Elo ratings calculations - why is home advantage 100 points

[Vilamoura04]

In the traditional Elo ratings calculations it is generally accepted that the home team gets a 'boost' of 100 points to account for home advantage. Why 100 points? Is this based on anything? A round 100 sounds more than a little bit arbitrary to me. Why not 80 points or 120 points? I have read papers where authors suggest 85 points is more accurate for a MLS Elo system, which I assume reduces the home advantage a team gets. Which is strange because the ML tends to have a big home advantage. Is there a way that this number could be calculated, as there are big discrepancies between countries and divisions in countries as to how much of an advantage playing at home gives.

[a1ehouse]

Re: Elo ratings calculations - why is home advantage 100 points I calculated that there were substantial differences between leagues and countries when I developed my ratings. Previously I used 100 for all home team advantage, but tried to tighten the model using division-specific points advantage. See here for details: http://www.punterslounge.com/forum/f21/elo-access-function-75510/#post1457007

[ELO Evangelist]

Re: Elo ratings calculations - why is home advantage 100 points IMO the 100 home advantage adjustment is arbitrary. It probably comes about because the Chess rating ELO models use a similar arbitrary value to adjust for their home advantage equivalent, white advantage. Because white advantage is more of an advantage at higher levels of play the figure changes depending on the level of play. The equivalent in soccer would be to say that the advantage figure changes depending on what league the teams are playing in. One advantage that soccer rating has over chess rating is that soccer has a lot more collateral form than chess and so I calculate separate ELO ratings for home and away performance. IMO you can not have an accurate overall value that gets added to the home team. Here are some examples of the difference between home and away ELO rating shown as a percentage that some EPL teams have. Chelsea 2.8% home advantage Man Utd 4.7% home advantage Arsenal 6.8% home advantage Liverpool 7.7% home advantage You will notice that it does not seem to make sense that as a teams ability increases it's home advantage decreases. This is because I have found that the notion of home advantage is misleading. What is actually happening is that teams suffer from away disadvantage not home advantage. But that is another story...

[Vilamoura04]

Re: Elo ratings calculations - why is home advantage 100 points I tried separating home and away ratings. In theory it makes absolute sense. I was very disappointed to find that results were worse, across 10 seasons of data and two separate divisions. Yield was down nearly 50%. Sticking to just using one rating I could improve things by changing the 100 figure, but it all felt so arbitrary and adjusting to fit that it didnt give me any confidence that it would work in the future.

[a1ehouse]

Re: Elo ratings calculations - why is home advantage 100 points Vil, it might be influenced by a couple of things like the points the team start with eg 1000 pts; and whether this is reset season after season or whether it's rolled over. If you're using a relatively small starting number then 100 pts might skew more. Bear in mind that because the advantage is only one small part of the equation the overall impact may not be as big effect as perceived.

[ELO Evangelist]

Re: Elo ratings calculations - why is home advantage 100 points When I first started my ELO calculations I found that they were not accurate and one obvious reason was that I was using a combined home/away rating that would need a home advantage adjustment to make it accurate. I was unable to find this figure and so decided to calculate two ELO ratings for each team, home and away, and do away with the home advantage issue completely, but still I could not get the ratings accurate. I noticed that even with a separate home and away rating there was still a home advantage skew present. When two teams played I would subtract the away rating from the home rating and come up with a rating difference which was used to calculate the predicted outcome of the match. When the rating difference was negative it meant that the away team had the ability advantage and when the rating difference was positive the home team had the advantage. Looking at the figures it became obvious that the home team has the advantage in the majority of cases. I decided to plot the Away advantage matches (the minority) on a separate graph to the home advantage matches (the majority). The results are shown in the graphs below: http://216.130.182.116/PuntersLounge/HomeAdvantage.htm http://216.130.182.116/PuntersLounge/AwayAdvantage.htm You will notice that the intercept point (the point at which the graph meets the zero μ rating difference) for the Home advantage graph is 31.9 and for the away advantage graph is 31.1, also that the slope of the two graphs is not the same. In theory the intercept point and the slope of the two graphs should be the same. I concluded that because the away advantage matches were in such a minority that the low sample size was causing inaccuracy and combining the two sets of values into the same graph was creating an inaccurate average value and therefore prediction. With the home advantage matches separated from the away advantage the home advantage match predictions now became more accurate than the away advantage matches due to the larger sample size. So in summary I would suggest: 1: Calculating separate home and away ratings. 2: Plot home advantage matches separate from away advantage matches.

[salford_student]

Re: Elo ratings calculations - why is home advantage 100 points

Is there a way that this number could be calculated' date=' as there are big discrepancies between countries and divisions in countries as to how much of an advantage playing at home gives.[/quote'] yes, if you want the "average" value for a league or a division. step 1) get the league table of the league you are interested making sure it is split into home/away format step 2) sum all the homes games, sum the wins and sum the draws. e.g. for premiership last year, there were 360 home games, which resulted in 193 wins and 96 draws step 3) using the chess method of awarded draws as half a win, i.e. 0.5. work out the proportion of home wins. e.g. for premeriship last years [193+(0.5*96)]/360 =0.67 step 4) using the elo formula work out what rating difference gives a prob of 0.67. the formula you need is R=400*log(1/p-1). e.g for the premiership last year this equalled 123

[Vilamoura04]

Re: Elo ratings calculations - why is home advantage 100 points Thanks for that. I ran the calcs. Resulted in reduced strike rate and significantly reduced yield. Why would that be?

