Betting Systems?

[stlseeeek]

Is there a universal way you guys figure out points, speed variations, and ratings?


example: say you give a horse 5 points for best jockey, add in avg of times at the distance, etc



I'd like to atleast try and learn of some systems, to see if it could help my handicapping, and get a better ROI.



thanx

[Overlay]

If what you're referring to or looking for is a measure that can be used to judge widely varying factors on a common scale in order to facilitate blending or otherwise dealing with them quantitatively, impact values associated with the rankings of horses compared to their competition in today's race for each of the various major handicapping categories (such as speed, pace, class, etc.) come as close as anything I've found.

[kenwoodallpromos]

No. And if there was, everybody would use it and make it hard on the ROI.
Statistics have shown me how much to value various what I call "variables" (angles, factors). For example, early speed EP type horses win much more in sprints than in routes. Certain types of races and horses at certain ods dictate how many points to give a horse.
I use early speed on fastest tracks; claimers and allowances on place and show; and the unpredictable races (cheap long, etc) for multiple longshots.
Horses witn certain attributes in their past performances tend to do much better in the "average" race and you can find stats on those; but be careful when handicapping oddball races (turf switched to dirt, races no entrants are experieced in), "muddy" track". Many of those do not conform to the norm.
What I tell people is to monitor your picks and bets and check results charts on Equibase.com to look for success and common factors. They are there most of the time!
Part of what makes horseracing so interesting is that tracks, rules, race conditions, and speed figures all use a wide array of criteria.

[RaceIsClosed]

Is there a universal way you guys figure out points, speed variations, and ratings?

example: say you give a horse 5 points for best jockey, add in avg of times at the distance, etc

I'd like to atleast try and learn of some systems, to see if it could help my handicapping, and get a better ROI.

thanx


Everything is worth x amount of time, in the final analysis. Your horse is going to run today's race with a final time of x. All handicapping factors are used to adjust your guess of what x will be. The horse with the lowest value for x is the one most likely to win the race.

The problem is, anyone who posts something truly profitable is going to be trashed as a know-it-all, or attacked as unproven. That's the "reward" for helping the fellow man win.

Do you *really* need to know how I found the 9-1 shot at Tampa that keyed a $1,600 superfecta on Tuesday? Even if you do, what do I get for telling you except flames from the haters and a devaluing of the methods I used to pick it?

[xfile]

What? Are we back in the 80's??? :cool:

[twindouble]

What? Are we back in the 80's??? :cool:

I knew some would eventually come around to my way of thinking. :lol:


T.D.

[PaceAdvantage]

Do you *really* need to know how I found the 9-1 shot at Tampa that keyed a $1,600 superfecta on Tuesday? Even if you do, what do I get for telling you except flames from the haters and a devaluing of the methods I used to pick it?

You mean like all that tremendous flaiming Boxcar is getting in his "Longshots..." thread????? :confused:

Come on man....you're exaggerating. When people get flamed around here, nine times out of ten, they had it coming to them....

[boxcar]

You mean like all that tremendous flaiming Boxcar is getting in his "Longshots..." thread????? :confused:

Come on man....you're exaggerating. When people get flamed around here, nine times out of ten, they had it coming to them....

Stay tuned....one unhappy camper on my thread swears he's been insulted by me. Things could get interestin'....

Boxcar

[andicap]

Stay tuned....one unhappy camper on my thread swears he's been insulted by me. Things could get interestin'....

Boxcar

Again, stop bringing politics into the racing forum
:D

[traynor]

kenwoodallpromos wrote: <Statistics have shown me how much to value various what I call "variables" (angles, factors). For example, early speed EP type horses win much more in sprints than in routes. Certain types of races and horses at certain ods dictate how many points to give a horse.>

A question. When you calculate the values, do you go by win percentages (how often the value in question wins), or do you relate it in some way to mutuel returns? That is not meant to sound foolish. It is fairly easy to find factors that predict the winner frequently, or that predict high mutuels; it is more difficult to find an acceptable percentage of winners at high enough mutuels to be worthwhile.

