Rolling average numbers


#1

I just put up a thread with multiple posts in it on FOS regarding our rolling average numbers for this season. I’ll recreate the posts here.

These are this season’s rolling average numbers along with an explanation of what they are since undoubtedly some people won’t have a clue what they are all about.

This will be a VERY long post (which is why I don’t publish the numbers very often - it can be a pain to create the charts and rehash the explanation ever time).

Because of it’s length, and because people might want to comment on different parts of it, I’m going to break the post up into a bunch of different segments (that will also be helpful in case something goes wrong - if I lose a post I won’t have to start all over from scratch.

So here goes (it will take me a couple of minutes to cut and paste each segment over from my text file and reformat it for this forum) bear with if you happen to be up early.


#2

What’s this Rolling Average I’ve been hearing so much about?

For the last few years, I’ve been looking at DeChellis’ performance from several perspectives and publishing the results. Philosophically, my premise has been that if the team continues to show continuous improvement, then DeChellis deserves to be retained (I borrowed this idea from the Japanese quality control model of Kaizen. Conceptually, that seems like a quite reasonable perspective. You might argue that the improvement that is shown isn’t sufficient enough, and I can understand that, but I would at least think that the idea of continuous improvement wouldn’t be too contentuous (what I have discovered, however, is that when it come to things DeCheillis, everything is contentious - kind of like politics).

One of the tools that I used looking at ED’s performance was to do a rolling (aka moving) average analysis.
The concept of a rolling average was foreign to some and they belittled it despite the fact that it is a universally accepted way of analyzing time series data. The approach is often used in the financial industry and you can find examples of it in stock market charts everywhere.

But it’s not just the financial industry where moving averages are used. They turn up everywhere. Here’s an article from an engineering statistics book that tooks about moving averages and how they can help uncover trends in time series data that might be hidden.

Here’s another example. This one from Microsoft where they show how to use their Moving Average Analysis Tool in Excel to forecast inventory levels for a hypothetical pharmaceutical company.

As I’ve said it’s a common technique.


#3

It’s looks awful complicated, is it really? Can you give me a simple example?

Sure it’s not that hard. Let’s look at an example. Perhaps something simple and non controversial like global warming? All kidding aside, yearly temperatures provide a good example. Suppose I wanted to find out if the general overall trend was toward increasing temperatures. If I simple take an average monthly temperature reading and compare it to the previous month that’s not going to tell me much. Chances are pretty good that March will be warmer than Feb, Apr warmer than Mar, May warmer than April, etc. I won’t be able to tell anything from those measurements.

However, instead of just recording the temperature for the month, what I really need to do every month is to compare the averages of the previous twelve months. So each month I will be comparing a year’s month of data, i.e. the average of the previous twelve months. February’s number will include the 12 months from the previous March forward. Then in March, I move the average forward (hence moving average) drop the previous March’s number and add in the current March’s number. Each month will have a years worth of data and it will help me see if the average temperature is getting warmer.

OK - how does this translate to basketball?

The idea is that it’s probably not totally realistic to expect a coach to increase his wins every year. Some fluctuations in the wins are to be expected. One logical way to look at this is to evaluate the team on their wins over a four year period. Why do I say that’s logical? Because it simplistically is looking at each recruiting class and seeing how many wins they accumlate over their four years on campus. You hear the talking heads on TV say something similar all the time. I’m sure that several times over the next month you will hear some analyst say player so-and-so’s class has won more games than any other class in the history of school xyz. That’s all the rolling average is doing. It’s looking at a school’s wins over the previous four years (technically, it’s averaging the wins instead of accumulating them but it’s just accumulated number divided by four).

Basically, I’m saying that if each senior class wins more games than the previous senior class, then clearly the program is improving.


#4

Who can argue with that?

Well, there are some legitimate concerns. The most obvious one is just focusing on wins. Given that the number of games a team plays can easily vary from one season to the next, it’s possible to accumulate more wins simply by playing more games. A way around that is to concentrate the measurement on winning percentage rather than wins.

