Amended Nationals Appeals Policy

[nicholasmaurer]

I do wish there was a better process for handling these types of situations. If I remember correctly, your scoring system does include some useful metrics than can help spot outlier data (e.g. if an event should be ranked with scores low-to-high, but is accidentally entered as the reverse).

However, I wonder if some additional review could be done if, for example, a team that is otherwise placing consistently in the top 10 has one event entered where they placed 50th. Is it possible they genuinely did that poorly? Sure. Anyone can have an off day. But I would think that result would be something worth reviewing to make sure there wasn't a clerical error along the way. This is less practical at invitationals, but since State and National Tournament awards are typically at a set time later in the evening, it is more possible there. Just a thought - maybe this already happens behind the scenes and I'm just not aware!

We do indeed look for major issues like reverse sort order, unexpected No Shows, etc. However looking for more nuanced outliers like a top 10 team placing low in an event is much harder. For example, if you look at the Div C nationals result from last year, you'll see that:

1. The gold medal team had 2 events they placed in the 20's
2. The silver medal team have 2 events they placed in the 30's
3. The 6th place team had 3 events in the 20's, 1 in the 30's and 1 in the 50's
4. The 8th place teams had 2 events in the 20's and 3 in the 40's
5. Conversely, the 38th place team got a silver medal in one event

I don't see an obvious algorithm we can implement beyond our normal process of carefully checking things multiple times. Of course if someone has a good idea, I'd be happy to try to implement it.

I would have to test out some different approaches to try and find an algorithm that doesn't turn up too many false positives, because that rapidly becomes too resource intensive for the scoring staff. Intuitively, I would start by looking for situations where dropping a team's worst event significantly reduces the standard deviation of their placements. It would likely have to be refined from there.

[nicholasmaurer]

I do wish there was a better process for handling these types of situations. If I remember correctly, your scoring system does include some useful metrics than can help spot outlier data (e.g. if an event should be ranked with scores low-to-high, but is accidentally entered as the reverse).

However, I wonder if some additional review could be done if, for example, a team that is otherwise placing consistently in the top 10 has one event entered where they placed 50th. Is it possible they genuinely did that poorly? Sure. Anyone can have an off day. But I would think that result would be something worth reviewing to make sure there wasn't a clerical error along the way. This is less practical at invitationals, but since State and National Tournament awards are typically at a set time later in the evening, it is more possible there. Just a thought - maybe this already happens behind the scenes and I'm just not aware!

We do indeed look for major issues like reverse sort order, unexpected No Shows, etc. However looking for more nuanced outliers like a top 10 team placing low in an event is much harder. For example, if you look at the Div C nationals result from last year, you'll see that:

1. The gold medal team had 2 events they placed in the 20's
2. The silver medal team have 2 events they placed in the 30's
3. The 6th place team had 3 events in the 20's, 1 in the 30's and 1 in the 50's
4. The 8th place teams had 2 events in the 20's and 3 in the 40's
5. Conversely, the 38th place team got a silver medal in one event

I don't see an obvious algorithm we can implement beyond our normal process of carefully checking things multiple times. Of course if someone has a good idea, I'd be happy to try to implement it.

I ran some quick calculations using the placings from the 2017 National Tournament. For each event placing, I calculated the percent change in that team's overall standard deviation if you excluded that event.

You can set various thresholds, but if you find all cases where a team's standard deviation falls by more than 15% because you exclude that one event placing, you identify 16 cases for extra review. Nine of these cases are teams who placed substantially better than their mean placing; I'd be inclined to ignore these. This leaves seven outliers where a team did unusually poorly. One of these is a no-show, which you are already verifying. That would leave six individual scores for the scoring team to double-check.

[nicholasmaurer]

For those who care, the seven specific events my algorithm flagged were:

#7 Northville - 38th Ecology
#9 New Trier - 53rd Electric Vehicle
#11 Harriton - 31st Experimental Design
#20 Clements - 50th Write It, Do It
#24 Munster - 48th Optics
#38 ABRHS - 60th Electric Vehicle
#50 Clark - 54th Write It, Do It

[DarthBuilder]

For those who care, the seven specific events my algorithm flagged were:

#7 Northville - 38th Ecology
#9 New Trier - 53rd Electric Vehicle
#11 Harriton - 31st Experimental Design
#20 Clements - 50th Write It, Do It
#24 Munster - 48th Optics
#38 ABRHS - 60th Electric Vehicle
#50 Clark - 54th Write It, Do It

Do we know the specific reasons for each flagged?

