Published · · Last reviewed · · 6 min read
English · 한국어
"Auto wins more often": what that statistic really means
"Most jackpot winners used auto picks" comes up whenever Lotto 6/45 is discussed, and it is true. Pool the 983 rounds from 262 to 1247 that had a first-prize winner and you get 9,336 winners: 6,187 auto (66.3%), 2,885 manual (30.9%) and 264 semi-auto (2.8%). The trouble starts with the next sentence. "So auto is better" and "auto is worse because you cannot choose" both fail to follow from these numbers. This page shows why, using logic and a simulation. The data themselves are covered in the split-risk article; here we focus on how to read such a statistic.
A similar mistake: customers at a café
Suppose a café reports that "70% of this month's refund requests came from people who ordered an Americano". Does that make the Americano a problem item? If 70% of the café's sales are Americanos, 70% of refunds is perfectly normal. What you need is the share of all orders that were Americanos. Without it, "70% of refunds" tells you nothing.
Auto versus manual is exactly this situation. "66.3% of winners were auto" has meaning only once you know what share of all games sold were auto. That figure is not published. It is an unverified item.
A deeper reason: the odds are the same
The café analogy has one weak link: menu items can genuinely differ in refund rates. A lottery cannot, by construction. An auto game is six numbers filled in at random by a machine; a manual game is six numbers chosen by a person. In both cases the game is one of 8,145,060 combinations of 6 from 45. The draw picks one combination at random and does not look at how any ticket was chosen. A single game therefore has a 1 in 8,145,060 chance of the jackpot regardless of method.
So the question is not "which method hits more often" but "what share of sales was auto, and did winners follow that share?" The simulation below shows it. Every game has identical odds; we vary only the auto share of sales and generate 9,336 first-prize winners.
With the auto share set to 66%, simulated histories give an auto share of winners that moves roughly between 65% and 68%. Set it to 50% and the result moves near 50%. The share of winners follows the share of sales; the method itself does nothing for winning. With 9,336 winners the 95% range is about ±1 percentage point (a standard deviation of roughly 0.49 points at 66.3%). The real 66.3% fits the assumption that auto was about 66% of sales, but that is a back-calculation, not data on actual sales.
The same number can support the opposite conclusion
The same statistic could be used to argue that manual is better. If auto were 80% of all sales, 66% of winners would suggest auto games underperformed. If auto were 50%, auto games would seem to outperform by a factor of two. Since the real odds are equal, those differences are illusions created by the assumed denominator. Without knowing the sales split, no conclusion can be drawn.
That is the core lesson. To interpret a number fairly, first ask what it must be compared with. A ratio without a denominator makes a good story and a poor proof.
What the auto/manual split does tell us
The statistic is not useless; it answers a different question. Look at how the manual share changes with the number of winners in a round.
| Winners per round | Rounds | Auto | Manual | Semi-auto |
|---|---|---|---|---|
| 1 | 12 | 75.0% | 25.0% | 0.0% |
| 2–3 | 59 | 80.2% | 19.1% | 0.6% |
| 4–7 | 314 | 72.4% | 25.8% | 1.7% |
| 8–15 | 492 | 68.0% | 29.1% | 3.0% |
| 16 or more | 106 | 55.7% | 40.7% | 3.6% |
In crowded rounds the manual share exceeds 40%. This is a signal that can be read without knowing the sales split, because the comparison is between rounds under the same method. It points toward "manual games tend to cluster on similar combinations". The difference is not in odds but in overlap. The right conclusion is therefore not "auto is better for winning" but "if you win, the chance of sharing the prize may be higher with manual picks".
A caution applies here as well. Grouping rounds by winner count builds in a selection effect, so this shows association, not causation. And manual players are varied: the point is not that every manual choice clusters, only that on average clustering is greater.
Four questions to ask of a similar statistic
- What is the denominator? Is the full population behind the ratio stated?
- Is there a structural reason? If two groups are built to have equal odds, a difference in shares most likely reflects sales volume or selection bias.
- Is there a selection effect? Analyses that split by outcome easily read cause backwards.
- Could the opposite conclusion also fit? If the same number supports the opposite claim, it is not yet evidence.
Questions that come up
Does buying auto change the chance of winning? No. Auto means the machine fills in the numbers for you; the game is still one of 8,145,060 combinations. There is no route by which choosing auto makes a match with the winning numbers more or less likely.
So are there simply more auto winners because more auto tickets were sold? That is the simplest explanation and it is consistent with the calculation here. We cannot confirm it without the sales split, but if auto were around 66% of sales, 66.3% of winners being auto is exactly what we would expect.
Is it true that manual picks are more likely to be shared? What the data show is an association: the more winners in a round, the larger the manual share. On average it suggests manual choices tend to cluster on similar combinations. It does not mean any individual's manual picks will overlap, and a selection effect is mixed in.
And semi-auto? Semi-auto means you pick some numbers and the machine fills the rest. Its share is a small 2.8% of all winners and it does not appear in the data before 2015, so this page offers no interpretation of semi-auto.
Can this statistic be tested directly? It could be, if sales by selection type were published. Until then the best we can do, as in the simulation here, is show how results change under assumed sales shares.
What the sample size tells us
The simulation range above is narrow because 9,336 winners are plentiful. With the same 66% but only 30 winners, the 95% range would be about ±17 percentage points and no conclusion could be drawn. More observations narrow the width due to chance, but a narrow width does not make the denominator problem go away.
Limits and what is not verified
- The sales split between auto and manual is not verified. The 66% estimate here is back-calculated from the equal-odds structure.
- The mapping of the data fields to selection types rests on one report on round 1128 (11 auto, 52 manual); we could not open the official field documentation (press report).
- We could not confirm the official definition of "winner" as people or games.
- The auto/manual breakdown is absent before round 262, so those rounds are excluded.
However numbers are chosen, one ticket's jackpot chance is the same, and the negative expected value does not change. The generator on this site produces random numbers in the same way auto selection does and does not raise the chance of winning. More on how choices affect sharing continues in The birthday-number trap.
LottoFlick is independent and not an official lottery operator. Numbers are random and cannot be predicted.
Sources and verification
- Donghaeng Lottery: Lotto 6/45 round results (first-prize winners by selection type) — verified
- LottoFlick aggregate output: 983 rounds, 9,336 first-prize winners — computed by LottoFlick
- Dailian report on round 1128 (field-to-label check) — press report