Thursday, June 28, 2012

Research study, part V: Some final notes about behinders

One pattern with the "behind" winners was so pronounced that I detected it even though I wasn't formally charting it. I decided to go back and capture the information after the fact, which is that large payoffs are a staple of the "behind" firster winner.

In part II I spoke of my control group of 238 non-state-bred dirt sprint maiden special weights for 2-year-olds or 3-year-olds only, covering mid 2007 to mid 2008 (there were some other minor requirements; see part II). The median winner in that dataset went off at odds of 3.20-1; by comparison, the 50% mark of the "behind" winners was 8.05-1. 13.9% of the control-group winners went off at 10-1 or more; 40.7% of the "behind" winners did. For 20-1+, the percentages were 6.3% and 16.3%; for 40-1+, 1.3% (3 for 238) and 4.7% (4 for 86). The biggest longshot winner among the behinders was 1998 Gotham winner Wasatch, who took his Santa Anita debut after running 12th early at 108-1.

Well-bet, first-out behind-winners also occur with some frequency; 8.1% were even-money or less (although this includes one winner who was coupled), and 30.2% were less than 3-1. The corresponding percentages of winners in the control group, though, were 12.2% and 45.8%.

Since the control group consisted of all maiden special weights, and my study only of ones won by first-time starters, someone might say that I am looking at the difference in mutuel support between winners who have run before and those who have not, but I am positive that is at least 80% untrue. I didn't keep the results, but 15 years ago I studied how first-time starters did vs. how they were bet, and as a group, they run formfully. Well-bet first-out winners easily outnumber overlooked ones. If I had the odds distributions for the first-out winners representing the other styles, I would be able to prove this, but as I said, I did not collect odds initially, so settled for the control group of maiden special winners as an alternative.


As we know by scanning the odds board before many a maiden special weight, bettors make very strong determinations about the chances of various first-time starters, and these determinations are by and large accurate. But they seem to have a bit of a blind spot when it comes to horses who will be coming from off the pace. How this can be is at first curious; after all, the style category of a first-time starter is undeclared and undiscussed.

It seems a good assumption, however, that first-time starters who win showing early speed in fact have more early speed than first-time starters who win from behind. If training rated highly leads to mutuel support, then the type of AM performance suggestive of early speed is more likely to be interpreted in those terms than the type of morning performance suggestive of another type of quality. Indeed, the lukewarm support for off-the-pace debut winners must mean that the indications of it are largely overlooked in training.

This could be a result of first-quarter dullness not leading to impressive workout times, and fans therefore overlooking the the potential “behinder” winner, or it could mean that trainers and horse people themselves are fooled by the different presentation in training given by the behinder. Trainers, for example, might miss that they have a good chance to win with their behinder, and not talk the horse up beforehand.

If this is true, perhaps the nature of workouts simply does not give trainers an opportunity to gauge some qualities that will be very important on race day. Or perhaps they could do a much better job of projecting and looking past early speed. In any event, the horse who can run but can’t run so much early often goes under the radar.

Another theory for the “behinder” longshot mystery is that, for whatever reason, trainers pick certain horses to be ones they try to win first time out with, and certain horses to be ones they will give a race. Trying to win first time out means having the horse in position early to win. (My statistics in Part II show this is a smart move.) Betting on firsters is indicative of trainer intentions as much as ability of horse, this theory says. When the behinder wins, he overcomes a deficit in preparation that is reflected in the odds.


If this theory were true, however, “behinders” should be superior to debut winners employing other styles. But the evidence suggests more uniformity in future performance relative to debut style than behinder dominance.

Additionally, being behind early in the debut would seem to be all about trainer intention and not about the horse’s inclination. Yet reviewing the well-known “behinders” on my list, my impression is they were by and large known as closers for their entire careers. If their trainers recognized that coming from well off the pace would put them at a consistent disadvantage, they wouldn’t have allowed the style to endure, if they had an alternative (meaning that the horse had more natural speed).

One theory that occurred to me about why come-from-behind winners end up doing well, despite their high odds in their maiden scores, and their narrow margins of victory, is that they may have overcome a problem at the start. How does a horse end up at the back of the pack? Sometimes, through tactics, often through lack of speed, but often because he or she didn't break well. This is particularly true of first-time starters, as any racing fan knows. And winning after a bad start might be a sign of a good horse. The lengths lost at the start might compensate for the small margin over the runner-up at the wire.

When I was going back and collecting the odds data on the "behind" winners, I also noted whether the horse was noted in the short comment ending the horse's row in the Equibase chart to have broken poorly or had a bad start. By my coding, twenty-eight of the 86 "behind" winners (32.6%) incurred difficulty at the start. While some eventual winners begin poorly and recover position quickly enough to be mid-pack or better by the first call, I am certain the count of troubled starts would have been lower for the other style categories, particularly for "led" and "pressed."

Nine of the 28 "behind" winners (32.1%) who had trouble at the start went on to be graded stakes winners, a good percentage. However, the "behind" winners who started cleanly went on to be graded stakes winners 25.9% of the time themselves, a higher rate than seen for the other styles. So "bad starts" does not explain the ultimate competitiveness of the "behind" winners.

Research study, part IV: Interpreting debut margin of victory

This section of the presentation considers margin of victory of the selected debut winners -- what it can tell us about them going forward, and what it reveals about maidens with young horses in general.

First, the sample drops by eight to 822 because I did not register a margin of victory for horses who won by disqualification. There was also the question of what to do with margins of under a half length: I counted nose and head wins as wins by a tenth of a length, neck wins as wins by a quarter of a length, and deadheats as wins by 0 lengths. Those conversions went into calculations involving margin.

Average margin of victory was 2.90 lengths and median margin of victory was 2 lengths. The most lopsided win was eventual Breeders' Cup Juvenile Fillies' winner Cash Run's 15.75-length score at Belmont in 1999.

Below I display wins by "lengths truncated." What this means is that wins by 1, 1 1/4, 1 1/2, and 1 3/4 all go into the 1-length bucket, wins by 2 lengths, 2 1/4 lengths, 2 1/2 lengths, and 2 3/4 lengths go into the 2-length bucket, etc. Percentage of total wins is the first number to the right of the group, followed by percentage of total wins by that margin or less (also known as cumulative frequency). The final column is the percentage of graded stakes winners among horses who won by the particular margin. Taking you through one row, 11.9% of debut winners won by 3-3.75 lengths, and 72.3% of debut winners won by 3.75 lengths or less. The horses in the 3-3.75-length bucket were graded stakes winners 19.4% of the time.