I have not found it particularly useful (a euphemism for "making a profit") to wager on win frequencies of less than about 20%. The lower win percentages tend to correlate with occasional boxcar mutuels, interspersed with incredibly long losing stretches. That is, records indicate a particular ROI for a particular factor in a given sample that is rarely replicated in subsequent samples. Or, in plainer language, betting on a pattern with an aberrant longshot or two tweaking the (apparent) results seems like a losing proposition.

I tend to cringe when someone describes betting on a pattern (angle/factor) with a historical success rate of 5-10% and a positive ROI. I have attempted such with a remarkable lack of success, and I would appreciate any insights or opinions you might have on this topic.
Good Luck

[Overlay]

A question. When you calculate the values, do you go by win percentages (how often the value in question wins), or do you relate it in some way to mutuel returns? That is not meant to sound foolish. It is fairly easy to find factors that predict the winner frequently, or that predict high mutuels; it is more difficult to find an acceptable percentage of winners at high enough mutuels to be worthwhile.

If you know how often a factor wins (in terms of percentage) or know its strength in relation to field size (as with the use of impact values), and particularly if you can determine that the factor's correlation with winning is too strong to be accounted for by random chance (that is, the factor by itself is responsible for the better-than-expected performance of the horses possessing it), you have a firm basis for using the odds of the horses in question to determine if wagering value is present (if the odds of the horse are higher than the winning percentage of the factor), and then betting when it is, and passing when it isn't. This same process can be used with individual factors, or with suitably distributed (non-redundant) combinations of factors from the major handicapping categories (speed, pace, class, condition, and so forth)(through multiplication). The use of combinations of factors increases the likelihood of finding value, since you're not keying in exclusively on any one variable, which can result in overbetting when practiced by too many members of the public-at-large.

[sjk]

Traynor,

I bet horses in the win slot if the price is right whenever I give him at least a 5% chance of winning and in the second slot (of the exacta) if I give him at least a 3% chance of runnning second. I don't think there is any intrinsic reason why a line cannot be accurate in every odds range.

Relating to your concern in the other thread about independence I bet almost all exactas and I act as though the horse running 2nd is independent of the horse running 1st. You would think (perhaps if both horses have early speed) that this could lead to problems but for me over tens of thousands of races it has never seemed to create an issue.

[JohnGalt1]

It is nearly impossible to create a universal system that fits every race.

Turf, dirt routes and sprints, maiden races are all different and different factors are more important in each race.

If you used a point system for jockey, trainer, speed, pace, class, etc, maybe a turf route would give more value to late pace, class and breeding, but a maiden race with all first time starters would only give value to breeding, (for debut and distance), trainers with first timers, workouts, but could not be ranked on speed/pace or class.

One flaw in a number ranking system is the Bris Prime Power Number. Foreign horses and first time starters get no number. Another flaw is a ranking for condition. A few years ago they ranked a horse #1 that was coming off a 1 year lay off who had run in $20,000 claiming races at SA was now in $3,000 Turf Paradise race with a single 3 f work out. If healthy and in condition he should've destroyed his competitors with his back class and faster races, but tired and finished up the track.

Look at each race as a seperate puzzle and remember that all puzzles can't be solved, and those are the races to pass.

[traynor]

overlay wrote: <If you know how often a factor wins (in terms of percentage) or know its strength in relation to field size (as with the use of impact values), and particularly if you can determine that the factor's correlation with winning is too strong to be accounted for by random chance (that is, the factor by itself is responsible for the better-than-expected performance of the horses possessing it), you have a firm basis for using the odds of the horses in question to determine if wagering value is present (if the odds of the horse are higher than the winning percentage of the factor), and then betting when it is, and passing when it isn't. This same process can be used with individual factors, or with suitably distributed (non-redundant) combinations of factors from the major handicapping categories (speed, pace, class, condition, and so forth)(through multiplication). The use of combinations of factors increases the likelihood of finding value, since you're not keying in exclusively on any one variable, which can result in overbetting when practiced by too many members of the public-at-large.>

I agree for the most part, with a minor exception; when calculating impact values, it is essential to factor out overlaps--races in which two, three, or four entries shared a specific factor. Simplistically, speed and class--the question is, "how often does top speed win when it is also top class, second in class, within top three in class, etc." It generates a completely different view, because you are essentially generating two or more related but discrete factors. The value of one factor is only valid when the secondary constraints or attributes of another--perhaps seemingly unrelated--factor are applied or removed.