Second, wins can also be be manipulated from a scheduling viewpoint by booking easier teams. That can be addressed by breaking out conference play and looking at those numbers separately. Most people would agree that how a team does in conference is more important than out of conference so it’s quite valid to look at these numbers no their own. However, because the number of in conference games changed from 16 to 18 in the middle of ED’s tenure, it’s important to look at winning percentage and not just total wins.

Third, some people believe in the what have you done lately for me philosophy. They have a hard time valuing what a coach did four years ago as being worth much. I can understand this viewpoint. One way to address this is by using what’s called a weighted moving average (WMA). A WMA considers more recent values as being more important and assigns them relatively more weight.

Similarly, If I’m trying to evaluate the performance of a senior class, chances are that they have made a bigger contribution to the current year than they did their freshman year so the results of the current year should carry more weight. For purposes of my analysis, I simply assigned a weight of one to freshman year numbers, two to soph, three to junior, and four to senior. There’s nothing scientificy about this. Anyone who doesn’t like those numbers can choose others of their own. I suppose a more accurate model could be built based on minutes played by each class but that seemed too complex.

What does that leave me with now?

Basically, a whole bunch of charts and graphs. What I’m presenting below is two different rolling averages (normal and weighted), for wins and winning percentages,
both overall and within conference.


#5

Enough already, show me the numbers.

OK - here you go.

And graphed versions of the above data.

What’s do the graphs tell me?

Nothing that you don’t already know. When it comes to wins and losses, the season was pretty bad - both from an overall perspective and from a Big Ten perspective.


#6

Can we use rolling averages to look at other data?

Certainly. If you are planning on making the NCAA tournament, one of the major criteria that the selection committee will look at is your RPI. Penn State’s RPI last year was one of the things that kept them out of the NCAAs.

Some people (Rokk, for one) don’t feel that RPI ratings are a particularly good measure of a team. Whenvever he quotes rankings or schedule strength, Rokk uses Sagarin’s ratings, most specifically Sagarin’s Predictor ratings since that measures team performance independent of their wins and losses.

I understand where he’s coming from (although I disagree with him on whether they should be used to predict how the NCAA selection committee works (they will use the RPI numbers). When it comes to performance measures I refer using Pomeroy’s ratings.

So here are those three rating services using actual, rolling average, and weighted rolling averages.


#7

Let’s graph them like before.

First the RPI:

Looks like business as usual. Our RPi collapsed this season. That should come as no surprise since a significant portion of your RPI is based on your winning percentage and our record wasn’t too good.


#8

Now Rokk’s favorites, Sagarin.

Hold on here one minute. Something looks different. I see the drop off in the rating for this season but what’s happening with the rolling average numbers? The rolling average shows a decline in all the other charts for this but it’s still going up in this one. What gives?

A little analysis will provide the answer.

All the previous charts provided data that was primarily based on a team’s won/loss record. Sagarin’s Predictor Rating is based on a team’s offensive and defensive performance not on whether they win or not. The idea being if you play well, the wins will come your way. Unfortunately, that’s not exactly what happened this year. Penn State’s wins and losses did not reflect how they were actually performing. Sagarin’s rating supports that.


#9

So what about Pomeroy’s numbers. Here’s his graph.

Looks similar to Sagarin’s. Again we have the dropoff in the rating for this season, but the rolling average continues upward. Pomeroy’s ratings are also performanced based. His Pythagorian Rating is based on a team’s efficiency ratings, both offensive and defensive. That’s similar to Sagarin so it shouldn’t be surprising that the graph of his numbers should look similar to Sagarin’s.

Remember the rolling average numbers are calculated by including one new data point (this season) while dropping off one old data point (in these cases, the 2005-06 season). Our record this season was worse than the 05-06 season so one would expect the rolling average numbers for win/loss stats to go down. However, our performance based numbers for this year are better than the numbers for 05-06 season so our rolling average for performance based rating systems such as Sagarin and Pomeroy should go up - and they do.


