statistical significance

[stancaris]

To all: If a certain handicapping factor gets 63% winners (12 winners in the last 19 years) and horses qualifying on this factor comprise 30% of the starters in the last 19 years what kind of statistical tests can be performed to tell me if the above stat is significant? By the way the impact value for the above stat is 2.10 (63 divided by 30 equals 2.10). Horses who qualify on this stat are winning twice as often as is expected. Statistical expectation would be .30 times 19 years or 5.7 winners in 19 years.

[dublino]

The most important statistic you need to now about handicapping 96% of people who gamble lose.

I would suggest an account at Betfair or Betdaq where you can lay horses to lose, ie have the field running for you agaist one horse.

There is no handicapping system out there or one that can be made up to make money by backing horses to win.

I know its not what you asked but what your system is throwing up is just a temporary coincidence not a system or handicapping factor.

[bdw0617]

To all: If a certain handicapping factor gets 63% winners (12 winners in the last 19 years) and horses qualifying on this factor comprise 30% of the starters in the last 19 years what kind of statistical tests can be performed to tell me if the above stat is significant? By the way the impact value for the above stat is 2.10 (63 divided by 30 equals 2.10). Horses who qualify on this stat are winning twice as often as is expected. Statistical expectation would be .30 times 19 years or 5.7 winners in 19 years.

i think you are going about it the wrong way.

in other words, you are trying to basically commercialize your staitustic so that you can hone in on it on a more normal basis to make more wagers/more money.

if you go this route, the stiatuatic will almost inevitably water itself down to the point where it's not as fiesable to go about pursing it.

what you shoudl do, and this is the approach I take with gambling, is instead of trying to find more races to wager on, save up your money and just unload on the few times that this comes around. i would rather bet 1000 on 1 race, then 100 on 10 races if i am that confident in that one race. i'll lose, shit i'll lose more than i win, but if i have a deep enough bankroll i will almost always turn a profit sooner or later, and a pretty significant one


for all of the advancements and new site stuff that they roll out, it's all regurgitated info, in other words, there should be a tool out there that within minutes lets you find out how fiesable your plan is. this is quite easy to do with stocks regardless of whatever stat you found. this is almost impossible with horse racing. I'm a lot like you; I'm more interested din stats than speed figures when it comes to horse racing. what if i want to find out the win percentage of say.. bob baffert when he teams with martin garcia on 1 turn races versus 2 turn races while at santa anita park. unless you keep your own extremely detailed stats, it's impossible. and this is pretty basic information. My handicapping methods over the years have developed to the point where they are pretty unique... i really don't give a shit how many lengths a horse finished in front of another horse that he's racing against today; to me that is almost irrelevant information. i don't care to spend hours dictating past races it's pointless; if the trainer isn't trying to wint he race it doesn't matter how many lengths the horse lost on the turn last time out

so i find myself wanting stats and i can't get them. i' going to have to do something about that.


anyway that's my advise. save the omney you would wager and unload on the times the findings do come around. it's a guaranteed profit

[stancaris]

To all: Same question as above--How can I tell if the stat is significant.

Neither of the above posters answered the question I stated.

[Fireslam]

Put your money where your mouth (stats) are. That will let you know quickly how significant it is.

[bdw0617]

To all: Same question as above--How can I tell if the stat is significant.

Neither of the above posters answered the question I stated.

if 2 people answer the question and neither is the answer you are lookig for, the problem isn't with the people answering the question

[bdw0617]

to clarify i am saying that you need to re structure your question so that we can understand what you are asking. it's not very clear.

[griff]

I once enrolled in a masters program At the Univ. of Redlands in CA and dropped out after an advanced course in Statistics. Some of that stuff stayed with me and I remember the question of significancy was considered very important.