Table 1

<1 25.9 25.9 16.9
1 and 1+ 19.2 45.1 24.1
2 and 2+ 15.2 60.3 25.6
3 and 3+ 11.9 72.3 19.4
4 and 4+ 9.4 81.6 24.7
5 and 5+ 5.0 86.6 39.0
6 and 6+ 4.3 90.9 20.0
7 and 7+ 2.7 93.6 31.8
8 and 8+ 1.5 95.0 8.3
9 and 9+ 1.2 96.2 30.0
10 and 10+ 1.0 97.2 37.5
11 and 11+ 1.8 99.0 60.0
12, 13, 14, 15+ 0.9 100.0 50.0

This is surprising to me, but horses who give their backers anxious moments and win their debuts by less than a length may really be a lesser group than horses who simply win by 1-2 lengths. As the first column after the group indicates, the sample sizes are greatest for those two situations, and the difference in percentage of graded stakes winners relating to them provides food for thought.

A strict reading of the graded-stakes percentages says that winning by 2-4+ lengths does not bode better down the road than just winning by 1-2 lengths, but the actual best read is tricky. The samples in the individual buckets, ranging from 77 to 158 there, are not big enough that we can take the individual graded stakes percentages too seriously. On the other hand, wins of more than 2 lengths and less than 5 comprise 36.5% of the data (300 cases), and wins of 1-5 lengths comprise 55.7% of the data (458 cases). The absence of movement in the GSW % between 1+ and 4+ lengths convinces me that with the desired added data, I would at least find no more than a weak trend between margin of victory and graded stakes winning between 1 and 5 lengths.

I'll offer a couple of other breakdowns: horses who won by 5 lengths or more became graded stakes winners a third of the time (33.1%); horses who won by 8 lengths or more became graded stakes winners 37.7% of the time (20 for 53). Horses who won by 5-8 lengths became graded stakes winners 30.6% of the time.

Beginning the breakdowns with 5 lengths is not arbitrary. Not only did the percentage of graded stakes winners shoot up in the 5 and 5+ bucket, but the steep drop in the 6 and 6+ bucket makes it indisputable that the sample sizes have become small enough to render bucket-by-bucket comparisons completely unreliable at that point on the chart. Picking 8 lengths  as the cut-off for the top margins was mostly arbitrary (or if it wasn't, I can't remember why it wasn't), but was not picked to jack up the graded stakes winning percentage; you can see the 8 and 8+ length horses went just 1-12 by the graded-stakes-winning measure. The sample size dropped from 22 to 12 from 7+ to 8+ lengths, and I think that, in addition to considering how many horses I might want to look at in the top group, guided my decision. The good thing is that you have all of the data, and you can cut and dice it anyway that you like, and I frankly suspect some of you will be far less disciplined and thoughtful than I was.

I'm a little bit surprised that margin of victory by itself cannot support more than maybe a 38% prediction of graded-stakes winning success. After all, narrowing to Hollywood Spring/Summer and Saratoga, without knowing margin of victory, can get us to 35%.

Part of the reason that margin of victory didn't factor more heavily is that it's not the best measure of performance. Speed figures are probably much better, and margin over horses back in the field probably is, too.

My study focuses on style in the debut win, not strength of win. That's why what I'm doing is an interesting approach. I'm asking whether wins that look of equivalent quality really are of different value because of how they are achieved. I included margin because it was so easy to record, and with enough sample size, could serve as a control for strength of win.

Another way of viewing the relationship between margin and success is to start with graded stakes-winning status rather than with margin. If you do that, you see that the average eventual graded stakes winner won by 3.53 lengths, and the average eventual non graded stakes winner won his or her debut by 2.70 lengths. That looks like a reasonable difference, but using something we statisticians and statistical dabblers call effect size, it actually comes out as small.

In Part II I showed that chances of winning decrease dramatically and consistently as horses get farther away from the early lead. Well, horses who do overcome the bias of being well behind early and win, win by less than horses who set or attend the pace. The exact trend seen in winning percentage reproduces itself in terms of margin.

Table 2. Mean margin of victory by style

Led 3.82 lengths
Pressed 3.09 lengths
Middle 2.29 lengths
Behind 1.58 lengths

Table 3. Median margin of victory by style

Led 3 lengths
Pressed 2.5 lengths
Middle 1.75 lengths
Behind 1.13 lengths

Table 4. Largest margin of victory by style

Led 15.75 lengths
Pressed 13 lengths
Middle 11 lengths
Behind 6 lengths

During the couple of months that I retrieved and compiled all of this data, I needed diversions to keep up my spirits. One was charting the biggest-margin "behinder" winner, and another was noting the top margin, which wasn't very large, by conventional standards.Don't Get Mad won, with a 6-length debut win. He was followed by Bella Bellucci, at 5 3/4. None of the other 84 behinders won by more than 4 1/2 lengths.

To get an idea of how stark the difference in the most one-sided wins was between "behind" and the other styles, wins of more than 6 lengths represented 11.8% of the sample over all. From 86 races won by behind-horses, we might have expected 10 wins of more than 6 lengths. Instead, we didn't get any.

Part III showed that the "behind" group has the highest percentage of graded stakes winners and the highest average earnings. We know now that horses who win by more lengths turn out to be better, on average, than horses who win by fewer lengths. We know that the "behind" horses often make it close when they win. So the success of the "behind" group is surprising in light of their typical margins of victory. Since future performance does not show strong signs of differing by winning style, but average margin of victory does differ by style, I think it would be reasonable to assess margin of victory (and probably speed figures too) entirely in the light of style. To say, o.k., this horse won by 3 lengths from behind, but from a normative standpoint, that's as good as winning by 6 lengths on the lead (the frequency of 3+ length wins from "behind" is the same as 6+ length wins among horses who won after leading at the first call).

We need to do another analysis. You might be wondering if margin of victory is really more important than I'm saying -- if the success that the "behind" have despite their low margins of victory is depressing the overall relationship between margin and success. The solution (at least the clunky one without benefit of a statistical model) is to compare the average margin of graded stakes winners and non graded stakes winners not just overall, but within each style.

Table 5.