[chalker]

For those who care, the seven specific events my algorithm flagged were:

#7 Northville - 38th Ecology
#9 New Trier - 53rd Electric Vehicle
#11 Harriton - 31st Experimental Design
#20 Clements - 50th Write It, Do It
#24 Munster - 48th Optics
#38 ABRHS - 60th Electric Vehicle
#50 Clark - 54th Write It, Do It

This is very interesting... I'll have to think about how I'd actually go about potentially implementing it to automatically check within excel - any idea what the formula might be to accomplish this? Of course one issue is that it will only catch single event mistakes of a certain threshold... Of the list I previously provided, I think it only flagged 1 of the teams. However even incremental improvements are still improvements!

[Unome]

Just dropping this link here, in case anyone finds it useful. I came up with this specifically because I found standard deviation to be unhelpful for SO purposes.

[Ashernoel]

For those who care, the seven specific events my algorithm flagged were:

#7 Northville - 38th Ecology
#9 New Trier - 53rd Electric Vehicle
#11 Harriton - 31st Experimental Design
#20 Clements - 50th Write It, Do It
#24 Munster - 48th Optics
#38 ABRHS - 60th Electric Vehicle
#50 Clark - 54th Write It, Do It

I can confirm that the 53rd in EV was due to a "construction violation" that was appealed but rejected.

The ruling was disgusting, relying not on proof but the word of the ES, and was based on an incredibly unclear section of the rules that had gone uncalled all year.

The run was also one of their best all year, and would have resulted in New Trier placing 6th or 7th overall.

The arbitration committee completely missed the ball and the ruling shows a failure in the national arbitration system.

[dxu46]

For those who care, the seven specific events my algorithm flagged were:

#7 Northville - 38th Ecology
#9 New Trier - 53rd Electric Vehicle
#11 Harriton - 31st Experimental Design
#20 Clements - 50th Write It, Do It
#24 Munster - 48th Optics
#38 ABRHS - 60th Electric Vehicle
#50 Clark - 54th Write It, Do It

What about #21 Ladue - 39th Electric Vehicle?

[Name]

For those who care, the seven specific events my algorithm flagged were:

#7 Northville - 38th Ecology
#9 New Trier - 53rd Electric Vehicle
#11 Harriton - 31st Experimental Design
#20 Clements - 50th Write It, Do It
#24 Munster - 48th Optics
#38 ABRHS - 60th Electric Vehicle
#50 Clark - 54th Write It, Do It

What about #21 Ladue - 39th Electric Vehicle?

The only 30somethings in the list is northville harriton, 2nd and 4th overall. Thier high ranking is probably why they were caught and ladue wasn't high enough for a 39th to count

[nicholasmaurer]

For those who care, the seven specific events my algorithm flagged were:

#7 Northville - 38th Ecology
#9 New Trier - 53rd Electric Vehicle
#11 Harriton - 31st Experimental Design
#20 Clements - 50th Write It, Do It
#24 Munster - 48th Optics
#38 ABRHS - 60th Electric Vehicle
#50 Clark - 54th Write It, Do It

This is very interesting... I'll have to think about how I'd actually go about potentially implementing it to automatically check within excel - any idea what the formula might be to accomplish this? Of course one issue is that it will only catch single event mistakes of a certain threshold... Of the list I previously provided, I think it only flagged 1 of the teams. However even incremental improvements are still improvements!

From the list you started with, it flagged the silver medal team placing 31st and the 6th place team placing 50th. It also technically flagged the 38th team placing second, but I excluded this as one of the places where a team did unexpectedly better. That list of seven is only teams where the outlier was worse than their mean.

In terms of calculation, I did all of my work in Excel. The formula is relatively straightforward. Here is a link to the math.

Just dropping this link here, in case anyone finds it useful. I came up with this specifically because I found standard deviation to be unhelpful for SO purposes.

Standard deviation may not be the best measure, but it's reasonable for this application and less arcane than the alternatives.

What about #21 Ladue - 39th Electric Vehicle?

Removing this event only lowered their standard deviation by 8%, so it didn't cross the 85% threshold I used. This is likely because they had many other events where they placed in the 20s and 30s.