Does the Draw Order Decide Horse Show Placings?

Every exhibitor knows the feeling: you draw late in a big class and wonder if the judge has already made up their mind. I’ve now got enough results in the equine-data database to actually test that question properly.

Across 1,748 classes and 191,605 placings, horses drawn later place marginally better on average — but the pooled effect is now small enough that it can’t be separated from zero. Class-to-class variation swamps it. Draw order is, at most, a tie-breaker, not a deciding factor.

The pooled weighted Spearman rho between draw order and placing is −0.005 (95% CI −0.018…+0.007), and only 50.3% of classes show any negative association at all. Here’s the honest picture.

Based on 191,605 placings · analysis refreshed 27 Sept 2026.

Updated 27 September 2026. This post was first published in August 2026 on 934 classes and 86,281 placings, where the pooled rho was −0.019 with an interval entirely below zero. The dataset has since more than doubled, and the headline got weaker, not stronger: mean rho drifted from −0.019 to −0.005 and the confidence interval now straddles zero, while the median class rho went from −0.028 to −0.004. The first-draw win-rate edge also eased from 1.45× to 1.25×. The last-vs-first draw tercile contrast is the one number that held its ground (−0.019 → −0.018, interval still below zero). Numbers below are the current ones; the August figures are noted where they differ.

Per-class correlation distribution

For each class with at least 10 riders I computed the rank correlation (Spearman’s rho) between draw order and placing. Negative rho means later draws placed better; positive means earlier draws did. The histogram below is the real spread across all 1,748 classes — not just the pooled average:

Green bins are classes where later draws placed better (negative rho); tan bins are the reverse. The distribution is a bell sitting almost exactly on zero — the tallest single bin is actually the first positive one (254 classes between 0.0 and +0.1) against 250 classes in the −0.1…0.0 bin beside it. Median rho −0.004, 50.3% of classes negative, and the full range still runs from −0.80 to +0.80. One pooled number hides all of that spread, and this histogram is the honest picture.

Draw terciles

Within each class I split riders into three equal groups by draw rank and averaged their place percentile (0.0 = win, 1.0 = last; lower is better). Pooled across all classes:

Later draws average −0.018 percentile points better than early draws across the whole range — the interval below is the bootstrap 95% CI on that difference:

The whole interval sits below zero, so this contrast is real — but it is −0.018 percentile points across the entire draw range, and it is the only pooled measurement here that still clears zero. That’s far smaller than typical class-to-class variation, and it is not the same statement as “drawing late helps”.

Does going first matter?

The most striking single number on this page is actually about the first draw, not the last. Among the 718 classes of 10+ riders where the first-drawn rider posted a numeric result (1,239 more classes are excluded because their first draw scratched or never placed), first-goers won 8.5% of their classes — against 6.8% expected if the draw carried no advantage (the mean of 1/n across the same classes). That’s a 1.25× win rate.

The direction survives larger fields but the gap has narrowed since the August cut (1.45× on 355 classes then, 1.25× on 719 now) — exactly what you’d expect if part of the earlier number was the small sample being flattering.

Broken out by discipline, the first-draw edge is concentrated in the western pattern and hunter classes:

The average placing is a quieter story: first-goers sit −0.035 percentile points from the rest of the field (95% CI −0.058…−0.012) — an interval entirely on the better side of zero, though far smaller than the win-rate gap.

Caveats. There’s only one first-goer per class, so this compares classes against each other, not riders within a class. The disciplines above have modest n’s and no correction for multiple comparisons across ~20 disciplines — with draws like these, one or two will look strong by chance. A higher-than-odds win rate is associative, not causal: the draw is a schedule artifact, and classes that draw first may differ in other ways.

By discipline

Pooled weighted rho per discipline for every discipline with at least 10 classes. Negative = later draws placed better; positive = earlier draws did.

The old three-discipline story got messier with more data. Horsemanship and Showmanship — the two clear “later draws don’t help” exceptions in the August cut — have converged on zero (+0.008 and +0.038, both barely off it), and Hunter Under Saddle flipped sign but stayed small (+0.019 vs −0.043 before). What survives is a split by class type: the largest negative figures now sit in cattle and speed-adjacent work (Cutting −0.103, Roping −0.083, Reining −0.051, Western Riding −0.043, Trail −0.041), while the positive ones are the timed speed events and Longe Line (Barrel Racing +0.121, Pole Bending +0.111, Barrels +0.094, Longe Line +0.127) — where going later means a chewed-up arena or a warmer clock, so the sign flips the other way.

Methodology & caveats

How it’s measured:

1. Within-class percentile normalisation — raw places are meaningless across classes of different sizes (6th of 8 is not 6th of 60), so each rider’s outcome is place_pct = (place − 1) / (n − 1); draws are normalised the same way.
2. Spearman’s rho per class — rank correlation between draw order and placing for every class with ≥ 10 riders (average ranks for ties; DQ and non-numeric rows excluded). 1,748 classes qualify today, up from 934 in August.
3. Pooled weighted mean + bootstrap CI — class rhos pooled as a size-weighted mean; 95% interval from a 2,000-iteration bootstrap resampling classes with replacement (fixed seed, reproducible).
4. Draw terciles — riders split into three equal groups by draw rank within each class.
5. First-goer test — only classes where the first-drawn rider posted a numeric result count; the benchmark is the mean of 1/n over the same classes, i.e. the win probability the draw alone would predict.

Read before quoting:

• Draw semantics vary by class type. In some classes the draw is a random start order; in others it’s a scheduled slot, and scratches/re-entries can shift it afterward. One pooled number averages over all of these.
• DQ rows are excluded — only rows with a numeric place and a draw number enter the analysis, using the Combined Judges roll-up placing.
• The pooled effect is no longer distinguishable from zero (−0.005, CI −0.018…+0.007). Only the last-vs-first tercile contrast still clears zero, at about 1.8 percentile points — far smaller than typical class-to-class variation.
• Doubling the data weakened the signal, and that’s the useful lesson — an effect that shrank toward zero as n grew was mostly small-sample optimism. Treat single-cut anomalies in your own results with the same suspicion.
• Per-class variation is the interesting part — the pooled mean hides a lot of spread, and the histogram above is the honest picture.

This is one of a series of data articles built on my equine-data results database — a personal AQHA show-results explorer that analyzes every class in the dataset.