Led

GSW 5.13 lengths
Non GSW 3.41 lengths

Pressed

GSW 3.54 lengths
Non GSW 2.95 lengths

Middle

GSW 2.51 lengths
Non GSW 2.23 lengths

Behind

GSW 2.03 lengths
Non GSW 1.41 lengths

This is a crude comparison, but the overall difference in margin for GSWs and Non GSWs was 0.83 lengths, and if we average the difference in the four groups above, the difference is 0.80 lengths. So the importance of margin in future success is not obviously greater than I found it to be before.

Making the comparisons of GSWs and Non GSWs by style had one unexpected benefit, though. It reaffirms that margin does matter for behinders, as it does for typical debut winners. And margin seems much more important for the wire-to-wire pattern than for the other progressions.

This makes sense. It is only when a horse is let go from the beginning that we see how fast he can really run, and not in part how fast his opponents were enabling him to go. It's as if the issue shifts to brilliance, or lack thereof, when horses lead from the start, and winning big is a sign of brilliance (effect sizes again suggest not to overstate the relationship, however).

A regional difference emerges charting average margin of victory by meet. Margin of victory is greatest at the New York tracks, and smallest at the California tracks. The pattern is almost perfect.

Table 6. Average margin of victory by meet

1. Saratoga 3.51
2. Aqueduct 3.51
3. Belmont Fall 3.31
4. Belmont Spring/Summer 3.14
5. Keeneland Fall 2.99
6. Gulfstream 2.90
7. Churchill Fall 2.75
8. Del Mar 2.70
9. Churchill Spring/Summer 2.63
10. Hollywood Fall 2.42
11. Hollywood Spring/Summer 2.31
12. Santa Anita Winter 2.25
13. Santa Anita Fall 2.16

Of the 260 debut winners at California tracks, none won by more than 11 lengths. Seven of the other eight meets had winners of more than 11 lengths, and six had more than one. Yet it was not just fewer lopsided or less lopsided wins that were depressing the averages in California, because median margin of victory also has three California meets ranked at the bottom, with another California meet tied for fourth-lowest.

Something seems to be going on here, but I'm not sure I know what it is. I can well believe that the Belmont maidens might have small fields, leading to large margins of victory. But would small fields be an accurate characterization of Saratoga's maidens, or even Aqueduct's? And if field size is the main explanation, are California's maiden races more dense with horses than Kentucky's or Florida's? That comes as news to me.

I entertained the idea that something I didn't track, percentage of debut winners from debut starters, might be lower in California, and that the winning margins the first-time starters register might in turn be more modest. But why then does Hollywood Spring/Summer have such a low average margin of victory, when the races there often pit first-time starters against first-time starters?

In general, the close nature of California maidens just speaks to me of their competitiveness. In California, you have the most graded-stakes-winning graduates, and the closest races. A win in a California maiden special weight means something.

I next tackled whether margin of victory differed in importance by meet. Again, there were indications of a regional difference. It's not a significant difference, but the California graduates who failed to win graded stakes won their debuts by more lengths than the California graduates who did win graded stakes (2.43 to 2.33). Outside California, graded-stakes winners outperformed non graded stakes winners on length of maiden victory, 4.22 to 2.82.

Again, this is a bit of a puzzler. Are there (or were there, since this was 1996-2005) differences in attitudes about runaway wins in California than there were at prominent meets in other parts of the country? Was it more common for winners to be eased up in California, rendering margin meaningless? You might think we're looking at effects of synthetic in California vs. non-synthetic elsewhere, but remember, these maiden wins came before there were any synthetic tracks in California.







Monday, June 25, 2012

Research study, part III: The future success of first-out winners, broken down by style

Without ado, the graded-stakes-winning success-rate of each style.

Table 1

Behind 27.9%
Pressed 23.6%
Led 23.5%
Middle 21.7%


Chi_Square confirmed that the variation here is not suggestive of an effect. In other words, it is a reasonable thought that style of victory has no bearing on future success, or at least that good horses can come from every initial style.

The "behind" percentage is most suggestive, standing out above the remaining three styles. However, let's take a step back: in the qualifying races, horses who won their debuts from behind went on to be graded stakes winners 28% of the time; horses who won another way became graded stakes winners 23% of the time. And that difference is suspect, since the sample for the "behind" winners was just 86, far less than half of any other group. If an aberration was going to occur, it was most likely going to be from this group.

The graded stakes rates for my group of debut winners are in the range of batting averages; the overall rate of .236 is close to the current typical batting average of .250. Think of the sample sizes as at-bats, ranging from 225 for "pressed," to 267 for "middle." We have here three players with about 250 at-bats each, well less than a full season's worth, with batting averages between .217 and .236. There is no likelihood the .236 hitter will finish the season with the best average. The players look indistinguishable. The "behind" player has hit better in just 86 at-bats, but at .279, he has hardly torn it up. If given a full-time job, he were to hit only .200, no one would bat an eye.

A standard error, or average amount the percentage is off just because of the limits of the sample size, can be calculated for each style. For behind, it's .048; for middle, the style with the largest sample size, .025.

Using earnings, "behind" remains on top using average, with almost $30,000 more per debut winner than any of the other styles. But "behind" falls to 3rd using median (although just $7,000 behind "winner" "middle").

Table 2. Mean earnings

Behind $284,931
Led $255,113
Middle $252,092
Pressed $233,606

Table 3. Median earnings

Middle $143,225
Led $137,426
Behind $136,098
Pressed $125,396


That the "behind" group shows better using mean than median suggests some "big horses" in that group, although this characterization must be considered in the light of the highest graded stakes winning percentage, too. In other words, the strong performance of the "behind" group is hardly limited to a handful of horses, even with the sample of just 86 (it contained 24 eventual graded stakes winners).

As an identifier of top, top horses I found the 75th percentile for earnings among all of the graded stakes winners. It was just over $800,000. In turn, I figured the percentage of $800,000 earners by style.

The "led" and "behind" groups may have an advantage here. The rate of the "led" winners may be particularly noteworthy, given the larger sample size pertaining to it.

Table 4

Behind 7.0%
Led 6.8%
Middle 5.2%
Pressed 4.9%


To keep you grounded, the percentages represent 17 $800,000 earners in the "led" group and six in the "behind" group. In order, the 17 top earners in the "led" group were Ashado, Surfside, Southern Image, Discreet Cat, Diabolical, The Cliff's Edge, Lady Tak, Madcap Escapade, Grand Hombre, Grand Slam, Cash Run, E Dubai, First Samurai, Harmony Lodge, Pomeroy, Texas Glitter, and Esteemed Friend.

