Rating History

The whole idea of rating lottery numbers came about around November 1995. I originally started collecting the drawn numbers, machines and set of balls facts, believing that the each machine or set of balls may have favourite numbers.
From the tedious task of browsing these stats manually, I wrote a computer program on my home Amstrad computer. This simply picked out and made a list of all the numbers that had been drawn with each machine and how often. Although this lead to some bland results it still produced three £10 wins within my first four plays!

With this in mind, I then set about modifying this program to output all the numbers that had appeared alongside each other number. This would help shorten the list of combinations chosen from the first program, down to just two or three good lines that I could play with.
Using the stats produced by this program, I developed an algorithm that would search all possible combinations, rating only those necessary, which would output the top 20 combinations of various lengths.

From here my Amstrad computer, with its astounding 64K(!) and 3.7MHz, was going to be too slow and a transfer from BASIC to Ada was made from home to the university computers.
The rating procedure used at this point ( Nov. 1996 ) was ad 'hoc, to say the least, and my winning streak was losing touch with what I had in mind.

So, on the 19th June 1997, I mathematically developed the formula shown on the derivation page to rate lottery combinations.

What does the combination rating do?

In plain English, this function identifies pairs of balls that have appeared together quite often in the past. Using the best number pairings it forms combinations of various lengths and rates them accordingly.

I am not trying to break any rules of statistics here:-
the value of the rating for each combination is not a probability of winning.
The rating just gives an indication of how much better one combination is from another.
A smaller rating denotes a better combination.

The ratings can be interpreted in one of two ways:-

• Good combinations are 'hot' and therefore should be used as staked lines.
  This approach is based upon the idea that the number selection is deterministic, i.e. the lottery machines in some way influence the balls that appear. This implies the results of past draws can be used to determine future ones.

• Good combinations have 'had their day' and should be avoided like the plague.
  This is the more conventional approach. Based on the assumption that the lottery machines are totally random, we can infer that combinations which are rated highly have appeared more often than others. So if an average, random nature is to occur, we might expect combinations with lower ratings to 'catch up' by appearing sooner.

How you use the data churned out by this program is purely personal preference and belief, although one thing is for certain:- we should not use either notion and the rating function to find 'good' lottery combinations, rather the function and past draw data to test each premise.
Until either premise is shown to be valid, we can not be entirely certain of how to use these results.

However, in the mean time, this is just as good as any other way of choosing your lottery balls.

On the 11th January 1998, I devised a way of altering the rating value of each combination to include a prediction factor for each ball.
The technical details of which can be found here.

Two very important points

The prediction scheme can only be used in joint with the second philosophy mentioned above. This is because the scheme is based on the assumption that the numbers drawn are totally random, which the first approach rejects.

The second and most important point is how, or rather when, the scheme is known to work.
Randomness suggests that all balls will eventually be drawn the same number of times. Of course we can only be assured of this after an infinite number of draws - a mathematical possibility, not a real option. Predicting that a ball is due to appear because of its lack of recent appearances is a dodgey game, and I can't say I'm all that convinced myself. In the short term ('cos I'm too impatient to wait for infinity) the scheme also works provided there is little variance in the random nature of the number selection. This notion actually comes back upon itself, since to calculate the actual variance you need to analyse the draw data for all draws up to infinity. All-in-all, this bit is a leap of faith, we'll have to wait and see whether the variance works out to be low enough.