spreadsheet got completed.

Here is the 1993 regular season updated from Squared2020's tweet - 79 games:

Read on Twitter

The 3 games that aren't logged are:

Read on Twitter

Here is the 1987 regular season updated from another tweet - 40 games:

Read on Twitter

And the 42 games that aren't logged are:



Those are the sources..

As for 1986 and 1995, it makes sense to do it either way. As long as people understand the issue of uneven samples and don't just combine the off sample from 1995 for instance (which is huge!) to other off samples in neighbouring seasons. That would be problematic but otherwise it's ok.

Okay, thanks! That all makes sense. Do you have any link to where Squared has all 82 games for 1996 (the above stuff you linked from Squared does include some 1996 stuff, but it looks like he only had 78 games at the time he posted those)? I’d probably prefer to use Squared’s version for that than the Pollack data that’s there now, because I believe the Pollack data is just contained in a random spreadsheet someone made at some point, rather than actually being pulled from the original data source. The numbers are very similar so it doesn’t really matter much, but they’re not exactly the same and if I can avoid using that spreadsheet for something, I’d prefer that.

As far as I know, Squared never publicly released the 1996 update. So if you want, just stick with the Pollack data for now. You are using it for 1995 after all. By the way the source for the raw plus minus numbers for 1996 is on this thread. Unfortunately the links to the images are dead.

viewtopic.php?t=1343246

[lessthanjake]

The playoff revision comes from my own tracking. I basically tracked all Jordan playoff games from 1986-1996 at some point. That's how the spreadsheet got completed.

Here is the 1993 regular season updated from Squared2020's tweet - 79 games:

Read on Twitter

The 3 games that aren't logged are:

Read on Twitter

Here is the 1987 regular season updated from another tweet - 40 games:

Read on Twitter

And the 42 games that aren't logged are:



Those are the sources..

As for 1986 and 1995, it makes sense to do it either way. As long as people understand the issue of uneven samples and don't just combine the off sample from 1995 for instance (which is huge!) to other off samples in neighbouring seasons. That would be problematic but otherwise it's ok.

Okay, thanks! That all makes sense. Do you have any link to where Squared has all 82 games for 1996 (the above stuff you linked from Squared does include some 1996 stuff, but it looks like he only had 78 games at the time he posted those)? I’d probably prefer to use Squared’s version for that than the Pollack data that’s there now, because I believe the Pollack data is just contained in a random spreadsheet someone made at some point, rather than actually being pulled from the original data source. The numbers are very similar so it doesn’t really matter much, but they’re not exactly the same and if I can avoid using that spreadsheet for something, I’d prefer that.

As far as I know, Squared never publicly released the 1996 update. So if you want, just stick with the Pollack data for now. You are using it for 1995 after all. By the way the source for the raw plus minus numbers for 1996 is on this thread. Unfortunately the links to the images are dead.

viewtopic.php?t=1343246

FYI, I’ve updated the OP to include your updated playoff data, as well as the additional Squared regular season data from 1986-87 and 1992-93. I believe everything should be right, but of course if you (or anyone else) notice any mistakes, please let me know.

The numbers for specific years/playoffs have sometimes changed a decent bit with this update (even small plus-minus changes can change “off” values in the playoffs a fair bit), and obviously the 1986-87 data is completely new, but the bottom-line numbers didn’t change a whole lot. The updated playoff data results in a slightly higher playoff on-off (now at over +16 per 48 minutes in the total-playoff data). Combining the updated playoff data and additional regular season data results in a slightly lower overall “on” value, slightly worse “off” value, and very slightly lower on-off.

Regarding 1996, I’ve stuck with Pollack’s data for now. I am aware of that thread you mention (and had actually linked to it in the OP), but I still don’t *love* citing a random spreadsheet that is corroborated by a thread with dead links, rather than having direct access to the data source. It’s not *that* concerning really (the thread does corroborate the data, and I think we all uncontroversially understand that spreadsheet to be an accurate representation of Pollack’s data), but if I had full data for that season that I could cite directly then I’d opt to do that instead.