Example is win percentage--it tends to be high in entries that either run fast or that run at lower grade races. Without evaluating or quantifying the grades or conditions under which the value of "win percentage" is earned, the use of that factor to predict the outcome of subsequent races is diminished.
Good Luck

[traynor]

sjk wrote: <I bet horses in the win slot if the price is right whenever I give him at least a 5% chance of winning and in the second slot (of the exacta) if I give him at least a 3% chance of runnning second. I don't think there is any intrinsic reason why a line cannot be accurate in every odds range.>

In your case, with the monster data base you use, it may work. For most, betting an entry because of a rate that amounts to 50 wins in a thousand races is not practical. Consider--if one or two of those 50 per thousand has an unusually high mutuel, "statistics" indicate profit. Even if the anomaly is repeated over a base of 5000 or 10,000 races, it is no less an anomaly; a handful of high mutuels (that are essentially unrepeatable flukes) scattered among the 5% can create a seriously jaded view of the potential of that model applied to other samples.

Again, your example is specific, because you concentrate on exactas. Exactas are not especially profitable when betting chalk, even at higher success rates. I like exactas, although I use a completely different model to determine worthwhile bets (from the model used for win betting). For win bets, anything under 20% success rate coupled with 20% or more net profit seems a bit too risky for someone intent on profit over action. (Action is great, but even better when profitable.)

I can only assume you have developed a method that is workable for you, and I wish you continued success with it.
Good Luck

[kenwoodallpromos]

Some hanicappers and authors believe as I do- % and price matters not as long as the return is more than the % chance of winning.
I have studied "Beat the Track", "150 Blue Ribbon Systems", "Flatstats", as well as info buried in books or on posts or observing results on PP's or results charts. Sometimes I know the exact % expected for an angle, sometimes it is a very educated guess. But I always rely heavily on facts, stats, and common sense to arrive at the %/mutuel ratio.
A couple of examples- Zippy Chippy was 8-1 M/l after going like 0 for 98; he was a big underlay. If a horse is 4-1 and won 4 of the lst 10 it is an overlay. Proud Accolade was to run in the rain and mud on a visiting circuit KY), jockey not on home circuit (Bailey) who has not been winning a great stakes %, a field of 11, and was 1-5. It was a big underlay.
Mules Black Ruby were 100 for 103 exacta record in Fresno and the box exacta and quinella odds were 3-1. They were massive overlays.
Of course other variable (factors, angles) apply, but it is helpful to know how strong to credit each variable. I credit rainy mud to be a strong variable. I credit ZP running at the same track and same level to not be a strong enough variable to alter the basic %/mutuel ratio.
If there is a horse who has won 5 of the last 10 at the same circuit and level there had better be a very strong reason why the tote says 2-5 and not 1-1! If it has been off 6 months or is up in class from a $10,000 claimer and today's is a $50,000 claimer optional, I had better see odds like 5-1 or better if the teller wants my money!
If you want to give a horse a 5% chance of winning then maybe it has won 1 of the last 20 races and that was at this level. If has lost by and average 1 1/2 lengths in the lst 3 races maybe give a 6% chance!
If you think a longshot has a 5% chance of winning, then maybe it and 4 other longshots in the same race have a total 25% chance of winning. In the 2005 KY Derby. the tote said 15 horses had 20-1 or longer odds- total 70% + or - chance of winning. If you had time to add up and average the payoffs expected for all 15 of those it was like $65 by my wild guess; $30 in bets covered them all. (Some of those longshots may have even had longer odds at post time then their futures pool odds!).
You can make your own table- list what you consider to be strongest variable for most races (purse level, distance, surface type, last race OTM, change of trainer, etc.) horizontal and vertically list racing situations (race condition, field size, horse record, last race result, par speed of track, etc.) and insert a value from 1-10. That is the basic situation value. Multiply by 1-10 depending on how well you think that horse will in that situation. That is your % chance of winning. Convert the % to mutuel odds and the result is minimum odds you will accept.
(For example: the horse won its last race at the same purse level= 10; Same distance as last race and a field of 6=6; that is a 40% chance of losing and 60% chance of winning= 2/3 odds). If that horse is even odds you may consider it an overlay.
The point is, you still have to decide based on experience and knowledge how much to credit each variable, then consider that in relation to the expected mutuel payoff.
Maybe a certain win does not pay off enough in the "average" race to be profitable. I know what my ROI is and needs to over many races, but I never bet and "average race". I only bet races where the win frequency is enough at the expected mutuel to produce a positive ROI over the long run.
Why bet an EP type in a route at a situation value (frequency) of 35% the next race is a 5 1/2 dirt at GP if the expected frequency of EP type favorite winning is 55% and everything else is close to equal? Just to get a bet down? THAT is gambling!