#10

Believe it or not, both Sagarin and Pomeroy say that this season’s team was actually the second best performing team under DeChellis. Unfortunately, their record did not reflect that. Looking back at all the close games that we had against quality opponents and it probably shouldn’t come as a surprise that our record isn’t as good as our play was. When the talking heads mentioned that no one wanted to play Penn State now, it was with the understanding that the wins were finally starting to come our way. When Curley was quoted yesterday as saying that the team’s record did not support how well they were playing, he was just echoing what Sagarin and Pomeroy’s numbers said.

In fact, when you look closer at both of those performance systems you start to understand how much different our play was from our record. Sagarin has multiple rating systems. One of them, the ELO-Chess system is based entirely on wins and losses, it does not reflect margin of victory (or loss) at all. The other one, his Predictor system that I have charted above takes the opposite perpective. It’s based entirely on
performance and not wins and loses. When you compare how Penn State ranks under the two systems, it turns out that the Predictor systems has us 81 places higher than the ELO-Chess system. That is the biggest delta of all 347 teams that Sagarin tracks.

A similar thing shows up under Pomeroy’s ranking. His Pythagorean System is supposed to predict a team’s winning percentage. When you compare Penn State’s winning percentage with what Pomeroy says it should be PSU ranks 346th of the 347 teams (we were 347th before Thursday’s game butslipped behind Holy Cross after the loss).

So Curley has actual data to support his view that the teams efforts on the floor aren’t relected in their won-loss record.

[font=Verdana]THE END[/font]


#11

[quote=“UncleLar, post:10, topic:968”]In fact, when you look closer at both of those performance systems you start to understand how much different our play was from our record. Sagarin has multiple rating systems. One of them, the ELO-Chess system is based entirely on wins and losses, it does not reflect margin of victory (or loss) at all. The other one, his Predictor system that I have charted above takes the opposite perpective. It’s based entirely on
performance and not wins and loses. When you compare how Penn State ranks under the two systems, it turns out that the Predictor systems has us 81 places higher than the ELO-Chess system. That is the biggest delta of all 347 teams that Sagarin tracks.

A similar thing shows up under Pomeroy’s ranking. His Pythagorean System is supposed to predict a team’s winning percentage. When you compare Penn State’s winning percentage with what Pomeroy says it should be PSU ranks 346th of the 347 teams (we were 347th before Thursday’s game butslipped behind Holy Cross after the loss).

So Curley has actual data to support his view that the teams efforts on the floor aren’t relected in their won-loss record.

[font=Verdana]THE END[/font][/quote]

Thanks for putting this all together, Lar. It’s interesting to see all of the data. As for the above comments, I think an argument can be made that this is a coaching issue. If the team is playing well, but losing tough game after tough game, and having close loss after close loss, at some point it has to fall on the coach. Why can’t the team get over the hump? What is the coach doing that is creating a situation where the team is losing over and over? Or, asked differently, what isn’t the coach doing to put the team in a better position to win at the end of games based on how well they’re playing throughout the games?

It’s been a pretty consistent thing for the team under ED. There is inevitably a 6-8 minute stretch in the last 10-12 minutes of games where the team just dies and the buckets won’t go in. Whether it’s misuse (or non-use) of timeouts, or poor offensive gameplanning, or misuse of personnel, it doesn’t really matter. After this long of time, it’s on the coach.

What I do think you can read from the numbers though is that ED has done a better job of putting together some talent than people tend to acknowledge. That’s why the team has progressively improved it’s play during his tenure, there has been a deeper collection of players who can play.


#12

Thanks Lar.

The eyeball guys without an agenda should be able to admit that this year’s team was much better than their record, and realize there’s plenty to be excited about next year despite the awful taste left in our mouths from the finale of this year.

I always find it funny when people want to use numbers to justify their arguments (like W/L record) but then want to ignore all the other numbers that don’t help their cause. Nobody defending keeping Ed around is going to ignore his W/L record, we’re just going to realize that it doesn’t and can’t ever tell the entire story.

The program is improving under Ed, it just doesn’t seem like it’s going fast enough for most people’s standards around here.