Based on the data I've reviewed, the "behind" group has plenty of graded stakes winners and a reasonable number of big stars, although not a huge advantage over the other styles in the latter. I thought that perhaps the puzzlingly low median was due to more "flame-outs" than in other groups. I am always on the look-out for traits that correspond to higher rates of injury. And if I found a higher rate of horses in the "behind" group who went on to do just about nothing, I thought that might also mean that there were more lucky, low speed-figure, "clunking up" winners in the group. But comparing the 10th percentile in earnings for the styles, or comparing on similarly low percentiles, yielded little.

Table 5. 10th percentile in earnings

Behind $36,677
Pressed $35,988
Led $35,539
Middle $31,951


If I had to venture a guess, I would say "behind" is on top of this chart because the average winning purse in those horses' maiden special weights happened to be the highest, maybe because of a preponderance of California tracks in that sample. Career earnings of $35,000 don't represent the spoils of much more than a maiden victory. The true "flame-out" percentage is probably higher than 10%, which is why I made comparisons as well at some slightly higher percentiles, without seeing anything terribly interesting. The fact that "middle" has the highest median but the lowest 10th percentile probably indicates that the 10th percentile sheds little light on why the median is what it is.

The picture becomes far more interesting when we compare success in terms of short and long graded stakes winners.

Table 6. Rated by percentage of short graded stakes winners

Led 18.3%
Pressed 16.8%
Middle 16.1%
Behind 12.8%


Table 7. Rated by percentage of long graded stakes winners

Behind 18.6%
Pressed 11.5%
Middle 10.9%
Led 10.8%


Looking at the above data in a different form, and using totals instead of percentages to emphasize the sample size element, here are the number of short and long graded stakes winners in each group, and the ratio between the two.

Table 8

Led 46 27 1.70
Pressed 38 26 1.46
Middle 43 29 1.48
Behind 11 16 0.69


The "behind" group had the highest rate of long graded stakes winners and the lowest rate of short graded stakes winners, with a ratio essentially the opposite of the other three styles. This makes intuitive sense: horses who win from off the pace in sprints are telling you something -- they want to go longer. Lately, however, a belief has taken hold that this assumption is knee-jerk. My data suggest otherwise (unless trainers are forcing their closers into longer distance races, making the assumption that so many sophisticates dislike). In any event, it would seem that the debut come-from-behind sprinters manage acceptably when they are run longer, irrespective of whether they are given ample chance at the stakes level in shorter races as well.

It's an open question whether the pattern of closing debut winners becoming good distance horses has its equivalent in front-running debut winners becoming good sprinters. Significance tests of the ratio of short to long graded stakes winners are complicated by the fact that horses can be in both groups (note there were 59 total graded stakes winners from the "led" group, not the 73 that are the sum of the short and long winners). Still, I will state with some confidence that the distance data for the front-running winners does not indicate special distance proclivities, short or long. Although the short/long ratio is higher for the “led” group than overall, the difference is not striking.

In part I I showed that long graded stakes winners make more money than short graded stakes winners. I therefore wondered now if the advantage the "behind" group showed in mean earnings was a result of the higher percentage of long graded stakes winners among the "behind." The overall superiority in mean earnings for "behind" remains when the sample is restricted to graded stakes winners, and in fact extends to median earnings as well for graded stakes winners. So I was wondering if the "behind" graded stakes winners simply outperformed their counterparts, or whether they just earned the most money because they did more of their good work in routes.

These data were not what I expected. In stark contrast to the overall numbers, in the "behind" category, the eight "short-only" graded stakes winners earned more than the 13 "long-only" graded stakes winners. The edge was $261,455 in mean and $139,810 in median. And among "short-only" graded stakes winners, the "behind" group was easily tops in mean and median earnings. The "behind" group was last in mean and median earnings among "long-only" graded stakes winners. Graded stakes winners in the "behind" group made the most money on the backs of sprinters and horses who went both ways, short and long, not on the strength of the predominant routers.

I went to the data and looked horse by horse at "graded stakes distance type" and earnings, also considering the ability of the horses in a holistic sense. The pattern I present does not stand out to my naked eye. There were some excellent sprinters in the "behind" group (Midnight Lute, Dream Supreme), but they seemed offset by just as many good distance horses. Perhaps if I examined the graded stakes winners of other groups, the difference would be more phoney graded stakes-winning sprinters in those groups.

The take-away question may be, if the "behind" group can produce some really outstanding sprinters, doesn't it stand to reason that it should be able to produce more of them? Perhaps we have evidence for the rare if veritable "come-from-behind" sprinter. Or perhaps the discrepancy between the frequency of graded stakes types and grade I types suggests that the low number of short-only "behind" graded stakes winners is flukish.

A couple of additional notes from the earnings data, at the risk of going too deeply into it given its likely non-significant nature. The key to the "middle" group's top median earnings was the earnings of the non-graded stakes winners, which were $111,494 on the median, over $16,000 more than any of the other styles...Athough the category's percentage of graded stakes winners was right on the overall study average, I'm convinced after reviewing the earnings' data that the "press" winners had the worst subsequent performance. The median earnings of $410,497 for the press graded stakes winners, compared to $552,784 for the remaining styles, is particularly glaring.

Research study, part II: Debut victories categorized by style


Holding off further discussion of debut winners and future success, my database provides good data on how often horses of each style win. The number of winners in the led, pressed, middle, and behind categories:

Table 1

Debut winners who led at first call: 251 (.302)
Debut winners who pressed at first call: 226 (.272)
Debut winners in the middle at first call: 267 (.322)
Debut winners behind at first call: 86 (.104)

Early leaders, pressers, and mid-pack first-time starters won in almost equal amounts, while back-of-the-pack winners were rare. However, the totals do not really give us the data we want. In order for the data to be of practical significance, we need to adjust for the fact that the three bottom categories usually comprise more than one horse in a race (see definition of the categories in Part I).