[sjk]

sjk wrote: <I bet horses in the win slot if the price is right whenever I give him at least a 5% chance of winning and in the second slot (of the exacta) if I give him at least a 3% chance of runnning second. I don't think there is any intrinsic reason why a line cannot be accurate in every odds range.>

In your case, with the monster data base you use, it may work. For most, betting an entry because of a rate that amounts to 50 wins in a thousand races is not practical. Consider--if one or two of those 50 per thousand has an unusually high mutuel, "statistics" indicate profit. Even if the anomaly is repeated over a base of 5000 or 10,000 races, it is no less an anomaly; a handful of high mutuels (that are essentially unrepeatable flukes) scattered among the 5% can create a seriously jaded view of the potential of that model applied to other samples.

Again, your example is specific, because you concentrate on exactas. Exactas are not especially profitable when betting chalk, even at higher success rates. I like exactas, although I use a completely different model to determine worthwhile bets (from the model used for win betting). For win bets, anything under 20% success rate coupled with 20% or more net profit seems a bit too risky for someone intent on profit over action. (Action is great, but even better when profitable.)

I can only assume you have developed a method that is workable for you, and I wish you continued success with it.
Good Luck


If you made an odds line on all the horses and your testing told you that your 12-1 horses were in the aggregate good bets but the 11-1 horses and 13-1 horses were not it would be absurd to assume that you have a good method but that it is specific to that level of odds.

I would expect a sound method of generating odds and betting sufficiently overlaid horses to work at every odds level. If it does I don't see how anomolies in the data could be responsible.

I do well betting chalk exactas and am suprised that you would write those off. I enjoy betting chalk over chalk for $12 if I think the value is there as well as chalk over 60-1 when the price is right.

I try to play value wherever I find it. If I think a 25-1 over a 40-1 provides value I will play it (and unfortunately it came in upside down at OP today).

While I'm at it I would also take issue with jgalt1. I use exactly the same program to play all dirt races and it seems to work out well. I would agree that turf is totally different and I have never tried to solve that one.

[traynor]

sjk wrote: <If you made an odds line on all the horses and your testing told you that your 12-1 horses were in the aggregate good bets but the 11-1 horses and 13-1 horses were not it would be absurd to assume that you have a good method but that it is specific to that level of odds.

I would expect a sound method of generating odds and betting sufficiently overlaid horses to work at every odds level. If it does I don't see how anomolies in the data could be responsible.>

I don't think it is reasonable to use that finely grained a distinction. Perhaps something more on the order of 5 to 1 or less, 5+ to 10 to 1, and > 10 to 1, with a possible chalk on the low end. While there is a loose correlation between odds and probability of winning, it is loose--a long way from a direct correlation, which is at best still only a correlation.

With an uncontrolled mass of raw data, extracting percentages like "5% chance of winning" seems highly speculative. If you do it successfully, more power to you. I do this for money, and a LOT of research indicates that using large masses of uncontrolled data lumped into a mega-model is misleading, rather than predictive. Unstated is how you arrive at the 5% figure--you may have a workable method that is beyond a simple regression indicating that one or particular factors, or combination of factors, have won 5% of similar races in your database.

Chalk exactas are not particularly interesting to me, because I use deltas--essentially a dutched pattern with specific amounts correlated to the exacta pool. It is intriguing on a good day, and a lot of hard work and stress on others, particularly at tracks where others are using the same approach and dump very large bets just before the last flash. I am not a big fan of adjusting bets at the last minute because someone else just laced the same delta pattern and cut the odds on each combination in half. In that case, I find it more profitable to use a truncated delta, only betting the "peak" end and discarding the chalk "base." Again, just a personal preference.

