How to determine a track bias

[EMD4ME]

As they hit the 1/4 of race 1 yesterday, I already suspected that a dead rail was in play.

My lone speed, who jogged to the lead on the rail (and who had evidenced that she has 2 moves) just stopped outside the 1/4 while running on that quicksand.

The rest of the card, featured a dead rail.

Annoyingly, the 2nd race had an inside post wire BUT the apprentice was smart enough to go wide immediately and STAY wide.

Race 3 featured a 3 horse field with the 1/5 outside speed winning.

Race 4 featured a terrible donkey benefitting from an outside stalking trip at short odds , arrghh

[classhandicapper]

I'm just bumping this thread for now. I want read it again and may want to continue the discussion.

[classhandicapper]

I make my own automated speed/closer bias ratings with fairly strict conditions.

Days are assigned a rating of either Honest, Favored Closers, Strongly Favored Closer, Favored Front Runners, or Strongly Favored Front Runners.

To get a bias rating a day requires a minimum sample of 5 races on that surface with the same track condition, 4 of the 5 races have to be strongly tilted in the same direction, and the average tilt in that direction to be quite strong. (6 races would 5-1, 7 races would 6-1, 8 races would 6-2, and 9 races would be 7-2)

Each race looks at the entire race flow and not just what the front runner did.

The standards are so tight only 5% of all races get flagged as probably biased. If you saw a list of the days and looked at the charts, it would be clear that something unusual happened that day by either by remote chance or because the track was actually biased.

I use this as a tool to help in my analysis when I watch replays and look at the charts more carefully.

In the last couple of weeks I added a feature that automatically upgrades and downgrades horses that ran on days identified as biased (assuming the race they came out of also flowed that way) .

Then I tested how all the upgrades and downgrades did when the came back going back to 2014. To my disappointment, they didn't do much in terms of outperforming the take on either side of upgrading or downgrading.

I'm of the view that no automated system can do as well as a close competent subjective analysis of a card like I do anyway, but my first pass at hard evidence was pretty disappointing in terms of finding value.

[MJC922]

To each his own when it comes to handicapping research techniques but I personally would never look at outperforming the take as evidence of anything predictive. Outperforming the take is a very high-level metric and would be relevant way downstream IMO, like your own bottom line profit loss at the end of the year while using the data is when that becomes relevant. It should be far removed from the low-level predictive research IMO.

What I would be looking for is for example with a contour bias if the rail is dead I'd want to have a y/n for whether the horse was on it. Then I'd want to see a performance rating for the races the horse has run, can be BSFs or whatever, choose your poison, so if it's Beyer and we have 87, 78, 87, 85 but the 78 was on what we think was a dead rail day then that is score one for a piece of 'evidence'. When you look at the 25 or so horses on the rail I'd want to see similar. The question would be how dead is it? So that horse was about 9 points slow, if the avg for the others is about 5 points slow that would be the metric I would record.

Seems to me something like that 'might' be useful when it reaches a certain point + or - Certain folks do make path variants. I've never done that myself but I thought I heard Paul M. did and maybe still does.

In the beginning I was more of a track bias guy, I mean as a student of Beyer and Davidowitz it would be hard not to turn out that way but as the years went on I moved more into the camp of it being mostly noise / coincidental.

[mountainman]

I've tinkered with formulas to represent potential short-term trends that impact the performance of particular running styles. Path bias, in my opinion, is too murky for such measures. Too many factors come into play, and the painstaking exactitude needed for info gleaned through visual observation could even be counter-productive.

Regarding all biases, I am old-school: trust your judgement, but error on the side of caution.

[classhandicapper]

Originally Posted by MJC922

To each his own when it comes to handicapping research techniques but I personally would never look at outperforming the take as evidence of anything predictive. Outperforming the take is a very high-level metric and would be relevant way downstream IMO, like your own bottom line profit loss at the end of the year while using the data is when that becomes relevant. It should be far removed from the low-level predictive research IMO.

I once did a manual study of horses I upgraded/downgraded based on biases using Beyer figures to see if the fluctuations in their figures were consistent with my upgrades and downgrades. They were, but not as strongly as I would have hoped. It was also complicated by the fact that so many of the horses were switching distances, surfaces, trainers, etc.. I also think on some level biases get built into the figures. I elaborated on that earlier in this thread.

The reason I chose to do it with ROI this time, aside from my prior experience with figures, was that I also want to know if there any value in it.

I just added a single condition to the test.

The horse had to be returning on the same surface.

That improved the results. So I am more encouraged to keep looking at the data.

[classhandicapper]

Here a couple of very preliminary Beyer tests on my strong upgrades and downgrades based on track bias (A lot more work needs to be done).