I didn't track field size in my sample. However, from another project, I have a dataset of 12,097 races from July 2007 to July 2008. I tracked number of betting interests there, which is very close to field size (probably differs by less than .1 on the average). From that sample, I was able to narrow to non-state-bred maiden special weights with 2-year-olds or 3-year-olds only, between 4 and 7 furlongs on the dirt, with five betting interests or more, and not originally carded for the turf. That left 238 races. By printing the freqency table for betting interests in those races, and multiplying the percentage frequencies by the number of led/pressed/middle/behind horses in each case, I was able to estimate an average number of horses associated with the led, pressed, middle, and behind categories in a race in my study. (The led category has one horse per race. The pressed category has one horse if the field has fewer than nine horses, and two if the field has nine horses or more. Behind is identical to pressed, except that the entries are two with fields of eight or fewer, and three with fields of nine or more. The pressed and behind average, then, is as simple as knowing the percentage of races with nine or more horses. It's 42.9%, so pressed has an average of 1.429 places in a race, and behind an average of 2.429. The average for the middle category is the only one that is an involved calculation.)

These numbers were

Table 2

Led: 1
Pressed: 1.429
Middle: 3.566
Behind: 2.429

The numbers sum to 8.424 (the average field size in the dataset). What we next want is is the percentage of places that each category covers (i.e., convert Table 2 into percentages).

These are

Table 3

Led: .119
Pressed: .170
Middle: .423
Behind: .288

Divide the number in Table 1 by the corresponding number in Table 3 and we get how much more likely than average each position is to win.

Table 4

Led 2.538
Pressed 1.600
Middle 0.761
Behind 0.361

Divide "Led" in Table 4 by the other categories in Table 4 and we discover that a first-time starter on the lead is 1.59 times as likely to win as a first-time starter who presses the pace, 3.34 times as likely to win as a first-time starter who lays mid-pack, and a whopping 7.03 times as likely to win as a first-time starter who comes from the back of the pack. The true rate of victories descends as the horses analyzed get farther back in the pack.

This is an expected conclusion, duplicating research done by I don't know how many people how long ago. The most effective styles in the debut itself were not something I was overtly studying, but since I have the data, I thought I would pass them along. Now I'm very curious to know, however, how these rates compare to non first-time starters in similar races, and to the rates in races with older horses. Does the winning percentage drop less steeply by descending category in those races? I would like trainers to know if they are particularly sacrificing a potential win if they do not send their first-time starter to or near the front, and are truly giving their horse a race by not doing so. This is something they should know, rather than just blindly sending their horse to the post, thinking the way he or she is ridden doesn't matter, and hoping for the best. Without comparable data for other kinds of races, however, it's unclear how strong of a message trainers should take from the rates I found.

Astute readers might pick up on one hole in the analysis. I assume that the led/pressed/middle/behind places are represented in the same proportion by first-time starters as overall. This isn't necessarily true. By their numeric presence, first-time starters may have a 28.8% chance of occupying one of the "behind" slots in a race. But if they are less sharp than horses who have run, they may in fact occupy the "behind" slots more frequently. If that is true, then the winning percentage of 36.1% average for behind horses is overestimated. They are even less likely to win than that.

                                                 The breakdown by meet

I compared number of led/pressed/middle/behind winners at each of the meets. While the evidence for real differences and patterns is strong, the meets cannot just be compared at face value. The percentage of starters that fit into each category are not exactly equivalent to table 3 but differ some depending on the particular distribution of field sizes at the meet. The smaller the average field, the higher the rate of "led" wins will be. Larger fields necessarily aid the aggregate win rates of the other styles, although the exact relationship between field size and expected win rate is unique for every style, and complex. Taking "pressed," for instance

Table 5. Col 1 fld size, Col 2  possible winners from style, Col 3 expected w%

5 1 .2
6 1 .17
7 1 .14
8 1 .13
9 2 .22
10 2 .20
11 2 .18
12 2 .17

Large fields are generally good for the pressed winning percentage, but very large fields are not as good as merely large ones, and five-horse fields are good as well. Given how nuanced the relationship between starters and style categorization percentage is, speculating on difference in biases from one meet is another is well-nigh impossible without hard data. My dataset of similar races might have shed some light, but the number of races per meet would have been frightfully small, since that dataset contained less than a third as many races as this one, many of which were at meets not included in this study. With just the raw data of number of winners by style, field size cannot be ruled out as a force behind different debut-winner style-distributions at different meets. Some of the quirks in the relationship between starters and expected winning percentage may diminish differences between meets, in any event. Again, this does not hold for "led", where there is a simple and inverse relationship between starters and winning percentage.

The style percentage of debut winners by meet, with meets presented chronologically, and percentages reflecting the led, pressed, middle, and behind styles from left to right:

Table 6

Belmont Spring/Summer 47.9, 21.9, 23.3, 6.8
Churchill Spring/Summer 34.1, 34.1, 27.3, 4.5
Hollywood Spring/Summer 35.4, 33.3, 27.1, 4.2
Del Mar 31.5, 28.8, 27.4, 12.3
Saratoga 37.2, 24.8, 28.9, 9.1
Belmont Fall 27.1, 22.0, 35.6, 15.3
Keeneland Fall 21.7, 32.6, 37.0, 8.7
Santa Anita Fall 21.9, 18.8, 37.5, 21.9
Aqueduct Fall 23.4, 25.5, 38.3, 12.8
Churchill Fall 23.6, 30.9, 36.4, 9.1
Hollywood Fall 16.7, 27.8, 36.1, 19.4
Gulfstream 30.4, 27.2, 33.6, 8.8
Santa Anita Winter 21.1, 29.6, 38.0, 11.3

I would attribute the very high "led" rate at Belmont Spring/Summer to a few factors. I believe Belmont's Spring/Summer maiden special weights typically drew very few horses. In that respect, the high percentage of wire-to-wire winners (well, first-call-leader winners; included are horses who lost the lead at one of the intervening calls) is artifactual.

Belmont's standard 2-year-old maiden special weight distances at that time of year are 5f and 5.5 furlongs, and those distances probably favor front runners more than 7f races, say. The emphasis of this study was on future success, and arguments can be made for either 5f or 7f maidens being a superior proving ground for top horses, or for neither distance being more telling in a first win. So I was comfortable including some range in distances. But it’s fairly intuitive that the shorter the maiden, the more important prominent early placement is in that race that day, making distance a variable that needs to be eliminated or controlled for in simple comparisons of wins by style. So you can count distance as one of those variables for which I now see there was some use, but left out of the initial data collection in the interests of time.