I think there is more money to be made on straight win betting than in exactas, with the exception mentioned above. I keep fairly detailed data models, and it is not just personal opinion or preference.

I am curious how you arrive at your expected odds. I use a strict probability model that has no correlation to current odds. In short, I don't care if the entry goes off at 5 to 1 or odds on--the pattern is based on a model, and I bet the model. I know this is counter to a lot of "horsey wisdom" by self-proclaimed experts, but it has the advantage of working--over extended periods of time--and generating a comfortable margin of profit.

I have seen MANY races in which "my" selection was at unacceptably low odds a minute to post, then went up to 2 to 1 or 3 to 1 after the race was off because someone's brother-in-law from Poughkeepsie bet his roll on another selection. Waiting until the last minute to decide whether or not to bet, based on someone else's actions, is a crazymaker. If you have a model, bet the model. If you are betting the tote board, you need to find a way to factor that action into the model. In the scenario above, if you decide only to bet based on odds, unless you factor out the races you would not have bet because the odds a minute to post were too low, but the final odds were acceptable (and so included in the "model"), the model does not accurately reflect reality.
Good Luck
Good Luck

[traynor]

kenwoodallpromos wrote: <The point is, you still have to decide based on experience and knowledge how much to credit each variable, then consider that in relation to the expected mutuel payoff.
Maybe a certain win does not pay off enough in the "average" race to be profitable. I know what my ROI is and needs to over many races, but I never bet and "average race". I only bet races where the win frequency is enough at the expected mutuel to produce a positive ROI over the long run.>

The "experience and knowledge" part is where it gets cloudy. I used to use multiple regression analysis to extract impact values. That gets very complex, very fast. Again, a simplistic example is speed with, or without, class. While each factor can have a distinct impact value, that can be "empirically derived" (otherwise called, "discovered by looking at"), the particular blending of those impact values may be unique for a given race. MRA involves a whole, crazy chain of "what ifs" and "yeah, buts" that, if overlooked or ignored, allows large numbers of otherwise astute researchers to generate essentially meaningless or misleading results.

I understand that expert systems can use layered iterations to craft a "finely tuned" set of impact values for various combinations of multiple factors; that is what I do on a daily basis. I also understand that "statistics," even when carefully crafted, can be highly misleading if the data model is off in some (apparently trivial) way. That may be why I am so skeptical of "database" handicapping and tweaking impact values to match history.

It would seem you are saying that you correlate win probability with current mutuels, then make a bet-no bet decision on the fly, based on the available odds?
Good Luck

[sjk]

I developed a program several years ago to calculate the odds line for the race. It took a considerable amount of work to develop and try to optimize the process. I have tried to incorporate all of the elements that I would consider relevant to handicapping a race and there are hundreds of Access queries used in the calculations.

I don't use the tote odds as an ingredient in the calculation of the odds line but make only bets where the prices are sufficiently overlaid so the tote odds are key to the decision of what to bet. As I said above I am willing to bet almost any horse in the race if the price is right.

Back-tests are useful but the real test is how things turn out in actual real money play. I don't see much difference between backtest results and real money ones that would relate to last minute tote changes.

[kenwoodallpromos]

It would seem you are saying that you correlate win probability with current mutuels, then make a bet-no bet decision on the fly, based on the available odds?
Yes. But the chart is for learning purposes. This is the kind of thinking I use for most of my betting; my place/show method is much different and does not depend on odds because the win % is always high enough.

[kenwoodallpromos]

http://www.horseracingsystems.net/index.php.
Lots of ideas!

[PaceAdvantage]

Back-tests are useful but the real test is how things turn out in actual real money play. I don't see much difference between backtest results and real money ones that would relate to last minute tote changes.

How close to post time are you placing your bets? I find what you are doing fascinating, but I myself have been rather unsuccessful when trying to incorporate betting overlays based on my odds line (and yes, I too wouldn't be adverse to betting 3, 4 or even 5 horses in a race if the odds say to do so...)

[highnote]

http://www.horseracingsystems.net/index.php.
Lots of ideas!


