Jump to content
Brian Enos's Forums... Maku mozo!

lstange

Classified
  • Content Count

    154
  • Joined

  • Last visited

About lstange

  • Rank
    Looks for Target

Profile Information

  • Real Name
    Louis Stange

Recent Profile Visitors

The recent visitors block is disabled and is not being shown to other users.

  1. To get classifier percent, your hit factor is divided by high hit factor for this classifier and division. The way USPSA determines high hit factor is not completely disclosed. For 19 series classifiers, it's the average of the top 10 results shot at 2019 Nationals where those stages were first introduced (except for Revolver, where they average fewer). For older classifiers, USPSA looks at historical data, throws out outliers (by some definition of outlier), then averages top-something remaining. Except for Carry Optics, where Production numbers are used instead.
  2. Seven weeks. Yes, cut out with scissors from some old plastic packaging. Seems to work fine, I ran a few thousand rounds with no visible shift.
  3. Yes, but people would shoot it with game-ready CZs and Tanfoglios rather than Glocks you can buy at a local gun store. That's what Production was supposed to be, and look what it became.
  4. Now there is. This is what just came back from service. Note different board color (blue instead of green), logo placement, and rotated center contact. I had to change the shape of the stopper to fit the new geometry.
  5. There is no new style board. When I sent the AF series DPP to Leupold, it came back with a note saying "replaced circuit board", but that "new" board looked exactly like the old board, with the same sticker.
  6. It lasted 10k rounds. A week ago started to lose brightness and occasionally turn itself off, sometimes coming back on after some shaking. Fresh battery, all visible contact surfaces are clean, the central contact did not slip. Sent back to Leupold.
  7. I had one lot of Syntech 124 grain 9x19 with PF=134 and another with PF=152. Same SKU, same gun, similar temperature, just different lots.
  8. On closer inspection it appears that the formulas used to determine HHFs for the 19 series classifiers are the same as for the 18 series, but there are at least two different formulas. In Production, HHF is the average of top 10 HFs. In Revolver, it's the average of top 10 HFs after excluding HFs that are below 90% of the stage winner. Also worth noting is that 19 series HHFs are different between CO and Production. Prior to 19 series CO mirrored Production. Maybe because 19 series classifiers are 12 or 14 shots with no mandatory reload, so a larger magazine gives an advantage.
  9. This calibration issue is easily addressed by converting classification percent to classification percentile. If your classification percent is x%, then classification percentile is approximately pnorm(x, mean=56.6, sd=17.9). USPSA can do it even better using actual distribution rather than normal approximation. Note that this transformation is monotonic, so it preserves rank ordering and does not affect discriminative power.
  10. Only if you try to game it. And there is no point in grandbagging, you'll be only fooling yourself. I certainly don't want to be like that now famous GM who finished 67th of 118 with 62% of the winner in area match. If you study math, you become better at math and your standardized test scores go up. If you study to pass standardized tests, your test scores can go up even faster, but you are not better at math as a result.
  11. The slope of the regression line is less than one. C class shooters get about 50% match percent at the majors, but for B class and above match percent is lower than classification percent. One would expect to see such bias knowing the way classification is calculated. There are no re-shoots at major matches, and all stages count. But rank ordering is roughly preserved. if you order, say, shooters at the Nationals by their classification percent and take 42nd from the top, chances are he'll finish around 42nd place.
  12. The dependent variable was match percent at major matches, but it determines match place so the story is the same.
  13. Nah, it's good enough. Existing classification percent predicts major match placement with mean absolute error of 8.6%. The best I could do (using trimmed mean classification percentile) was 7.5%. This juice is not worth the squeeze. There are a few classifiers that are unusually hard, e.g., 03-11, 03-12, 09-03 in Production. But they are not nearly as lopsided as in your example. Apparently USPSA throws out robocop runs when setting HHFs.
  14. We already have that, it's called classification percent. No matter how you slice it into six bands, you inevitably lose information. A percentile would be even better. Without knowing the shape of the distribution it's hard to say whether 42 is a good classification percent, but "42nd percentile" has an intuitive explanation - in a match with 100 other shooters selected at random you would expect to beat 42 and be beaten by 58. And it's not hard for USPSA HQ to do, just one SQL query. That's by design. Results 5% below the bottom of your current classification band are not used in the average, two worst results out of six are discarded, if classifier is re-shot then the better result counts, all those things bias classification percent up. As long as everyone knows it and is treated the same, it's not a big deal. No matter how hard a B class shooter tries to hero-or-zero, he's still not making a GM. Ask me how I know
  15. It's very noisy, single digits. Not enough to draw conclusions. Not that many people make it to GM (who would have thought).
×
×
  • Create New...