Dismal Numbers

System Dynamics with NetLogo

2011-05-01T07:19:00.000-07:00

About 1 year ago I purchased an old book by Jay Forrester on dynamic systems. I really enjoyed the book and looked around for more information on modeling these systems. I ran into NetLogo . It has some basic models already built and it works on Macs and PCs well. My favorite model is the wolf, sheep, and grass simulation. Basically, the grass grows, the sheep eat the grass, and the wolves eat the sheep. Parameters are grass regrow speed, breeding dynamics, and how much energy is gained by eating. Each animal has an energy level that is reduced each "tick" of the simulation. If it runs out of energy, it dies. You can also change the starting # of sheep and wolves.

They have the simulation setup so if you run it, it eventually settles into a volatile but sustainable steady state between grass, wolves, and sheep.

The interesting part is changing some of the parameters to see what happens. When you do, you quickly learn that intuition takes a holiday in system dynamics. For instance, I changed how much energy the sheep gain from eating grass from 4 to 5 (for whatever reason, they improve their utilization). What do you think happens in the steady state values of grass, sheep, and wolves? I expect that the same amount of grass would sustain more sheep and wolves due to the sheep extracting more energy every tick.

Turns out, the wolf population goes down! Change the sheep gain to 6 and the wolves die off totally after a while. What happens is the additional sheep create increasingly large oscillations in the wolf population that eventually wipes them out. Certainly gave me something to think about.

Bowl Game Predictions

2010-01-01T10:54:00.000-08:00

http://spreadsheets.google.com/ccc?key=0As1wRbiaO_HWdHpzNmdVU2pMUkJjSkE0c213SnBTV2c&hl=en

Using a simple system to adjust each teams scoring and defending (based on ranked teams played), I average a teams scoring with another teams defending to get the scores. These are the bets I would have placed using same money on each bet. It may be worthwhile to bet more on games with more disagreement with my forecasts.

I was shocked as to how close the Cin/FLA game was and how favored my model said Penn State is.

Really SOur Grapes..

2009-11-22T13:57:00.000-08:00

Unreal, Browns lose with no time left. A few things I really don't like although I don't have any data for it.

I hate the Pro pass interference penalty, give me college anyday. Roughing the passer is 15 yards but some bs judgement call by a referee during a hail mary pass is good for 40 yards? Please... It puts too much of the game in the referees hands. If someone intentially hits someone in the endzone, flag them for a personal foul, that would make it 25 yards and an automatic first down. That's more than enough.
I hate the prevent defense or if you must, at least the 3 man rush. Why not just drop everyone back in coverage? What would be the difference? Rush 4 at least everytime.
And for the final lethal combination, being ultra conservative even though you can't run the clock out anway. Make 2 rushes for .5 yards and then throw on 3rd down anyway? Why not throw when they don't expect it? Or just run your normal offense? Anything but the 3 and out guarentee.

Think of this end game the next time someone goes for it on 4th down.

The offense looked good at least. It would be nice if they can keep that going.

Sour Grapes?

2009-11-16T18:57:00.000-08:00

After having to listen to how good Cincinnati (college in this post) is and how they would "kill" Ohio State if they played this year, I decided to do some analysis. Basically, how do you rate teams in college FB when they play such a wide variance of team skill levels? I didn't have much time but through the magic of Export to Excel (IE8) and http://www.cfbstats.com/ , I was able at least get some idea how you could levelize teams using some hard numbers. I settled seeing how teams performed against "ranked" teams vs how well they did against unranked teams. Some teams only play 1 ranked team a year so you have an individual team analysis won't work. Instead, I used split data from 2004-2008. The overall average was that teams score about 11 points less against ranked teams than their average against unranked teams. Ranked teams score about 11-12 points more than their average defense when facing unranked teams. The more ranked teams a team plays, the range of outcomes gets smaller and the impact is less.

I then adjusted Cincinnati and OSU's offensive and defensive average scores/game to levelize them to ranked teams. Since OSU played 3 ranked teams this year, the adjustments to their numbers was less. I then created points scored by each team as an average between their expected offensive points and the opposing defensive allowed points for each team. As it turns out, it is actually pretty close with a slight advantage to OSU. Still, this analysis could be improved but doing a strength of schedule analysis or maybe using vs Winning Teams split data.

Overall, I was a little shocked as to how teams can pad their stats by playing unranked teams. It can make a real difference when comparing teams. I also love the cfbstats website, great work.

Here are the data and analysis links
http://spreadsheets.google.com/pub?key=tyxi63nAKZsVaJ6jptw1rjg&output=html

http://spreadsheets.google.com/pub?key=tQHvnEBJUAVGj80yzkC74nw&output=html

Crazy Game, blows up model, or does it???

2009-10-11T04:58:00.000-07:00

What a crazy game the OSU/Wisconsin game was. Using the model as is, we get Wisconsin winning 24 to 10! The key missing stat was Return Yards and Interception Returns. Add those back in and you get OSU winning 35 to 24. Actually, Wisconsin only scored 13 points which is much lower than expected. Any comments as to why their scoring was so low? Bad field position maybe?

http://spreadsheets.google.com/pub?key=pwYkO8gv1ZMhdjSpMq2nnSA&single=true&gid=4&output=html

3 Qbs from last year

2009-09-13T04:34:00.000-07:00

Taking the 3 Qbs from last year, which one could be considered "better" when it counts and not better at padding stats during other games.

For a simple analysis, I looked at points per game vs ranked teams and unranked teams. Also, I compared their points per game average vs the conference average. The SEC gave up almost a TD+FG less per game than the Big 12 so the difference is material. Using these simple stats, I would rank them Bradford/Tebow/McCoy. Bradford and Tebow are very close so maybe call it a tie.

It is interesting to see the drop-off in points going from ranked to unranked opp. Next I may do winning vs non winning opp.

This is getting to be a bad habit!

2009-09-13T04:16:00.000-07:00

More dismal numbers although the game did live up to the hype at least. Could have used Beanie this year! I just started reading The Hidden Game of Football again by Palmer. Kicking a field goal on the opp 5 yard line is very much frowned upon. May have not made a difference looking at the drive logs but I'll try anything at this point.

As for the losses against ranked opp., I am not in panic mode yet. The Penn State, Texas, and now this USC game could have gone either way. 1 team has to lose and if you can play a top 5 team to a 50/50 chance of winning, I would say you are a good team. What winning % can you expect against top 5 teams realistically? Still, might be nice to win 1 from time to time!

Golf Statistics Analysis

2009-06-27T04:48:00.001-07:00

I thought it might be fun to try some golf statistical analysis. The PGA website makes it very easy to download data into excel.

The main statistics I used were driving distance, greens in regulation, driving accuracy, average puts per round, and average score. The driving distance and accuracy combine to make the % greens in regulation and then putting and %greens in regulation drive average score.

Here is the %greens in regulation stats and the average score regression.

What can we learn from this?
Before that, I also graphed totals winnings vs average score.

Wow! That last stroke is what makes Tiger a Tiger.

Let's look at using this information now. Can we advance a player to the "Tiger" line? Let's pick on Jim Furyk for no reason other than he is sponsored by a power company. Lets plot where he is relative to the line that would get him the same score as Tiger.

First we notice something, Jim is very good at getting to the green in regulation. Getting better in that area while certainly possible, is going to be very difficult and probably still won't do much for him. Look at the putting now. Now we have some room to improve. There are many players that have better putting stats than Jim does so it should be easier in relative terms to improve in this area. I plan on looking at Jim's stats since this analysis (2007 data I think) and see if the pattern repeats or not.

I didn't realize how easy it was to get golf stats now. Very interesting data.

Some Fiesta... (not!!)

2009-01-18T07:41:00.001-08:00

Well, I at least should be happy that model didn't do too bad. The score was quite a bit lower than expected but the spread was dead on. Still, this is getting a bit old. Maybe OSU should pass more? 7 yard per att. vs 5 per att running the ball?

Link Test

2008-12-28T06:56:00.000-08:00

http://spreadsheets.google.com/pub?key=pwYkO8gv1ZMgyj_wn4IpZPA&output=html&gid=0&single=true

2008-06-01T08:30:00.000-07:00

I found this chart interesting. It is the On Peak Weekly Average (Day Ahead) for SE Mass in the NE ISO from 2004-2007. It does need to be indexed to natural gas prices in NE but look at 2006, 2007. The exponents of the power law equations of all the years (except 2005) are fairly close. 2005 had Katrina which really spiked the prices at the end of the year. I expect that when I index the prices to nat gas, the chart will be even more behaved. You can also see that the highest and lowest weeks don't quite fit. They could be partitioned out to make the fit better. Also, there could be different ways to analyze this data.

new coal plant tracking

2008-06-01T07:06:00.000-07:00

http://www.netl.doe.gov/coal/refshelf/ncp.pdf

Dismal Numbers Indeed Part 2

2008-05-04T06:56:00.000-07:00

While doing some other work, I wanted to see the average household income by census region. Being in wonderful East North Central, I focused on that region. This website is the source for my chart.

http://tonto.eia.doe.gov/cfapps/STEO_Query/steotables.cfm?periodType=Annual&startYear=2005&startMonth=1&endYear=2009&endMonth=12&tableNumber=4

Sad, we were a solid 4th and now we are 6th going on 7th. We have had the lowest average growth over the period.