Check out Nick Mordin's book "Winning Without Thinking". The title is tongue-in-cheek. What he says in the book that when developing a system you are doing your thinking up front so that when it comes time to bet it's just a matter of checking the odds to see if your horse warrants a bet.

His book has lots of good ideas on developing systems.

[kenwoodallpromos]

Thank you! I will check on that. Since I have no problem deciding to cancel a bet if need be, 2 thngs would make his book interesting! (For me cancelling a bet if scratch or my pick is overbet late is fine- I consider it $2 profit!)

[highnote]

Nick's book, "Betting For A Living" is a classic. He's funny, witty and informative.

Two of his books are out of print, but Betting for a Living and Winning Without Thinking are still in print.

Check out http://www.nickmordin.com

He's got some exciting stuff coming soon -- Global Speed Ratings. He realized years ago that racing was becoming a global sport. He packed up his tack and moved from England to Upstate NY to write about US and International racing for a UK publication.

I don't get any money from his products, but he did pay me a little something to write software for him that helps him make International Speed Ratings.

If you want to bet on The Cheltanham Gold Cup, Epsom Derby, Grand National, Breeders' Cup Turf, etc., he's the man to talk to.

He understands foreign form better than anyone I've met. Whenever I make a pick 6 bet in SoCal I always call Nick to ask about the foreign horses. The amazing thing, he can tell me about their past performances without even looking them up.

He's not very well known in the States, but he is in England. "Betting For A Living" is the best selling handicapping book of all time in England.

[sjk]

PA,

I usually start thinking about betting at 3 min to post and bet from there in. If there are 4 races going off in 4 minutes at the different tracks I am playing I can't wait too long to get moving or it won't all get done. On the other hand if there is only one track at that moment and the bet looks like it might change (borderline play, small handle track) I will wait until inside a minute.

I use little keyboard macros to read the on line tote-tboard and prepare the data to import to Access. An Access macro compares the prices with the requirements and prepares a list of bets. As long as I have all of the right windows open it takes no more than 15 seconds from the time I sit down at the computer to have the list of bets which I then punch in manually.

It is common that I get home or return to the computer from doing something else and see that it is 1 mtp somewhere. Plenty of time to bet the race.

[Overlay]

I agree for the most part, with a minor exception; when calculating impact values, it is essential to factor out overlaps--races in which two, three, or four entries shared a specific factor. Simplistically, speed and class--the question is, "how often does top speed win when it is also top class, second in class, within top three in class, etc." It generates a completely different view, because you are essentially generating two or more related but discrete factors. The value of one factor is only valid when the secondary constraints or attributes of another--perhaps seemingly unrelated--factor are applied or removed.

Example is win percentage--it tends to be high in entries that either run fast or that run at lower grade races. Without evaluating or quantifying the grades or conditions under which the value of "win percentage" is earned, the use of that factor to predict the outcome of subsequent races is diminished.
Good Luck

If I understand you correctly, that was what I was trying to convey by referring to the factors I would use as being "suitably distributed" (i.e., diverse enough that their respective influences would not be duplicative, and strong enough in their correlation to winning probability that they would function independently of other variables), and by using factors where only one horse in each field would possess a given rank, with the intent of weeding out the kind of overlapping that you refer to.

I agree with your comment about needing to consider the race conditions under which a factor's performance was observed, so that findings are not applied too broadly. I try to address this by breaking my sample down into a sufficient number of subgroups so that I can analyze the specific effects of individual factors in each group.

Thanks for your comments.

[PaceAdvantage]

I use little keyboard macros to read the on line tote-tboard and prepare the data to import to Access. An Access macro compares the prices with the requirements and prepares a list of bets. As long as I have all of the right windows open it takes no more than 15 seconds from the time I sit down at the computer to have the list of bets which I then punch in manually.

It is common that I get home or return to the computer from doing something else and see that it is 1 mtp somewhere. Plenty of time to bet the race.

Man, you and I need to hook up. I could make your life a whole lot easier....lol

[xtb]

If I understand you correctly, that was what I was trying to convey by referring to the factors I would use as being "suitably distributed" (i.e., diverse enough that their respective influences would not be duplicative, and strong enough in their correlation to winning probability that they would function independently of other variables), and by using factors where only one horse in each field would possess a given rank, with the intent of weeding out the kind of overlapping that you refer to.