The relationship between time of year and effectiveness of style is difficult to gauge in general. In the spring/summer, Churchill's and Hollywood's front runners trail Belmont's dramatically, yet their percentage of debut wins is only surpassed by one of the other nine meets. The baby races at Churchill Spring/Summer and Hollywood Spring/Summer are usually short, too. Are the spring/summer 2-year-old maidens kind to frontrunners because early speed is at a premium with very young horses, or because early speed is at a premium in very short races? The bright side of distance and time of year being as strongly related as they are is that there are relatively few races at under 6f late in the year when one want separate estimates of winning based on early position.

A factor that might separate Belmont in the spring and summer from Churchill and Hollywood is that the Belmont 5f races feature a very short run to the first turn, and not being on the lead often means going wide and giving away lengths.

The spring/summer races are also distinct for their unkindness to "behind" runners. The low percentages in that category for Belmont, Churchill, and Hollywood don't seem to belong with the other meets.

There were nine such winners at the spring/summer meets, and I checked to see if they were guaranteed stars, but an expected two of nine went on to be graded stakes winners, and four earned less than $100,000. (One year before the study, however, there was a future Kentucky Derby winner who came from last place to take his debut at Belmont in June, and by 5 lengths -- Grindstone).

It might be coincidental, and the sample sizes are the study's smallest, but I saw some symmetry in the high "behind" winning percentages at Santa Anita Fall and Hollywood Fall. I can't think of a good artifactual explanation, of a characteristic of the qualifying races at these meets that would favor debuting closers more than other meets. So the success of the debuting closers may really indicate that good horses of that kind tend to debut at those meets.


It certainly appears that as the crop ages, speedy debut types have less and less of an advantage. Actually, at risk of analyzing the data too finely, I would submit that the front-runner advantage is particularly pronounced in the spring and summer meets, somewhat reduced at the boutique summer meets, and then stable at an again lower level through Gulfstream and Santa Anita Winter. The breakdown appears this way.

Table 7. Proportion of debut winners who led at first call

Spring/Summers 40.6%
Del Mar and Saratoga 35.1%
After Del Mar and Saratoga 24.6%

Given the general large fields and the carding of races 6f and longer, the 37.2% front-runner proportion at Saratoga might be the most notable of all in that category. We can perhaps attribute some of it to Saratoga frontrunners being potential superstars in the making, but if you see part III in my series, not very much. I'm such an opponent of the track-bias perspective, that for all the times I've been there, I can't tell you whether Saratoga is generally thought to be a speed-favoring track or not, but a speed-favoring track could certainly be a part of the high "led" proportion.

















Sunday, June 24, 2012

Research study, part I: The future success of first-out winners, as moderated by time of year and winning track

                                      What I was thinking and what I did

I am now going to embark on a series of posts about debut winners. The general topic is, how can we tell if one is going to be good? And what is the general outlook for them?

I had to create a database. When you create a database, the general rule of thumb is to include as many variables as you can, because you'll be surprised when one of them will come in handy. You don't know what your final analysis is going to look like, and the focus may shift. On the other side, in the slow, largely manual way that I compile databases, I have learned that too many variables make the data entry unmanageable. Additionally, while the array of variables leaves the world open to you, that also can be a curse, and the analysis, while, far-reaching and powerful in its conclusions, takes a long time to complete.

So in this study, I limited the number of variables I collected, fully aware of all of the questions I would be leaving open. I was also interested in a very specific aspect of debut winners' prospects: namely, which running style(s) translated to future success? Countless times I have looked at a Racing Form, and fancied this second-time starter or that, depending on whether leading the field from the gate seemed impressive to me at the moment, or coming on like a Zenyatta or a Grindstone. Horses run fastest at the beginning of races, at least on dirt, so perhaps that is how true abilty is most often signaled. But the horse who comes from behind and wins does something rare, and may be the one with potential down the road. The closer may be the late bloomer, the diamond in the rough. The truth was that, without doing research, I just didn't know which style was more impressive.

In choosing the parameters of my study, I wanted them to mirror the exact conundrum I had been in countless times at the races. I restricted myself to studying dirt, sprint maiden special weight winners. State-bred races were excluded, as were maiden claimers. I only looked at horses who won at the 'A' meets in the country, which I defined as Gulfstream in the first three months of the year; Santa Anita in the first three months of the year (plus the 12/26 - 12/31 period); all NYRA racing, June through the Thanksgiving weekend; Churchill in June and July; Keeneland in October; Churchill in occasionally October, and always November, through Thanksgiving weekend; Hollywood June and July; Del Mar; and Santa Anita and Hollywood in the fall. Certainly, Keeneland in April is an 'A' meet, for instance, but I was interested in tracking the best 2-year-olds, and April is very early for good 2-year-old racing. I studied 2-year-olds, plus 3-year-old-only maiden special weight winners for the winter Santa Anita and Gulfstream meets. I had a couple of other picayune rules: no off-the-turf races; no fields of fewer than five horses. Sprints were defined as 7f or less (the 'about' 7f distance at Keeneland is 7.28 furlongs and so was not included).

I wanted to study the most recent data possible, but I needed to use a number of years to obtain adequate sample size, and I also needed enough years to have passed since the winners' debuts to see how they turned out. I needed them to be retired. So I used the years 1996-2005.

I believe some truths that I found to be timeless, basic realities of racing, and others to be of questionable longevity. Not being current is a problem with longitudinal research.

I used Equibase's on-line historical charts, found the debut winners who fit my criteria, and then recorded their career accomplishments as shown by "Equibase Horse Search." My outcome variables were whether the horse was a graded stakes winner; whether he or she was a graded stakes winning in races up to a mile; whether he or she was a graded stakes winner in races over a mile; and the horse's career earnings. I did not record whether the horse was a plain stakes winner; there are lots of phony stakes winners going around. Graded-stakes-placed might have been nice to have. But I don't think other outcome variables were going to improve but very marginally over graded stakes-winning status, particularly not when earnings supplemented it.

Grade I winners are just too rare to be able to make comparisons across groups; plus, again, I had earnings. This enabled me to look at the very high end of accomplishment if I wanted to.