#13

Lar - You are a quant guy - what does it even mean to plot the RPI (not the ranking, the index itself), over different years? It has barely useful interpretation from one year to the next, because all that really matters is the ranking relative to other schools! a .59 RPI that puts you around 40 in one year is better than a .59 RPI that puts you around 60 in another year. You know this!

So let’s look at a comparison of the first 7 years of Dunn and DeChellis. I can’t make the case that he has things going in a better direction than Dunn Did. It seems to be worse actually

(green and yellow are 4 year moving avg)

Let’s also look at the strength of schedule. Dunn was definitely taking a more aggressive approach overall on the scheduling front - that means a better RPI and a better chance of making the dance in good years - which is why, I hypothesize, that Dunn had been in the NCAA 2 times and the NIT 2 times by this point - and Ed has just seen the NIT. They went in opposite directions! Ed is failing the university with his scheduling and in game coaching. These stats suggest that it was stupid to fire Dunn if we were going to replace him with this guy

Curley’s statements are a farce! He brought Ed in, supposedly, to improve the program. He has done the opposite relative to his predecessor!


#14
I can't make the case that he has things going in a better direction than Dunn Did. It seems to be worse actually

You don’t have to, it was made already by lar. you’re proving nothing here except your ignorance.


#15

You don’t have to, it was made already by lar. you’re proving nothing here except your ignorance.[/quote]

You know you’ve been had when the usually friendly Craftsy goes out of his way to shut you down. Score 1-0


#16

You don’t have to, it was made already by lar. you’re proving nothing here except your ignorance.[/quote]

I remember a time when a coach had to make the Sweet 16 to save his job (and then only for another 2 years). Looks like these days with the hundreds of ways to make a fart look impressive all it takes is a 3 win Big Ten season. Talk about grade inflation.


#17

You don’t have to, it was made already by lar. you’re proving nothing here except your ignorance.[/quote]

My ignorance? Please. Not worth a response. just look at the charts that I posted. Lar posted some interesting stuff, but only the ranking relative to other schools really matters as a trend. The index by itself doesn’t tell you much over time, especially the RPI. I think Lar would admit this, too.


#18
[quote="Craftsy21, post:14, topic:968"][quote] I can't make the case that he has things going in a better direction than Dunn Did. It seems to be worse actually[/quote]

You don’t have to, it was made already by lar. you’re proving nothing here except your ignorance.[/quote]

You know you’ve been had when the usually friendly Craftsy goes out of his way to shut you down. Score 1-0

I’m willing to wager that my quantitative background is equal to or greater than Lar’s.


#19
[quote="Craftsy21, post:14, topic:968"][quote] I can't make the case that he has things going in a better direction than Dunn Did. It seems to be worse actually[/quote]

You don’t have to, it was made already by lar. you’re proving nothing here except your ignorance.[/quote]

My ignorance? Please. Not worth a response. just look at the charts that I posted. Lar posted some interesting stuff, but only the ranking relative to other schools really matters as a trend. The index by itself doesn’t tell you much over time, especially the RPI. I think Lar would admit this, too.

Should the charts actually be showing up I might be able to take a look at them. And if we’re not judging ourselves relative to other teams, I guess titles don’t mean much either now? You should know with your “quantitative background” that it might take more than one series of numbers to describe a situation. Sure RPI isn’t the only thing that matters, but that’s why he also considers Pomeroy and Sagarin. Your beef with his numbers makes no sense.


#20
[quote="Craftsy21, post:14, topic:968"][quote] I can't make the case that he has things going in a better direction than Dunn Did. It seems to be worse actually[/quote]

You don’t have to, it was made already by lar. you’re proving nothing here except your ignorance.[/quote]

I remember a time when a coach had to make the Sweet 16 to save his job (and then only for another 2 years). Looks like these days with the hundreds of ways to make a fart look impressive all it takes is a 3 win Big Ten season. Talk about grade inflation.

There’s no question this year was a step back in the win column… seems like you continually miss the point in favor of throwing out another fart joke, yet continue to assume everyone else is somehow “kidding themselves”.