Wouldn't impact values handle this? Or better yet, Gordon Pine's modified impact values? Most of my variables are rankings but not all such as ascending speed figures.

I agree with your comment about needing to consider the race conditions under which a factor's performance was observed, so that findings are not applied too broadly. I try to address this by breaking my sample down into a sufficient number of subgroups so that I can analyze the specific effects of individual factors in each group.


I do something similar using track, surface, distance, msw (y or n), sexofrace, sloppyormuddy, ageofrace, class, in that order. I shoot for all these categories but take what I can get while keeping my sample size at a certain level. My problem is the tradeoff between what is a sufficient sample size and keeping my subcategories. I want to add purse and class conditions but even with almost 30k races, I don't have enough data yet. And, I really don't know what a sufficient sample size should be. I'm finding the order of my subcategories could use some adjusting for different variables and it's becoming quite a challenge.

[sjk]

Man, you and I need to hook up. I could make your life a whole lot easier....lol

Lets talk about it someday.

[Overlay]

Wouldn't impact values handle this? Or better yet, Gordon Pine's modified impact values? Most of my variables are rankings but not all such as ascending speed figures.

Yes, that's why I use impact values (if I didn't mention that before). I just meant that I try to avoid applying them to separate factors that are similar enough that using the impact values for each of them would be redundant, and would affect the accuracy of the final result.

I do something similar using track, surface, distance, msw (y or n), sexofrace, sloppyormuddy, ageofrace, class, in that order. I shoot for all these categories but take what I can get while keeping my sample size at a certain level. My problem is the tradeoff between what is a sufficient sample size and keeping my subcategories. I want to add purse and class conditions but even with almost 30k races, I don't have enough data yet. And, I really don't know what a sufficient sample size should be. I'm finding the order of my subcategories could use some adjusting for different variables and it's becoming quite a challenge.

I've had that same difficulty as far as where to draw the line in a manner that would give me a feasibly small number of handicapping categories to work with, sufficient specificity of results, and enough data in each category from which to draw valid conclusions. I finally found that I could live with somewhat greater generality on my race categories (keying on distance and running surface) as long as my performance measures covering class, form, speed, and pace showed similar effectiveness over all those categories.

[xtb]

Ah, got it. I didn't read closely enough. I've started combining variables too, with good results by excluding negative factors. Best of luck.

[highnote]

I use little keyboard macros to read the on line tote-tboard and prepare the data to import to Access.

I don't understand what you mean by using keyboard macros to read the online tote-board. Can you explain that? Sounds interesting.

[sjk]

I save the odds board to a known location on my drive where I can open it up in Excel. Then I name the ranges where I know the relevant data will be and save as xls. Access imports the spreadsheet and finds the data in the named ranges.

Easy enough to do by hand but quicker if you use keyboard macros. I bought a program called EZMacros about 10 years ago for $29. I don't know if they are still around but I'm sure many programs do the same thing.

[highnote]

So the macro software lets you record a series of keystrokes and assign the macro to a single keystroke?

Then you have the toteboard open and hit a button on your keyboard that activates the previously recorded macro and that macro emulates the series of key strokes and mouseclicks that are used to save a file to your hard drive?

[sjk]

Correct. Actually I have to hit four keys to make sure the operations don't step on each other but if you were good at it you could probably do it with one key.

Should be easy to find a program if EZMacros is no longer out there

[traynor]

xtb wrote: <Wouldn't impact values handle this?>

By definition, an impact value is a comparison of two very simple numbers; the expected frequency and the actual frequency. The sticky part comes in when the impact values are viewed in isolation. Again, simplistic categories of speed and class offer the best example. The impact value for speed is only "meaningful" when the overlapping area of class is factored out. In the overlapped areas (graphically, a Venn diagram in Excel), it is unclear which of the factors actually had the impact. Rather than two distinct IVs, there are three; speed, class, and the overlapped area of speed and class.

That simple strategy--assuming distinction in the data that does not exist in the real world--has resulted in a lot of otherwise reasonable research projects being hopelessly confounded by a mix of data types.
Good Luck