Debut winners were placed in one of four categories, according to the following method. Only the first call of the chart was used (in other words, the first one where there's a rendering of how far back or in front the horse is). Horses on the lead went into the "led" category; horses who were 2nd at the first call in fields of 8 or fewer horses, or 2nd or 3rd in fields of 9 or more horses, went into the pressed category; horses who were last or second-to-last at the first call in fields of 8 or fewer horses, or last, second-to-last, or third-to-last in fields of 9 or more horses, went into the "behind" category. All other debut winners went into the "middle" category. These divisions were decided from the beginning of the data collection, and were not a post-hoc rendering to make particular points. I in fact did not record the precise data of the horse's initial position, instead just the category.

A final variable I collected was margin of victory.

                                                Overall success rate, and success rate by meet

A first question does not compare debut winners but instead looks at the general prognosis: what is the ordinary outlook for the types of good prospects I collected? We have all heard bearish statistics, and know the chances of a graded stakes winner for a starter never hits 10% no matter who his sire is, but how does this percentage change with a successful debut?

From a sample of 830, my overall finding was 23.4% graded stakes winners, $251,375 mean earnings, and $133,942 median earnings. I think some owners will find these numbers encouraging, and others will find them discouraging. I suppose I should have asked racing-enthusiast friends for predictions, so that I could understand which way the surprise factor leans.

Perhaps you are wondering how to tweak the earnings statistics for the 2012 environment....Looking at how much money the debut winners went on to make from each year, there isn't a pattern of more recent debut winners going on to do better than the overall mean and median. Data for the 2005 graduates were a $236,314 mean, for instance, and a $135,720 median, with 21.3% eventual graded stakes winners. The years above the overall median in earnings were 1997, 1999, 2002, 2003, and 2005; the years above the overall mean, 1999, 2000, 2002, and 2003. Multiple years in the first half of the study made the top half for both mean and median.

While I'm on the division by years, the range in the number of qualifying debut winners was narrow, ranging from 75 to 89 a year. So we would expect each year to represent 10% of the sample, and in fact the range is from 9.0% to 10.7% of the sample. Changing foal crop numbers, and the emergence of turf racing, have not appreciably changed the number of maiden special weight sprints on the dirt for young horses. Or perhaps these races are more infrequent, but there are more debut wins now.

If I had continued the study past 2005, I would have had a decision to make about including synthetic races, and if I had decided that in the negative, the pool of races would have decreased. But none of the meets I studied had synthetic tracks in the '96-'05 period.

16.7% of the debut winners eventually won short graded stakes, and 11.8% eventually won long graded stakes (see opening section for definitions of short and long). 5.1% were dual short and long graded stakes winners. So horses beginning successfully in the classic way, by winning their sprint debut, were more likely to win short graded stakes than long graded stakes.

It would be worthwhile to compile all graded stakes by distance, and see if there are also more short than long. If not, then does winning at a sprint distance signal a sprint proclivity, setting aside style? Does winning first out indicate that a horse is almost "too fast" for his own good, and is not likely to matter much at longer distances? Or maybe, while beginning in routes is comparatively rare, perhaps that is the pathway for horses who most often do best in consequential routes down the road?

The issue would be hard to assess. I saw few winners in my sample logging top turf careers, and the overall ratio of long to short graded stakes is raised greatly by turf races. The fact that dirt sprint debut winners almost always stay on dirt is interesting in itself, particularly if this trend is as extreme as it seemed to me. The idea that a talented turf horse may sometimes start as a dirt horse, or fool his trainer first out with a bang-up dirt performance, may not be play out often in reality....

This is not specific to debut winners, and may be better addressed with more expansive datasets, but if you can get a graded stakes winner, what kind of money are you looking at earning? Median earnings for graded stakes winners were $484,761.50. Horses who only won graded stakes at a mile or less had median earnings of $358,148.50, while those who only won graded stakes at over a mile had median earnings of $586,789. Part of the advantage for graded stakes routers probably consists of higher purses in those races; another possibility is just that the graded stakes routers are better, more consistent, or more dominant overall than their counterparts. Short and long graded stakes winners, guaranteed of being multiple graded stakes winners, had the highest-earning careers of all, with median earnings just shy of a million ($965,346).

Despite my hopes of including tracks all of a certain level, there is evidence that the 23.4% graded stakes winners rate is not always 23.4%, and does depend where the horse wins. Indeed, using Chi-Square, graded stakes status by Meet was significant at the .05 level. Obviously, I wasn't necessarily expecting Aqueduct Fall debut winners to grade out as well as Saratoga ones, but I wanted general comparability so that Meet would not be an overriding factor. To the extent that it is, at least it is one I can account for.

                                         Rated by percentage of graded stakes winners

1. Hollywood Spring/Summer 35.4%
2. Saratoga 34.7%
3. Del Mar 28.8%
4. Belmont Fall 25.4%
5. Hollywood Fall 25.0%
5. Santa Anita Fall (Oak Tree) 25.0%
7. Santa Anita Winter 22.5%
8. Keeneland Fall 21.7%
9. Belmont Spring/Summer 19.2%
10. Churchill Spring/Summer 18.2%
11. Gulfstream 16.8%
12. Churchill Fall 16.4%
13. Aqueduct Fall 8.5%

Although sample sizes are small, the order is strikingly consist with reputation. It doesn't always seem to me that Saratoga debut winners run faster than Belmont debut winners, for instance, but either they do, or they have qualities that don't come out in the race time that allow them to do better later.

The surprise may be Hollywood Spring/Summer. Winners from qualifying races from the meet who went on to be graded stakes winners were A. P. Warrior, Buffythecenterfold, Came Home, Diplomat Lady, Dixie Union, Double Honor, Essence of Dubai, Exchange Rate, Hookedonthefeelin, Inspiring, Miss Houdini, Purely Cozzene, Roman Ruler, Ruler's Court, September Secret, What a Song, and Worldy Manner. It's natural to think the success of the graduates may have had more to do with the time of the year than anything else, but if you can find a pattern with time of year and eventual success, looking at the other meets, you're seeing something I don't.

Although only 36 maiden winners from the meet made the study, Hollywood Fall also rated very well. When we shift from graded stakes winners to earnings, Hollywood Fall is on top for both mean and median (barely, but still). In general, the ratings with earnings followed the ratings with graded stakes winners closely. For instance, the meets 2nd, 3rd, and 4th in mean and median earnings (although not in the same order) were Hollywood Spring/Summer, Saratoga, and Del Mar -- 1, 2, and 3 on graded stakes winner percentage.

