Some Historical Plus-Minus

[colts18]

I remember that a Youtube user uploaded all 82 games for the Bulls for 1991 and 1996. I wish I could find it because he was posting all of the games he had from the 1990-1993 timeframe.


Btw, We have MJ playoff on/off. Thinking Basketball posted it in their Youtube Video that had MJ playoff data.

That YouTube user could not have been recent. The most recent I remember was "Hoffman's Habitat" which posted a ton of Bulls games, but there were no games he uploaded that I did not already have. His account was terminated after the Broadband TV copyright purge 8 months ago. Curious to identify who you are thinking of.

That might be the guy I'm thinking of. I remember watching his videos a year ago and he was consistently uploading all of the 90s Bulls games.

[Squared2020]

.

[Squared2020]

.

[Squared2020]

.

[WestGOAT]

It's not necessarily the best measure. We could use nonabelian transforms to wring out multicollinearity a little better.

Two possession stints results in ratings of 0, 50, 100, 150, ... (on the 25's; should have said on the 50')
Three possession stints result in ratings of 0, 33, 66, 100, ... (on the 33's)

This is where the inflated variance comes from: using a Gaussian to approximate these jumps. Here's what the distribution of ratings look like for approximately 200 games from 1995-96:

Why is it that two possession and three possession stints always result in those specifc ratings (multiples of the reciprocals of 2 and 3)? As for ratings, do you mean net ratings of a lineup for a continuous stretch of time: the difference between a team's offensive rating (points scored per 100 possessions) and defensive rating (points allowed per 100 possessions) ?

I'm no hardcore stats guy, so not sure if I follow, and I have never calculated RAPM myself (at least not yet). I might use your website as a resource to figure out how to perform the calculations

[Squared2020]

.

[Squared2020]

.

[ceiling raiser]

Question -

What percentage of Bulls games are available each year on tape from 84-85 through 95-96?

Wondering what the maximum possible dataset for MJ’s on-off would be.

Thanks again for your work.

These are approximate guesses on what I have:

1985: ~35 games
1986: ~15 games
1987: ~40 games
1988: ~40 games
1989: ~50 games
1990: ~50 games
1991: ~55 games
1992: ~65 games
1993: ~75 games (79)
1994: ~20 games
1995: ~25 games
1996: ~80 games (82)

I know I could dredge up about another 20-25 games from friends and family.

Follow-up: How many have you tracked thus far from each year?

I also wonder if the 97, 98 days can be trusted. Some of the pbp based data seems off those first few years.

[70sFan]

Moses Malone Plus-Minus

Do you happen to have something from George Gervin and Sidney Moncrief?

[Squared2020]

.

[ceiling raiser]

One of my favorite threads on this board, going to dive deeper in the offseason.

[ceiling raiser]

.

Hope you're doing well. I saw on your site that 87-88 (20%), 90-91 (26.6%) and 95-96 (20.02%) are really coming along from a tracking POV.

Have you been focusing on those years specifically because of availability of footage, or have you looked at some other years in the late 80s-mid 90s as well?

[Squared2020]

.

[homecourtloss]

Everything has been manually tracked from video over the past decade or so. I have basically from the 1979-80 season to the 1995-96 season. As you could expect, the number of games get smaller and smaller as we go back in years.

I have between 200 and 400 games for each season, rolling back to about 1987. Then it dwindles pretty fast. I think the early 1980's there's only maybe 50 games per season. But of course, with careful notation, some data is much better than no dat; but context of that data is really important. Side note: There was a post here about two years ago where a poster explained how to interpret the results even better than I could when folks were complaining about small samples.


Anyways, the reason I picked those years to post were because of Ben Taylor asking about certain years of data.

1984-85: Jordan's rookie season
1987-88: Pippen/Grant rookie season; first year out of the first round
1990-91: First NBA Finals
1995-96: Last NBA Finals

I've been converting notebooks over for all the seasons and stop to gather more data from games. For example, I have about 400 games from the 1995-96 season but only have about 300 logged.

This is what a log looks like:



There's two games there. Starters are in the header. Left columns are possessions played for the stint. Right columns are points scored during the stint. Far right are substitutions that end the stint. Tally marks at the top indicate how many shared stints there are in the game.

I then translate the stint data to a csv file, with redundancies to help with error checking. The data file looks like this:



Once I get to 10/15%, I make a public post.

I am unbelievably thankful that you have made this data publicly available — thank you.

I wish there were funding available from somebody who loves basketball enough to get some college interns and start tracking some of the play-by-play data for footage that’s available as you have done.

[rk2023]

Everything has been manually tracked from video over the past decade or so. I have basically from the 1979-80 season to the 1995-96 season. As you could expect, the number of games get smaller and smaller as we go back in years.