Strong Upgrade Turf - Next Beyer averaged 4.22 points higher
Strong Downgrade Turf - Next Beyer averaged 1.15 points lower

Strong Upgrade Dirt - Next Beyer averaged .14 points higher
Strong Downgrade Dirt - Next Beyer averaged 4.27 points lower

So clearly, the automated Bias Ratings and automated Upgrades/Downgrades are picking up useful information. The key from here might be to refine bias method, refine the upgrades/downgrades, break it down further between speed favoring tracks and closer favoring tracks, sprints/routes, put some rules in place for the horse coming back on the same surface and a similar distance etc.

No matter what though, even at this basic level, you can see that bias related flow info is a valid concept. It's just a matter of tweaking it to maximize the results and understanding where it works best.

[Jeff P]

It's definitely a valid concept.

Once you work your upgrades and downgrades into a model --

Or if you aren't using a model, once you work your upgrades and downgrades into your overall process --

You should start seeing incremental improvement in your results.

Imo, as you go forward, and as you acquire data for new races, it's also important to keep making incremental improvements to the upgrade/downgrade process itself.


-jp

.

[classhandicapper]

Originally Posted by Jeff P

It's definitely a valid concept.

Once you work your upgrades and downgrades into a model --

Or if you aren't using a model, once you work your upgrades and downgrades into your overall process --

You should start seeing incremental improvement in your results.

Imo, as you go forward, and as you acquire additional new data, it's also important to keep making incremental improvements to the upgrade/downgrade process itself.


-jp

.

That's exactly what I've been doing with my pace/race flow process for several years now.

I've had automated bias ratings for the same period of time, but I just used them to flag me towards days I might want to review manually and then put in my notes if I felt there was actually a bias. Now I've automated the upgrade/downgrade for bias too. So I can actually test both the bias ratings and the ups/downs and improve both like I did with the pace/race flow stuff.

I wanted to publish this much at least so that people can see this is not just some theory. Here's some data.

[46zilzal]

Easy: whenever chronic quitters hold on race after race....Most prominent when the conditions are both cold and wet. Speed biases usually only in the East as that is where the cold wet weather mostly is.

[Aerocraft67]

I perused this epic thread to see what I'm missing. I accept track bias, but I only use basic factors available in PPs.

I've tried to identify short term bias (probably seeking an excuse for why my clever bet lost). But my feeble attempts haven't produced anything discernible from other factors.

I initially thought jockeys would mitigate the impact of bias. Surely they're aware of any consequential bias. They'll adjust accordingly. But Ruffian made a compelling post against this. Although I still find it hard to accept that jockeys can't adjust to bias at least as well as handicappers can identify it.

How do we translate post position into path position? The jockey can choose the pace and path, but not the post. If the second path is most favorable, surely it's easier to reach from the 2nd post than the 9th. But if the 3rd path is unfavorable, it's not like post 3 is doomed to race there.

Maybe pace bias is less surmountable than path bias. And more important to consider accordingly. You can't make a closer a frontrunner to exploit a speed bias without compromising her performance. But jockeys can still time their moves to adjust.

Which leads me to consider trip analysis trumps bias analysis. If he was stuck on a dead rail or moved too soon into a more tiring surface, isn't that performance a function of trip as much as bias?

Given a bias, it seems tougher to predict whether a runner succumbs to it. You have to identify the unfavorable condition as well as the likelihood of avoiding it.

Identifying a short term bias seems as worthy as any other special factor in the race. But bias also seems perilously indiscernible, obvious, or fleeting. No doubt sharps are beating me with it. But they do that many ways.

A few posts here suggested that bias analysis is most useful in evaluating past races and adjusting the merit of those performances. That seems more consistently useful, if not as exceptionally lucrative.

Maybe what's at least as valuable as a confirmed bias hiding from the crowd is a more popular but vulnerable bias that you can bet against. Also suggested in the thread. I'm as apt to take a basic bias statistic at face value as I am to forgive in looking for a longer shot.

[classhandicapper]

A few posts here suggested that bias analysis is most useful in evaluating past races and adjusting the merit of those performances. That seems more consistently useful, if not as exceptionally lucrative.

I think we are slowly getting to the point where it can be a very significant factor in the outcome of a race but getting progressively more difficult to detect and use effectively.

A a lot of tracks they run 4-5 turf races and 4-5 dirt races each day. That automatically gives you a smaller sample size of races to evaluate on each surface.

Some tracks have multiple turf courses (which could be playing differently).

There are 1 turn and 2 turn races that could b playing differently.

All this is going on while the maintenance crew is adding water between races or allowing it to evaporate and changing the surface during the curse of the day.

So even you are sharp enough to recognize a legitimate bias after watching just a few races, buy the time end of the card comes along, the maintenance crew could have flipped the bias on his head. The rest of the time the sample is so small you can't be sure.

[classhandicapper]

I just reread my last post. I need to start proofreading my spelling and grammar a little more.