Measuring Turquoise Underwear

To recap:

1) I hypothesised a potential mathematical mechanism to explain a small amount of predictability in lottery draws based on past data. I implemented a spread*** to generate predictions, ranking each of the 49 balls of the UK lottery. I checked the rankings of the drawn balls, and no predictive effect was apparent. However, it seemed to my visual inspection that the rankings of the drawn balls tended to be widely dispersed.

2) For each of the approx 14 million possible 6/49 combinations, I calculated the population standard deviation. Then I calculated the average of the 14 million standard deviations, obtaining a result of just over 12.

3) I calculated the population standard deviations of the ball rankings of successive lottery draws, and using a formula hoicked from a stats book, evaluated the probability that these came from a population with the mean calculated in 2), using a 2-tailed t-test. Currently p = 0.000808.

I'm a long way from certain that the above tests are statistically valid - do any math gurus have any comments? The population standard deviations of the 14 million combos are not a normal distribution - the minimum is probably found in 44 cases; 1-6, 2-7 ... 44-49. I guess the maximum is probably only found in one case: 1,2,3,47,48,49.

I once found a web site that allowed you to type in a p value and returned you an equivalent number of standard deviations, and vice versa. Unfortunately I can't remember the URL and a quick google didn't disclose it. If anyone knows of a website offering this facility, I'd be grateful. Most sites just seem to offer a table, usually giving up at 3 SDs. As long-term subscribers and correspondees know, I've had a number of healthy-looking systems keel over and die after reaching 3.5 SDs or so, so I'm not going to blow any trumpets until at least 4 SDs.

Evil Nigel

.

Loading