APBRmetrics

hoopseng · Joined: 13 Oct 2006 Posts: 68 Location: Basketball Research

Gabefarkas,

I respect your study. I didn't use a statement which indicates winning or losing depends only efficiencies. There are plenty of independents, plenty of factors out there. Margin of victory, road/home, rest days, players in the rotation and many other splits have all relationships with winning and with efficiencies. What's more, as you all know efficiencies include point differentials.

We should better perform a regression analysis to get deeper results.

Mine was just a Pearson product-moment correlation coefficient between "two variables only". Article needs to get improved. Would you like to contribute? That'll apreciate me. Just let me know if you want to.

Magicmerl,

I'll convert it to playoff version for sure but I want to do it timely, after the regular season.
_________________
http://www.nbastuffer.com

Kevin Pelton · Posted: Tue Jan 22, 2008 10:16 pm Post subject:

I don't have any issue with using correlation with team performance to measure the relative importance of offense and defense, but what's problematic is using only one season. These correlations have a tendency to fluctuate a great deal. I looked them up for a column I did after the 2005-06 season, and here's what I found:

Mountain · Joined: 13 Mar 2007 Posts: 537

The 10 year dataset shows variation and even-handed. Offensive with a lead this year. No clear bias there.

There appears to be a skew or movement toward weaker correlations with both efficiencies. But is it a trend caused by anything or just a subtle tilt pattern in a 10 point dataset?

gabefarkas · Joined: 31 Dec 2004 Posts: 984 Location: Durham, NC

hoopseng · Joined: 13 Oct 2006 Posts: 68 Location: Basketball Research

Power rankings charts should look better now. Some visual improvements have been made.
_________________
http://www.nbastuffer.com

Mountain · Joined: 13 Mar 2007 Posts: 537

I wonder if a third chart showing change between season numbers prior to last 5 games and last 5 games with a line representing the 2 prior and new values and an arrow showing the direction of movement and destination would be another helpful addition. Maybe do it by east and west if the chart is too busy with everybody. Just an idea if you want to try it.

You can go back and forth between the charts and figure out the movement but it is a little slow and I usually just do a few, not all.

This time I see Sacramento moved from average offense to league best in most recent 5 games. And Spurs fell back a good distance to below average on offensive efficency recently mitigated by some defensive improvement and the combination suggests the type of games they were in. But what other stories are in there that I miss unless I shuttle back and forth for everybody?

hoopseng · Joined: 13 Oct 2006 Posts: 68 Location: Basketball Research

Good idea. I can add a table showing the movement stories. We should better not use a chart for this purpose because amount of the arrows (30) will make it seem busy and tough to comment.
_________________
http://www.nbastuffer.com

Mountain · Joined: 13 Mar 2007 Posts: 537

Ok, your plan makes sense. Look forward to seeing it.

hoopseng · Joined: 13 Oct 2006 Posts: 68 Location: Basketball Research

Mountain · Joined: 13 Mar 2007 Posts: 537

I think you presented it well and it is a valuable new source of information.

I see, at least for last 5 games, that the offensive movers to the upside moved more than the decliners while on defense slippage was greater percentagewise than the movement of those who clamped down better. I will interested to see if these patterns are common or if this was just the way the last 2 weeks went.

Ed Küpfer · Joined: 30 Dec 2004 Posts: 646 Location: Toronto

hoopseng · Joined: 13 Oct 2006 Posts: 68 Location: Basketball Research

Ed Küpfer · Joined: 30 Dec 2004 Posts: 646 Location: Toronto

mikez · Joined: 14 Mar 2005 Posts: 39

One other thought on the Season and Last 5 games charts: make the chart square, instead of (just) rectangular. Right now a good (vs league average) offensive team is further from the origin than an equally good (vs league average) defensive team -- if the chart were square, total distance from the origin (assuming equivalent scales on the axes, which you have) would indicate overall point differential, which might be useful.

(Or is the graph non-square due to the "offense wins games" point you make earlier in this thread? If so, ignore this comment - I'm not going to comment on that point per my usual policy.)

Just a thought.

-MZ

Mountain · Joined: 13 Mar 2007 Posts: 537

Using the charts Cleveland is the "hottest" team on both efficiencies. Not the best team, the most improved team compared to season average.

Other 2 way hot teams (in both top 10s): New Orleans, Chicago, NY, Orlando and the Clippers.

2 way not hot: Portland, Denver, Seattle and Washington.

If you wanted to carry the work even further I guess it might be possible to adjust for strength of schedule and home & away.

I know there are strength of schedule and home & away adjusted power rankings but has anyone presented strength of schedule and home & away adjusted offensive and defensive efficiencies? Are you interested in blazing that trail? That would help put the different 5 game experiences of the teams on a similar basis and help try to isolate "improvement" over the impact of the schedule. It would of course still be worth seeing the raw data but having both available would lead to sounder impressions about improvement and less false signals arising from the variable degee of difficulty of schedule.

Regarding different horizontal and vertical scales, in the past when others have presented efficiency charts it was stated tha tthey varied the scale to fit the dimensions of the available screen and to help make the horizontal text as readable as possible. But it does influence perception and I'd prefer a same scale graph too, assuming it works on space and readability.