Methodology is described here

Power/Score Calculations

Let me explain what runner power is and how one's mind can be wrapped around it.

Power always existed in every ranking algorithm calculations, but hardly ever been surfaced. I think power is very interesting number and it helps me a lot in evaluating somebody's performance. To distinguish power from ranking score I always show power in brackets (100) throughout the report pages.

While converging, "Algorithm" asigns each person one number (Power), which is average of ALL his recorded scores. Algorithm uses power to evaluate race Gnarliness and perform all other steps until converged.

Race score for each race is calculated by dividing race GV by the runner time in minutes. Of course all scores will be different. Faster person gets higher score. By definition if one takes ALL his scores and averages them this will be his power as used by algorithm.

Ranking scores can be calculated using various algorithms from the race scores. Most standard one is "trailing": where four best races are counted and then "trail" is added as a half of remaining races. In both cases if number of races is less then four runner officialy doesn't qualify, but his score is average of existing scores

Currently I have two implemented (last drop-down) T(default) stands for Trailing score. ~T - no trail option only counts four best races. BAOC Sprint series uses third scoring system where course setter is awarded once by his best score of the season and you have to have 5 finished races to qualify (or 4 + course setter award)

While Power is crusial for algorithm convergeance and is genuine characterstics of the person's performance from the Algorithm point of view, ranking score have nothing to do with convergeance of the algorithm at all. Ranking score is calculated from race scores and depend on the races selected. Ranking Scores are generally higher then power, because best races are chosen, but Ranking scores can be lower than the power in some very interesting cases.

For those who like examples

Example 1: if I ran 6 races with scores (100,100,100,100,50,30), my power will be 480/6 = 80, my ranking score (trailing) will be 450/5 =90. NOTE that at the same time if first 4 races I ran were A-meets and last two were local meets - my USOF Ranking score will be 100, as 50 and 30 will not be even in the picture

Example 2: if I ran 8 races with scores (100,100,100,100,50,50,50,50), my power will be 600/8 = 75. Assume now that first four races were longs and last four were sprints My "Total" ranking score will be 500/6 ~ 83. My "Long" score would be 400/4 = 100!. While my "sprint score will be 200/4 = 50!

Performance alert: Above examples emphasize important point as many people tend to worry about how local race or Sprint race performance is dropping their ranking... All you need is to pick right races and calculate score using them - and you got right adequate score!

HINT: If you compare you ranking score with your power, you might notice that sprints are your weak (OR strong) point, while local events rating and A-meets are the same (OR local are much worse)

Normalization details

As one might notice, scores and powers are relative quantities, defined up to a multiplication constant. This means only ratio of scores have sence. Person with score 100 is twice fast as person with score 50 and twice as slow as person with 200. This raises issue of reasonable and stable normalization. I did research on this topic and found that (currently used by USOF) normalization to top 3 runners having 100 points in average can be improved.

All runner Powers in the database are calculated using standard algorithm and normalized so that average runner has power 50. One can check it by going to the Absolute power rating page Copy text file into Excel and calculate average of power column.

There was a discussion about reasonability of the power calculations for all runners at the same time or for each class of runners separately or separately for White-Orange and Brown-Blue. I finally decided to combine everybody into the same pool.

Because all runners are in the same pool, the power is absolute over all events and classes. Power changes as you run more races and your old races are getting obsoleted. Flip first drop down on Power report Page and watch your power recalculated from EOY2008 to current. Compare with your feeling about your performance. My score grown couple points up during last year and it well corresponds with my "very slowly getting in shape" feeling

Number of runners is quite large (2000+) and average number over the population is very stable, so I expect no sudden jumps in runner scores

As always runner with power 100 runs on average twice as fast as average runner in the database during the period in question. We do have a few runners with power in 130+ range. They are about 130/50 = 2.6 times faster than average runner.

It can be noted that fastest women has score of 100+ which makes them about 30% slower than fastest men

Many other interesting discussions can be started based on the database results presented

Here is the link to typical Power Curve (PDF) (last updated 2009/05/28). All runners in the database who run at least one race during a year back from last update moment are taken into account.
Zoomed version for top runners is here (PDF)

Case Study: Positioning of top White runner in overall ranking

Yes, of course there will be some weird-at-first-glance results like our best F-10 J.D. runner placed around place 1000+ with score 50+.

Some people were worried that:

"Take two runners who run the same speed, one with 20 years orienteering experience and one who has run only white courses. Put them on a white course and they will run the same time © feet

My analysis show that White course is close to a short sprint. Expect slightly lower scores as from every race with low GV and grant our top girls and boys high scores earned in competition with Big Runners. And do not forget to add one more stone in the baskets of UNIFORMITY, CONSISTENCY and CONTINUITY - no need to separate white; scores for the Big Runners at White races are the same as at other events; and juniors transfer to higher ranks seamlessly

Effect of Sprint Events on overal runner Score

There is constant discussion about Sprints spoiling the Classic scores by being lower on average. In this case it seems reasonable for top runner not to run Sprint at all or request it being not counted.

Here I present actual numerical proof that Sprint scores are generally lower than the Classic ones. One can see that this statement is especially true for top blue runners. Several (not many) people clearly show this "winner pattern". For the rest of us it is more or less irrelevant. I did not go into big effort counting, but one can do using following preselected searches. I recommend opening them in separate windows and compare side by side

Sprint + Classic, Blue; compare with Only Sprint, Blue and probably compare with Only Classic

Sprint Events as part of general picture. Effect of race GV on the Winner Score

It rarely happens that there are seven different sprint courses during the event. Recently, typical course pattern is: Sprint 1 for Blue/Red, Sprint 2 for Green-Orange and Sprint 3 for White-Yellow. Other configurations are possible. In order to provide CONSISTENCY I made a desision to combine all runners running physically same course into the same pool, indicate what courses sprint was designed to and color the sprint with highest allowed course color. Green-to-Orange will be Green and Yellow to White will be Yellow. Blue to White will be Blue.

Let's analyze how race GV affects winner score. This will help us explain why one should or should not run sprint if one wants to maximize score

Here is a PDF with graph showing how Winner score depend on race GV

In general there is a trend for winner to get higher score with increased race gnarliness. Sprints usually have lower gnarliness, therefore they generally bring lower score to the winner. But there is no visible jump between sprints and other races. More general statement will be true: The more difficult course is, the higher winner score will be (under proposed system where normalization is done for the whole population)

Most probably the same thing happens currently for each of the courses Blue to Brown with the only difference that each color is normalized to its own 100 points. I present separate plots for Blue, Red and Green subsets of races. Each color has tendency to climb up with gnarliness