I have between 200 and 400 games for each season, rolling back to about 1987. Then it dwindles pretty fast. I think the early 1980's there's only maybe 50 games per season. But of course, with careful notation, some data is much better than no dat; but context of that data is really important. Side note: There was a post here about two years ago where a poster explained how to interpret the results even better than I could when folks were complaining about small samples.


Anyways, the reason I picked those years to post were because of Ben Taylor asking about certain years of data.

1984-85: Jordan's rookie season
1987-88: Pippen/Grant rookie season; first year out of the first round
1990-91: First NBA Finals
1995-96: Last NBA Finals

I've been converting notebooks over for all the seasons and stop to gather more data from games. For example, I have about 400 games from the 1995-96 season but only have about 300 logged.

This is what a log looks like:



There's two games there. Starters are in the header. Left columns are possessions played for the stint. Right columns are points scored during the stint. Far right are substitutions that end the stint. Tally marks at the top indicate how many shared stints there are in the game.

I then translate the stint data to a csv file, with redundancies to help with error checking. The data file looks like this:



Once I get to 10/15%, I make a public post.

I am unbelievably thankful that you have made this data publicly available — thank you.

I wish there were funding available from somebody who loves basketball enough to get some college interns and start tracking some of the play-by-play data for footage that’s available as you have done.

As one with lay-off time from graduation -> full-time work, certainly second the notion here

[ceiling raiser]

I have between 200 and 400 games for each season, rolling back to about 1987. Then it dwindles pretty fast. I think the early 1980's there's only maybe 50 games per season. But of course, with careful notation, some data is much better than no dat; but context of that data is really important. Side note: There was a post here about two years ago where a poster explained how to interpret the results even better than I could when folks were complaining about small samples.

This is incredible, thanks for your work.

So if I'm reading correctly, not only do you have these games from other seasons, but you've *already* produced stint data and are just in the process of digitizing them?

This is probably the single greatest statistical/analytics achievement in the history of the NBA. The sheer amount of time, effort, and intelligence you've put into this is unprecedented.

[homecourtloss]

I have between 200 and 400 games for each season, rolling back to about 1987. Then it dwindles pretty fast. I think the early 1980's there's only maybe 50 games per season. But of course, with careful notation, some data is much better than no dat; but context of that data is really important. Side note: There was a post here about two years ago where a poster explained how to interpret the results even better than I could when folks were complaining about small samples.

This is incredible, thanks for your work.

So if I'm reading correctly, not only do you have these games from other seasons, but you've *already* produced stint data and are just in the process of digitizing them?

This is probably the single greatest statistical/analytics achievement in the history of the NBA. The sheer amount of time, effort, and intelligence you've put into this is unprecedented.

In the Internet age, when it’s usually impossible to say things such as a bolded and it not be considered hyperbole, THIS IS TRUE. I cannot even imagine how many hours have been spent doing this.

[Moonbeam]

Thanks for the hard work, Squared2020! Nearly 20 years ago, Steve Hellmuth had mentioned that the NBA was looking to digitize every single game, with plans to offer fans the option to purchase access to older games.

https://www.theglobeandmail.com/technology/nba-mounts-a-full-court-digital-press/article18242514/
https://www.tvtechnology.com/news/nba-moves-to-monetize-archive

However, as far as I know, the options are still very limited. It would be awesome if there was an update about this. Should these games become more widely available, I could imagine a community project to log stint data for these games.

[Squared2020]

.

[WestGOAT]

Everything has been manually tracked from video over the past decade or so. I have basically from the 1979-80 season to the 1995-96 season. As you could expect, the number of games get smaller and smaller as we go back in years.

I have between 200 and 400 games for each season, rolling back to about 1987. Then it dwindles pretty fast. I think the early 1980's there's only maybe 50 games per season. But of course, with careful notation, some data is much better than no dat; but context of that data is really important. Side note: There was a post here about two years ago where a poster explained how to interpret the results even better than I could when folks were complaining about small samples.


Anyways, the reason I picked those years to post were because of Ben Taylor asking about certain years of data.

1984-85: Jordan's rookie season
1987-88: Pippen/Grant rookie season; first year out of the first round
1990-91: First NBA Finals
1995-96: Last NBA Finals

I've been converting notebooks over for all the seasons and stop to gather more data from games. For example, I have about 400 games from the 1995-96 season but only have about 300 logged.

This is what a log looks like:



There's two games there. Starters are in the header. Left columns are possessions played for the stint. Right columns are points scored during the stint. Far right are substitutions that end the stint. Tally marks at the top indicate how many shared stints there are in the game.

I then translate the stint data to a csv file, with redundancies to help with error checking. The data file looks like this:



Once I get to 10/15%, I make a public post.

I am unbelievably thankful that you have made this data publicly available — thank you.

I wish there were funding available from somebody who loves basketball enough to get some college interns and start tracking some of the play-by-play data for footage that’s available as you have done.

Definitely doing God's work, and I'm actually pretty sure he can get a college intern to do the gruntwork since he's working for/with the league office. If that's what he wants is another story of course