For example, if I buy 10 lottery tickets have I significantly improved my chances of winning and the answer is no.. Th point being there is not a significant difference whether you divide 1 by 10 to the 6th ot 7th power or if you divide 10 buy that same number.. The same goes for 100 lottery tickets, and so forth

However, by applying very thing I learned in that advanced class in statistical Analysis I was able to find that you could obtain a very significant increase in your chances of winning if you by one ticket rather than zero tickets;ni.e., try dividing 1 by zero.

probably not in line with this trend but what is?

griff

[Dave C]

I don't feel like digging out my stats texts books right now so I'll just suggest you search wikipedia. Search for statistical significance. They should at least give the equations for t-tests, and probably for f-tests and z-tests. I would think that the bigger problem will be in establishing the validity of your samples.

[xfactor fan]

stancaris,

Looks like you are trying to test the statistical validity of your observation. Which given past posts is probably related to a particular stallion line in a pedigree.

So in your database, with X stallion, you have a pattern.

The next step is to do a random selection not based on bloodlines and look at the results.

For example if your data base is arranged by race, and post number, not order of finish, take every third horse listed, and run the same sort of analysis on that pool.

Any method that give a random selection would work. However you need to be careful to rule out accidental bloodline clusters. Color, and names can both suggest of sirelines. ie: Storm, Unbridled

IF your random population shows the same pattern, or close to it, then your results don't have significance. If the are much less, then you may be on to something.


Good luck.

Another

[zinn21]

Let me see if I understand this and not sure my math is correct. Out of 100% of the starters, 30% qualify under this handicapping factor. From those 30% who qualify, 63% win and the number of winners is 12 over a 19 year period. So 12 won from a total of 19 (12/19ths =63%) who qualified. And if 19 comprised 30% of total starters, the total number of starters (19/63rds=30%) was 63?

Roughly 30% of all horses win as the favorite. Obviously your handicapping factor is very narrow to only come up with 63 horses who qualify under this factor in 19 years. I've always thought the only way to beat the races is to somehow discover some very filtered handicapping concept that won at a high percentage with a high return on investment. The discipline to wager on 63 horses only over a 19 year period is beyond my patience level.

[stancaris]

X Factor Fan: Lets say for argument sake that my handicapping method concerns a Beyer figure of 98 or higher. Horses with a 98 or higher Beyer fig in their last start are winning a certain stake race as follows: 12 of the last 20 winners had a 98 Beyer figure or higher in their last start. These types comprise 60% of the winners from only 20% of the starters. The impact value is 3.00. This means that runners fitting this handicapping angle are winning the hypothetical stake race 3 times more often than statistical expectation.

In doing the research I looked at every runner in this particular stake race over the last 20 years and put every runner into one of two groups: group 1 were those that had the 98 Beyer in their last start and group 2 were those that lacked the 98 Beyer in their last start. I do not believe that random sampling applies to the above situation. Wouldn't it be much more accurate to include all the runners in the sample instead of every third horse?

Same question as before: How does one find out if the above data is statistically significant?

[stancaris]

Zinn: your interpretation of the number of starters is incorrect. In the study that I did which is found in my first post on this thread there were 345 starters over a 19 year period and in that time frame there were 103 starters that qualified on this handicapping angle. This translates into 30% of the starters qualifying on this angle. There were 103 of 345 total starters that qualified NOT like you said that there were only 63 starters.

[xfactor fan]

In order to figure out if you have identified a pattern, you need to look at two groups. Your selected group, and a random group. Putting together a true random sample can be done in a couple of ways, like every third horse in the database, or horses with one word names, two word names, three word names.

This second group provides your control. What kind of numbers do you get from a random group? If your Beyer figure numbers are better than the random group, you have discovered an "angle".

The important this is the comparison, Lets say the random group produces 6% winners. and your Beyer group produces 26% winners. Then you can say that your angle is better than chance.


Just looking at your Beyer data alone and out of context is like those TV commercials that say "20% more". Ok, 20% more than what? You need to have a baseline to compare the data to.

Does this help?

[zinn21]

Stancaris then 12 winners out 103 qualified starters. That's around 9% win rate from qualified starters. You mention 63% win rate in your initial post so something does not equate.. In order to break even you would need an average $17 return for each $2 dollars wagered from those 12 out of 103 starters who won.