An exception to the correspondence between earnings and graded stakes winners was Oak Tree; tied for 5th with Hollywood Fall in graded stakes winner percentage, Oak Tree was 10th among the meets in average earnings, and last in median earnings. The smallest sample size in the study (n of 32) is just one reason I wouldn't necessarily expect the poor performance to repeat if I examine later years.

Sprint/long graded stakes success is difficult to make much sense of on a meet by meet basis (again because of sample size), so I tended to aggregate in meaningful ways to find trends. One may be that the California tracks had a higher percentage of their graded stakes graduates earn the distinction sprinting. Aggregating the five California meets, 56 debut winners proceeded to win short graded stakes, and 32 proceeded to win long graded stakes (a ratio of short to long graded stakes winners of 1.75). At non-California tracks, the corresponding numbers were 82 and 66, for a ratio of 1.24.

The ratio is different, but not overwhelmingly different, to my eye. The most probable cause is simply the ratio of short and long graded stakes in California run compared to at other tracks over the past 15 years. Another possibility, and an intriguing one, is that the trainers who win first out in California (i.e., Bob Baffert) are more speed than distance trainers (not that Baffert's nine Triple-Crown race wins are something to gloss over when assessing his distance-training ability).

When I speak of non-California tracks, Saratoga is far and away the biggest contributor of graded stakes winners, so must be assigned much of the credit (or blame) for the sprint/long graded stakes winner ratio for the whole sample. (The Saratoga ratio was 29-24).

Saratoga's rate of 19.8% qualifying winners proceeding to win graded stakes over a mile at some point in their careers is very impressive. It also makes it clear that Saratoga's 34.7% 2nd-place overall rate of graded stakes winners is not a function of horses returning in one of the couple of graded stakes that complete the meet. Those races are sprints, and the 19.8% long graded stakes winners displays that Saratoga is not making its own, phony graded stakes winners (I guess my worry here was a little absurd, franky, since plenty of other meets have 2-year-old stakes intended for previous maiden winners).

Getting back to differences in sprint/route success in California vs. elsewhere, an interesting comparison is the Santa Anita winter meet vs. the Gulfstream winter meet. Santa Anita Winter had 13 sprint graded stakes winners, and just 7 long ones; Gulfstream had 13 of both.

There is additional data that I don't have that could certainly enable us to better understand the findings. It would be nice to know the relative chances of a firster and a non-firster winning at each meet, and it would be nice to know the future success of the non-firsters winning at each meet. Which meets have the best 2-year-olds and early 3-year-olds in their maiden special weights, and which just have the most formful maiden special weights (i.e. have the eventual top horse win first out the most often)? These are fascinating questions to me, but more descriptive of a process, than revealing about future prospects.




Vindication for Silverbulletday -- yikes!

Silverbulletday's 3-year-old daughter Elusive Jackpot was bet off the board yesterday at Hollywood for her debut but was no match for her stablemate Distracting and ended up 6th after showing excellent quicks. It sure is a shame Silverbulletday was wasted twice on Vindication. In fact, Elusive Quality, sire of Elusive Jackpot, isn't quite a match for Silverbulletday, either, when one remembers she was one of the greatest filly competitors ever, and wasn't ancient at the time of the matings. Silverbulletday's role in her several unraced foals can't be dismissed, however.

Saturday, June 23, 2012

Repole's claiming frenzy

I wonder if Todd Pletcher is wholeheartedly into Mike Repole's current claiming frenzy? Are number of stalls, size of operation not something Pletcher has to worry about? If there is a Calibrachoa or a Caixa Electronica in the group of recent claims, I guess he can live with them, but the program seems like a distraction from what he is trying to do, or what I perceive he is trying to do, which is to be the most successful trainer in the United States. I'm not sure that having this high-end claiming division goes along with that. I suppose Pletcher has a ways to go to have as high a percentage of claimers as Steve Asmussen has, however. Repole alone can't move him in that direction, but this development has been odd because it's been so extreme.

It seems like Repole is trying to dominate the owner standings the way that Sheikh Mohammed has at times tried to dominate the yearling sales. What he's doing comes off as egotistical and selfish, as trying to accentuate how powerful he is.

It might be good for the economics of racing in New York to have so many horses moving, but on the other hand, it could be keeping down entries because people are nwe truly afraid of losing horses. This could be checked or argued about in relation to field size for various races, although that is complicated by other factors, such as the increased purse money compared to years prior.

Tuesday, June 19, 2012

When horses are no better than what they're seeking: the Dispute rule, as demonstrated by Astrology on 6/16

It didn't occur to me until a quarter of a mile into Saturday's 3rd at Churchill, but race favorite Astology fit a tenet I've developed through the years, which is that good established horses running in soft spots are not generally good bets. The first race I saw that made me consider the merits of this philosophy was an ungraded stake in the winter of 1994 with Dispute in it. Dispute had won the Kentucky Oaks, Gazelle, and Beldame the previous year, all grade Is. As the layoff was not particularly long, and the competition not particularly, she struck me as much more of a 1-5 shot than the 7-10 she offered, and I lamented the surrounding stupidity. But Dispute finished 3rd that day and contiued to run sluggishly, until taking the Spinster in her final start of the year.

Race conditions are designed to produce evenly matched horses. We do have heavy favorites sometimes, however, and legitimately so. Connections still tend to be conservative with young horses and try an allowance race rather than a stake in the aftermath of a strong maiden win. There's nothing generally suspicious about a 3-year-old maiden winner proceeding to a nw1x allowance. We also find maidens who ran fast in their previous start but ran into the wrong horse. There's nothing suspicious in their taking advantage of the available maiden race once again. Favorites in starter races seem to be most reliable of all.

But conditions do not do all of the work in leveling a race's playing field. If Astrology was the horse who won or placed in five consecutive graded stakes, culminating with the Preakness, over a 9-month period, he wouldn't have been running in Saturday's allowance/optional claimer. Not after already having his race back as a 4-year-old, an allowance, and having won it. On form, Astrology may have been better than a nw3x allowance horse, but Steve Asmussen was saying by running him in the race that he wasn't, or at least not markedly so.

The public responded to Astrology with some caution, making him 3/5. If he had been lower, the race would have been very playable. With that amount of betting on Astrology, the race was generally unplayable by the percentages, despite a $1,781.20 trifecta resulting. You see I am not saying that the playing field was entirely level, or that Astrology should not have been the strong favorite. Only that his form could not be taken entirely at face value with the chosen spot.