[Vilamoura04]

Re: Elo ratings calculations - why is home advantage 100 points

So in summary I would suggest: 1: Calculating separate home and away ratings. 2: Plot home advantage matches separate from away advantage matches.

Ive done the first bit. I have separated each teams home and away ratings so they have a different rating for each and then for each match it is the home teams home rating vs the opponents away rating. I also find that this predicts a lot more home wins than away wins, and also the home wins are much more profitable. But still significantly less profitable than the averaged rating. I really want to believe this is the way forward because it makes perfect sense, but I cant get the figures to add up. Am I missing something? I wasnt quite sure what you meant by point 2.

[ELO Evangelist]

Re: Elo ratings calculations - why is home advantage 100 points Sorry for not getting back to you earlier, I have been enjoying the delights of a Burmese casino. 1. Separate home and away ratings so that each team has two sets of ratings. 2. For past results calculate the rating difference between the home team's home rating and the away team's away rating. 3. When the rating is positive this is a home team advantage match put all these results in one data set and when the rating difference is negative the away team has the advantage put all these results in another data set. 4. For each data set group the results by rating difference. So if you have 1000 home advantage results with rating differences from 0 to 200 split the results into 20 sub sets from 0-9.9, 10-19.9, 20-29.9 etc. 5. For each sub set calculate the number of wins, losses and draws as a % and plot the three results on a graph of rating difference by %. The results should look something like this: http://216.130.182.116/PuntersLounge/EnglishPremier310810.htm This is for Home team advantage matches you should create one for Away team advantage matches also.

[Vilamoura04]

Re: Elo ratings calculations - why is home advantage 100 points Thanks for the reply and I hope you left the casino with enough to pay their staff. I have done all that up to the charting. Im getting my predicted home win % for an upcoming match directly the observations of matches with similar match ratings in the past - so it isn't smoothed over by a line of best fit. I understand you are suggesting taking the estimated home win % from the line of best fit instead. I would be very interested in seeing if this increases profitability. Is there an excel function that allows me to get the result from a data point along the trendline once charted? Eg how would I find the home win % of a game with a +75 match rating from the trendline without putting a ruler up against my screen like a twat!!!! Is linear or polynomial distribution better? I get a better r2 from polynomial but they arenoth around .73 which isnt brilliant?

[Machine]

Re: Elo ratings calculations - why is home advantage 100 points When you click on add trendlines go to options and tick display equation on chart this will give you your y = mx + c equation.(x being the rating you want to check). :ok A linear trendline should best serve the type of data you have.

[ELO Evangelist]

Re: Elo ratings calculations - why is home advantage 100 points

Thanks for the reply and I hope you left the casino with enough to pay their staff....

I would not bet in a casino if you stuffed free money into Lucy Liu's panties and told me to bet with it, I was only there for a meeting/junket.

...putting a ruler up against my screen like a twat!!!!

LOL been there done that. What you are looking for is the LINEST excel function, here is a tutorial: http://www.scribd.com/doc/11361260/Excel-Tutorial-Using-LINEST-function-Plotting-a-graph-Adding-Error-Bars And here is one of my English Premier graphs using the function: http://216.130.182.116/PuntersLounge/EnglishPremierXLS020910.htm As for Linear vs polynomial? You can see that the linear R squared is good but clearly as the rating difference increases the trend line wants to go above 100% and below zero which suggests a non linear curve. I usually stick to linear trends discarding the last 10% of the data (the highest rating differences). But... I know the future lies in calculating an accurate polynomial fit. The problem for me is that the maths required for a linear fit are within my maths abilities and so I write my own software to take advantage of this the maths required for a polynomial fit are beyond me but I am in the process of mastering it and will then code a non linear solution.

[ELO Evangelist]

Re: Elo ratings calculations - why is home advantage 100 points

When you click on add trendlines go to options and tick display equation on chart this will give you your y = mx + c equation.(x being the rating you want to check). :ok A linear trendline should best serve the type of data you have.

Yes forgot to say how the formula is used. If you look at my graph you will see for the 'chances of a win plot' (blue), the formula is y=7.9668X + 33.891. Y = the % chance. X = the rating difference. 7.9668 = the slope. 33.891 = the intercept. So with a rating difference of 3.0 (my ELO rating system values): (7.9668 * 3.0) + 33.891 = 57.7914%

[FleetFanatic]

Re: Elo ratings calculations - why is home advantage 100 points Found this site earlier, and it's mention of ELO ratings made me think of you guys. http://www.prematches.com/

[ELO Evangelist]

Re: Elo ratings calculations - why is home advantage 100 points

Found this site earlier, and it's mention of ELO ratings made me think of you guys. http://www.prematches.com/

Cheers. I always like comparing other peoples ratings to my own. This site is for real stats freaks.

[FleetFanatic]

Re: Elo ratings calculations - why is home advantage 100 points

This site is for real stats freaks.

Understatement of the year. :lol I'm amazed that I hadn't come across that site before, but it looks quite amazing.

[lauriejay]

Re: Elo ratings calculations - why is home advantage 100 points I've got the rating system sorted and have subtracted the away rateform from the home rateform but how do i get the calculation for the correct values for a win, home,away or draw??

[JohnPR]

Re: Elo ratings calculations - why is home advantage 100 points

I've got the rating system sorted and have subtracted the away rateform from the home rateform but how do i get the calculation for the correct values for a win' date=' home,away or draw??[/quote'] When you say you have subtracted away ratings from home ratings do you mean you have two tables. One fpr away ratings and one table for home ratings.