Is there any method to how propagators are picked out of a field? It would seem that some players would be better propagators than others, mathematically speaking. For example (actually two examples): a player with a scoring range with less deviation makes a better comparison, and a player whose rounds occurred more recently makes a better comparison.

Just wondering if there's any method to how propagators are chosen...

Any player whose rating is over 799 and is based on 8 rounds. Nothing fancier. We're hopeful that the rating limit can be dropped to 750 when we check that out this winter. We do prevent any propagator's score from being used in a round to determine SSA if their score is rated more than 70 points below their rating.

In the original plans for the ratings system, only those with ratings over 900 were going to be allowed as propagators with their number of rounds at least 12. However, as I've mentioned before, the ratings process decisions have been biased toward member service over increased accuracy. Who can become a propagator is one of those tradeoffs. Consider that ratings are designed to group players into broad ratings ranges, not to be as precise as possible mathematically.

If propagators were limited to the original plan, it would take a long time for some members with ratings under 900, which is a large part of our membership, to ever get rated rounds especially when many times they play course layouts/tees different from higher rated divisions who would have many more propagators. Certain new PDGA locations especially emerging countries would take much longer for anyone to get ratings in those areas with fewer propagators. We have considered eliminating a few players as propagators if their standard deviation is greater than 'X' but we haven't studied that yet to figure out if it's worth it or what 'X' would be.

Here's what I mean:

Say you have 2 players rated 940. One has been playing for years, and has developed some consistency--his typical round is between 910 and 970. The other has played in 5 tournaments total, and has shot rounds rated up to 1010, but has also had some blow-up rounds rated as low as 870. Both players are rated 940, but the former makes the better propagator...

This idea holds up at any ratings level since the discriminating factor is a player's scoring deviation, not their rating. I suspect that as the rating goes up, the deviation goes down, and that would be the thing that might need to be studied and tweaked, were this kind of thing to be explored.

I think it stands to reason that a player with 20 rated rounds within the last 6 months makes a better comparison than a player of the same rating with 20 rated rounds from last year and none this year. This idea also works for any ratings range.

I understand the concept and agree from an accuracy standpoint. But the tradeoff of removing propagators with higher standard deviations is offset by the benefit of more propagators from a stats standpoint. If you justify this move, it would then justify not having any propagators below a higher rating level than we do know if you were looking for a certain level accuracy in all of our decisions made to produce ratings. The fact that we use as few as five props to do the calculations makes me uneasy from a stats standpoint. But like I said before, the decision is based on customer service versus the accuracy tradeoff. All in all, I think most feel the policies and the results are reasonable.

There are enough events with more than sufficient large numbers of players where the accuracy is more than acceptable for any stats person that it improves the iffy values the more you play. The more you play the more your rating is used as an Am. So the ratings of those most active tend to be the most accurate which is what you would want.

This sort of process (scoring deviation as a discriminant) doesn't have to prevent players with lower ratings from being propagators. I'm not talking about using only players with the lowest deviations as propagators--I'm suggesting we use the players with higher consistency and "recentness" factors at all ratings ranges. I do suspect there is a correlation between rating and deviation, but that doesn't matter too much. My idea is to pick the best propagators for all ranges.

Rob-

The one issue that you aren't considering is that you aren't getting to choose who plays the event.

Consider this scenario:

I have a new course and I need to determine an SSA for the course and I have 6 players to choose from. 3 who for the last year have shot every round between 925 and 975 and 3 players that shot between 900 and 1000.

If you tell me that I have to choose only 3 players to get my number, then obviously I'll choose the first 3 and get a more accurate number. But what if I can choose to have all 6 play the course?
Because I've doubled the number of data points, I think I'm more likely to get a good answer.

I would actually have to go back and look at the math a lot harder than is warranted on a Saturday morning. I'd guess that at some standard deviation adding 3 less consistent golfers will make things less accurate, but I'm not sure that the average distribution of golfers is that widespread...

The sample size is the issue in your example. Anybody that knows anything about stats knows the relevance of sample size. I think Chuck has said in other posts that he prefers at least 5 players to generate ratings. At any rate, if only 2 players showed up for your tourney, I'd use the more consistent player to generate ratings for the other...

Also, I thought course SSAs came from the ratings....do you have the cart behind the horse in your "I have to figure the SSA" example?

As I stated in my last post, I'm suggesting that we use the best props available. In my dream world, I'd only prevent a few players from being props at any given event. Maybe you guys think I'm suggesting that being a propagator for statistical evaluations is some sort of elitist perk that should only be available to the few. Wrong. In Utopia, everybody can be a propagator. I'm just saying we should weed out the statistically insignificant, just like we already do with rounds 100+ points lower or 2(?) deviations lower than your rating.

Happy Monday!

The sample size is the issue in your example. Anybody that knows anything about stats knows the relevance of sample size. I think Chuck has said in other posts that he prefers at least 5 players to generate ratings. At any rate, if only 2 players showed up for your tourney, I'd use the more consistent player to generate ratings for the other...

Also, I thought course SSAs came from the ratings....do you have the cart behind the horse in your "I have to figure the SSA" example?



Chuck can correct me if I'm wrong, but:

I think Chuck has said that statistically, 5 propagators makes him cringe, he'd prefer 10+ propagators...

The SSAs do come from the ratings.
-I have the established ratings of all of the propagators coming into the tournament.
-I use their ratings and their scores to establish what, based on each of their performances a 1000-rated golfer should shoot.
-The average of those numbers is what the SSA is set to.
-That SSA is then applied to what each player did shoot and is used to give that round for that player a rating.



As I stated in my last post, I'm suggesting that we use the best props available. In my dream world, I'd only prevent a few players from being props at any given event. Maybe you guys think I'm suggesting that being a propagator for statistical evaluations is some sort of elitist perk that should only be available to the few. Wrong. In Utopia, everybody can be a propagator. I'm just saying we should weed out the statistically insignificant, just like we already do with rounds 100+ points lower or 2(?) deviations lower than your rating.

Happy Monday!



I understand exactly what you are saying and don't think it is elitist at all. The problem is that unless someone is a really, really inconsistent golfer compared to the rest of the field, you get better numbers by including the inconsistent golfers and getting the extra data points.

Is there any method to how propagators are picked out of a field?




Police lineup :cool: