Sounds good, except I recommend using a clear, pre-defined way of scoring the guesses.
The most straightforward way, imo, is to either:
1) Count each digit match as one point, and tally it up for the three digits. (giving a score between 0 and 3)
2) Tally up the distance between real and guessed for each digit, as a total "deviance" (inaccuracy) score. (between 0 and 27)
This is helpful because you can then compare your scores each night precisely/robustly, as well as compare it directly with the scores that you'd get from complete randomness.
For example, if you use scoring-system 1, random chance would give:
* A 27% chance of getting one or more points/digit-matches, each attempt. (1 - (.9 * .9 * .9) = .271 = 27%)
* A 0.1% chance of getting 3 points/digit-matches, each attempt. (.1 * .1 * .1 = .001 = 0.1%)
(I'm not a statistician, so don't know/remember how to calculate the probability of 2 or more point/digit-matches, apart from doing brute calculation of the different combinations of two-digit-matches, which I don't feel like doing.)
But anyway, if you stick with one of the two criteria above, you can then tally the number of times you get those 1/3 point matches, and plug that tally in each time to a binomial calculator, and calculate the long-term probability that you'd get results that accurate if it were based on nothing but chance. You can find a free online binomial calculator here: Statistics
(For those interesting in how the binomial calculator works, you can read the tutorial on it here: Binomial Distribution)
Should be interesting to see either way, but the above-described tracking makes it even better, as we can then see how much better-than-chance your results are, as you perform the experiment. (you'll probably get at least a slightly-better-than-chance running result, since people have a shared inclination toward some digits, such as 7)
|
|
Bookmarks