Very Interesting Book!

2008-05-04T06:23:00.000-07:00

Someone sent me this link. If you don't read anything else, read Chapter 2. It has some great insights that really made me think. The transistor forecast method was very interesting. My company is in the business of selling electricity which customers don't really use/need (actually lethal if they tried). Also, the comments around rain dances being all about the dance and not the results was also very interesting. It all starts with the ultimate objectives, not the methods to meet the objectives.

http://www.forecastingprinciples.com/Long-Range%20Forecasting/contents.html

Offense vs Ranked Unranked

2008-01-19T05:55:00.000-08:00

Not much different rsquared wise and the error is about the same. Still, I am more inclined to believe the impact on offense than I am the defensive numbers when trying to estimate a teams performance against ranked teams when using unranked statistics.

Unranked vs Ranked Defenses...

2008-01-18T19:59:00.000-08:00

I wanted to see how much of the defensive statistics against unranked teams matters when you play a ranked team. Using cfbstats, I can split the data by games against ranked/unranked teams. It has data back to 2004 and I filtered out teams which didn't play at least 3 ranked teams that year. On average, ranked teams score about 10 more points than unranked teams. Using regression, the r squared is 32% so basically the non ranked statistics matter just not as much as maybe you would think. The standard error is almost a touchdown so their is a lot of randomness that needs to be understood. We would have to figure out what part of the statistics may give us clues to how well a team will do against ranked teams as compared to unranked teams. Here are the charts and regression stats...

2008-01-13T07:09:00.001-08:00

Here are some sample teams through the years. Some things stick out. USC defense is basically just as effective against ranked teams as unranked teams. Anyone know why or is it just randomness? LSU is about average except the 2007 team was just as effective against ranked as unranked. Overall, LSU appears to be quite stingy on defense regardless of who they play. WV had some weak defenses in 2005-2006 but the # of games is so low it is hard to tell. The other years were about average so maybe it is just randomness.

College Football Defense vs Ranked Opponents

2008-01-13T06:43:00.000-08:00

While on my favorite college football stats site, I saw you could segregate some stats based on ranked opponents vs non ranked. I am just getting started but this is what I see so far. The ranked teams score about 10-11 points per game above non ranked teams. This was constant since 2004. I am going to see how well defensive rankings (based on points allowed) change when comparing ranked vs unranked teams. Basically, how much do statistics against unranked teams matter when playing ranked teams?

some more dismal numbers

2008-01-08T03:43:00.001-08:00

Once again, OSU falls in defeat in the BCS title game. Still, you have to be there to lose so it can be all bad. Anyway, how close was the game, did the score reflect expected performance. As you can see, LSU wins by 7 points according to the regression. Turnovers killed OSU although the fumble was on 4th down but the blocked FG and a the roughing the passer penalty can be considered a turnover in some sense. Basically, you give the ball back without the benefit of a punt. I don't know if blocked kick stats are out there but that would help. Penalties are tough to model as well, in some cases they don't matter while in other situations they are basically turnovers if the other team's drive stays alive. Dropped passes are another thing not modeled but still OSU scored more points than expected by 4 and LSU by 11. Well, onto next year as the saying goes, with no Browns, Bengals, or a BCS win by OSU, it is going to be a long DISMAL winter. :-)

Mich State vs BC

2007-12-28T19:21:00.000-08:00

Did the 4 Mich State interceptions decide the game? Yes according to this analysis. Reducing the interceptions down to 2 would have been enough for Mich State to win.

so much for that

2007-12-02T07:10:00.000-08:00

I was hoping the punt yards would explain some of the error, it did not. Now I think interception and fumble return yards would have to modeled. but for this game, there was only 1 turnover returned for 22 yards. That would net 2 net points which is in the right direction but does not explain much of the error. I suppose kick off returns would be next. I think the key was the Mizzo field goals instead of touchdowns and the poor field positon Mizzo had due to OK's offense even when they did not score.

Why did Mizzo punt twice in the 4th quarter??

Mizzo OK game drive log method

2007-12-02T06:48:00.000-08:00

Here is the detail for the Mizzo/OK game. It did eventually get the winner right but the errors were high. I think it is due to the model not taking into account field position and punts. Punts would be the easy to model either directly or as offensive yards. While punts don't create pts, they do impact the differential which is what is modeled here.

