BGonline.org Forums

42P-43: more data

Posted By: Nack Ballard
Date: Sunday, 14 February 2010, at 2:13 a.m.

In Response To: 42P-43: more data (Daniel Murphy)

>> (3) This is a reminder of how much variance there still is in 5k or 10k trials (whatever bots may claim about the confidence interval).

I don't disagree with anything you wrote, Nack, but would point out that because of the confidence intervals, the disparate Snowie results shouldn't be too surprising. Snowie doesn't report the 95% confidence interval for the Janowskie-adjusted equities, but I think we can assume that they are about the same as the live cube equities...

This seems to be a commonly held assumption, but actually, no. Live cube equities have a considerably higher confidence interval than Janowski-adjusted equities; a factor of 2:1 or more. Open a live cube rollout and compare them (i.e., check the info in Snowie's yellow box) and you'll see what I mean.

Here, the (Janowski) confidence interval for one of Stick's (5k) plays is reported to be +/-.007 and the other +/-.006. The (Janowski) confidence intervals in Paul's rollout for both (10k) plays is reported to be +/- .005.

By deduction, it's even tighter than that. I think it's reasonable to assume that Stick's play intervals were both close to .0065 and rounded in opposite directions. Dividing that by sqrt2 means that Paul's .005 was rounded up from .0046. In other words, my best guess for the (alleged) intervals to another decimal place are +/-.0065, +/-.0065, +/-.0046 and +/-.0046. (In any case, they're a far cry from +/-.014, and I'm sure you can see the effect on the math.)

This sort of difference in independent rollout results isn't commonplace, but it happens often enough that I (and Paul, and Stick) stopped believing the confidence intervals that the bots report long ago.

Cheers,

Nack

Messages In This Thread

• 42P-43 update
  neilkaz -- Friday, 12 February 2010, at 7:34 p.m.
  • 42P-43 At Match Scores
    Paul Weaver -- Friday, 12 February 2010, at 10:31 p.m.
    • 42P-43 At Match Scores
      Stick -- Saturday, 13 February 2010, at 6:32 a.m.
      • Weaker vs. Stronger ~ GS vs. GG?
        jdg -- Saturday, 13 February 2010, at 2:17 p.m.
        • Weaker vs. Stronger ~ GS vs. GG?
          Stick -- Saturday, 13 February 2010, at 3:53 p.m.
        • Weaker vs. Stronger ~ GS vs. GG?
          Rich Munitz -- Saturday, 13 February 2010, at 7:20 p.m.
          • Weaker vs. Stronger ~ GS vs. GG?
            jdg -- Sunday, 14 February 2010, at 1:54 p.m.
    • Are we sure weaker vs stronger matters here?
      neilkaz -- Saturday, 13 February 2010, at 4:36 p.m.
      • Are we sure weaker vs stronger matters here?
        Stick -- Saturday, 13 February 2010, at 4:46 p.m.
        • Are we sure weaker vs stronger matters here?
          David Rockwell -- Saturday, 13 February 2010, at 5:36 p.m.
  • 42P-43: more data
    Nack Ballard -- Saturday, 13 February 2010, at 3:07 a.m.
    • 42P-43: more data
      Daniel Murphy -- Saturday, 13 February 2010, at 6:04 p.m.
      • 42P-43: more data
        Nack Ballard -- Sunday, 14 February 2010, at 2:13 a.m.
        • 42P-43: more data
          Chuck Bower -- Sunday, 14 February 2010, at 5:23 p.m.
          • 42P-43: more data
            Nack Ballard -- Sunday, 14 February 2010, at 7:46 p.m.
            • 42P-43: more data
              Chuck Bower -- Monday, 15 February 2010, at 8:47 p.m.
              • 42P-43: more data
                Nack Ballard -- Monday, 15 February 2010, at 11:51 p.m.
        • 42P-43: more data
          Timothy Chow -- Sunday, 14 February 2010, at 5:49 p.m.