Drive Log Method

2007-12-02T06:19:00.000-08:00

Using the same regression data, I applied to drive log data using the yardage and turnovers. Using the WV/Pitt game, I tallied the differential by quarter compared to the actual pt difference. Interestingly, the error started out being a touchdown but ended up at 0. I will test the Mizzo/Sooner game as well. Here are the screenshots...

chaos hits the BCS

2007-12-02T05:08:00.000-08:00

After these 2 games, things in the BCS are really messed up. Ohio State is back in it (yeah!) but who do you pick for #2? On a personal level, I like this part of college football and it is why I buy NCAA football for my Xbox 360 and not Madden Football. Count me in the non play-off crowd for sure, we have the NFL for that. Still, do we have to play on Monday night after New Years? I hated it last year and I'll hate it this year. TV money I guess... Oh yeah, here are the regressions on the 2 big games yesterday. No big surprises except the 3 fumbles by WV. That costs you 5 pts each but Pitt had 2 interceptions which cost 4 pts each so the turnovers pretty much canceled out.

another great website from the DOE

2007-11-28T07:31:00.000-08:00

http://tonto.eia.doe.gov/STEO_Query/app/

All about the incentives..

2007-11-27T03:42:00.000-08:00

While working with statistics, game theory, economics, and decision theory; you will find cases where people seem to make counterproductive decisions even when you know they have to know better. I am not referring to simplifications that yield very close to the optimal results but instead outright seemingly counterproductive decision behavior. Most people think that winning the game, what is best for the family, and/or more money is all people care about. People are much more complicated than that, they are much more motivated by self interest and risk avoidance in the sense of social/political status. They will carefuly consider how their actions will be perceived now and in the future. What is good for the group now, may not be in their best interest for the future.

Doing things counter culture can be very risky especially if you fail. There is kind of a safety net by sticking with the crowd even if you reduce the value of your decisions. In essense, you are calling everyone who has stuck with the status quo a dummy of sorts. There are many cases (medical, sports, business) of this where good ideas had to fight through culture that did not want to accept it. I just read Ghost Map which is about a cholera outbreak in London. Back then, the main theory for illiness was that it was caused by "bad air" which is strange considering people that worked in sewers all day where quite healthy and were not the ones getting ill. Getting doctors to wash their hands has always been an issue throughout medical history. In baseball, it is the sacrifice bunt and the stolen base that comes under fire. People still talk about batting average when it is easily shown that On base percentage is a much better indicator of performance. In football, going for it on 4th down or kicking a field goal is coming under fire by people using statistics/simulations to analyze football decision making. In areas with lots of movement/turnover/high bonus payouts, incentives can distort people's decision making dramatically. For those reasons, be careful when designing bonus programs and evaluating decisions made by people who you would think are always trying to do what is best for the team/company.

#3 and climbing!

2007-11-26T19:06:00.000-08:00

With Ohio State having a real chance now of getting to the BCS title game, I was wondering how I might calculate the %chance of them making it. My model is that if Oklahoma and/or Pitt wins, Ohio State advances to the BCS title game. Oklahoma is favored by 3 while WV is favored by 28. I would need some history to convert these spreads into percentages which I have seen for NBA basketball I think. I did find this website and it suggests using a normal curve with a standard deviation of 14 points. That gives OK a 58% of winning and Pitt a 2% chance. Therefore, Ohio State has a 59% chance to advance to the BCS title game.

This is where I estimated how to convert point spreads to win probabilities.

http://links.jstor.org/sici?sici=0003-1305(199108)45%3A3%3C179%3AOTPOWA%3E2.0.CO%3B2-V

This also has an interesting formula.

Win%=PointSpread/29.64988+.5 ( do we need all of those decimal places?)
Anyhow, this converts to a OK win % of 60%. clearly this method is better. Pitt has no chance of winning using this method.

http://www.smartcapper.com/article_converting_nfl_point_spreads_to_money_lines.html

not so dismal numbers

2007-11-17T19:06:00.000-08:00

Not the prettiest game but still great news!

School Daze

2007-11-14T20:32:00.000-08:00

Earlier this year, Bob Dyer wrote an article about the tale of 2 school districts. One school district was growing and prosperous while the other was shrinking and short of cash. Considering there is a ton of school data on the web. http://www.schooldatadirect.org/ and for Ohio http://www.ode.state.oh.us/GD/Templates/Pages/ODE/ODEDefaultPage.aspx?page=1 I thought it would be interesting to review the data.

To start with, the school district have measurable difference in average income for the people in the area. Highland averages $45k/year while Springfield averages $29k/year. That may explain the shortness of funds especially so when Highland actually spends less per student than Springfield does. $7.9k/studentyear vs $9.7k/studentyear. The main difference comes from a cost called instruction which means teachers.

The teachers in each school average about the same per year ($55k/year) so that isn't a major difference. The difference is the student/teacher ratio. Highland averages 24.5 students per teacher vs 20.5 for Springfield. Ok, looking at the data, can we see why? Turns out yes. Springfields enrollment has been declining at a rate of 48 students per year while Highland grows at 78 students per year. Highland has averaged 5 new teachers per year but Springfields teacher count has not changed in a material way since 1995. Therefore, the teacher pay is being spread over a shrinking student population which drives the average spend per student higher each year.

Strangely, it appears (regressing all public schools in Ohio) that high student/teacher ratios and low spends per student are can be a sign of a decent school. I have reviewed other districts and when there is declining enrollment there is not responding change in teacher count which drives up the spend/student and therefore it creates a financial situation for the school because it can be expected that the tax base is decreasing also.

Considering the importance of the subject, I am suprised a "Moneyball" approach has not been applied to school spending and performance measuring.

very interesting website energy data

2007-11-11T04:46:00.000-08:00

This DOE site always amazes me. Each time I check it out, I always find something interesting and useful. This is the latest section I found interesting and useful. http://tonto.eia.doe.gov/cfapps/STEO_Query/steotables.cfm?periodType=Monthly&startYear=1993&startQuarter=1&startMonth=1&endYear=2008&endQuarter=4&endMonth=12&tableNumber=20

You select the granularity you want and the data you want and download it to Excel. I plan on doing some simple gasoline analysis and post it here.

Dismal Numbers Indeed

2007-11-11T04:16:00.000-08:00

With a tear in our eye, we must do the cold calculas and analyze the Ohio State loss to Illinois. Using the model previously posted, Illinois should have won 36 to 18! while the actual score was 28 to 21. I think the major difference was the 3 interceptions. If I zero them out, the expected scores change to 30 to 23. Ohio St. still loses but the model performs much better. Judging from this, while turnovers obviously hurt OSU, Illinios could be expected to win the game.

What could be the cause for the model error? As mentioned in the Hidden Game of Football, interceptions thrown 5 yards down the field count the same as ones 40 yards down the field. In most games, the interception count is low or at the least the same on both sides. In this game, OSU had 3 vs 0 for Illinois. The 3 interceptions amounted to 71 yards total (the last one was a 51 yard pass). Adding that into the stats, makes the game 36 to 25 for an expected difference of 12. Still overstated, but much closer. I am guessing that Illinois trying to hold the ball at the end of the 4th quarter instead of focusing on scoring, may account for the rest of the model error. How does that happen by the way? 8 minutes?!?! oi...

other ideas

2007-11-07T07:47:00.000-08:00

Something that would be interesting to try is correlating current statisitics to current odds for the next game. I imagine there is a pattern and some relationships must hold when comparing good defences to good offences. Does anyone actually track historical odds?

Drive Statistics Database

2007-11-05T07:57:00.001-08:00

FYI I would be interested in a source for drive statistics for college and NFL football. Yahoo Sports has them so I am trying to learn Pearl to see if I can get them automatically.

College Football

2007-11-04T06:55:00.000-08:00

For my first post, I thought I would start with something interesting (if you are a college football fan) but not that serious. I downloaded college football stats from this website http://www.cfbstats.com/ using excel's web query. I created a model inspired by The Hidden Game of Football http://www.amazon.com/Hidden-Game-Football-Next/dp/1892129019/ref=pd_bbs_sr_1/104-5055249-0746347?ie=UTF8&s=books&qid=1194188310&sr=8-1 I have an older edition but basically I regressed average points per game in the 2006 season using these independant variables. Rushing yards, Passing Yards, interceptions and fumbles lost and gained, and penalty yards incurred and incurred by the other team. The r2 is around 80% and the standard error is 3 pts per game. I also created a much simpler model and the standard error is still around 3 so you could ditch most of the inputs except yardage gained and interceptions. Penalty yards only had a slight impact on average pts per game. Turnovers were costly though, a fumble creates a 5 pt change while interceptions created a 4 pt change. I am guessing because interceptions occur and average of 10 yards further down the field. I looked a few games and it did miss the Michigan/Michigan State game quite badly. It did model the BC loss to Fla St., interceptions being key to the game. Also interesting was how close passing yards match rushing yards. I was expecting passing yards to be more important because less time would run on the clock. It does appear that this was the case. Anyhow, enjoy and let me know what you think.

http://spreadsheets.google.com/pub?key=pwYkO8gv1ZMhdjSpMq2nnSA