How to Analyze a Lottery: 20 Methods and When Each One Helps
In serious lottery analysis there is no single magic indicator that tells you "bet on these numbers." Any statistic drawn from the draw archive is noisy — deviations from a uniform distribution are inevitable, even when the drum is perfectly fair. The player's job is not to find a "lucky" number but to look at the archive from several different angles and make a decision where the methods agree. Our site brings together about two dozen analysis methods — and each one captures its own aspect. Let's go through them group by group.
Why one method is not enough
Imagine you are deciding whether to trust the number 17 in the MillionDAY lottery. If you look only at frequency, 17 may come up rarely — and that looks like a "cold" number. But if you open the Z-score, it turns out that 17's deviation from the mean is statistically insignificant — just ordinary noise. And the draws-between-hits view will show that 17 has been missing for only two draws in a row — no intrigue at all.
This is a typical situation. Each method sees one slice of randomness and may contradict the others. An experienced analyst does not trust a single indicator — they build a consensus from three or four different methods and look for numbers that stand out in several at once. In the US Powerball lottery, where the bonus ball is drawn from a separate, small drum of numbers, this approach is especially clear — the pool is small, and random deviations stop being noise the moment you layer several filters at once.
Basic metrics for every ball
Everything starts here. Five indicators that everyone needs — from beginners to players with a decade of experience.
Frequency — how many times a ball has come up across the whole archive. The simplest metric, but a deceptive one: with a large archive the frequencies of all balls converge toward the average, and the differences become noise.
Hot numbers — those whose frequency is above the average. Sorted by descending frequency.
Cold numbers — the opposite: balls with a frequency below the average.
Draws-between-hits — how many draws have passed since a ball last appeared. If a number has not shown up for 40 draws, its draws-between-hits value is 40.
Ball weight — a derived metric equal to draws-between-hits divided by frequency. The idea is simple: if a ball comes up often but has not appeared in a while, its weight is high, and under the "regression to the mean" theory the chance of seeing it soon is greater than for a consistently "cold" one.
Formally, weight is computed as:
Weight = Draws-between-hits / Frequency
The larger the value, the stronger the ball's "skew" toward a long absence despite overall activity in the archive.
Ball weight often gives a more meaningful picture than frequency: it combines two signals on its own. But it is no panacea either — we'll discuss why further on.
Deviations from the uniform distribution
These methods answer the question "how much does a ball's real frequency differ from what would be expected under fair randomness?" If the deviation is large and stable, there is reason to take a closer look. If it is small, it is noise, and betting on such a number is no better than a random pick.
Z-score — the normalized deviation of a ball's frequency from the average across all balls. A Z-score of 1.5 or more is a significant deviation, above 2 is strong. A Z-score of -1.5 or less is a significantly "cold" number.
Pearson's test — the classic chi-squared test. It shows how well the distribution of frequencies matches a uniform one. A large value → it makes sense to look at which balls exactly are deviating.
Bernoulli — the probability that a ball came up exactly as many times as it did, under fair randomness. A low value means that this number of hits is unlikely to occur by chance.
It is convenient to combine them: Z-score gives a quick filter, Pearson gives an overall assessment of how "correct" the archive is, and Bernoulli helps gauge the reliability of a single specific ball.
Time series: how a number's behavior changes
While the previous methods look at the archive as a whole, these look at its dynamics. Useful when a lottery runs often and the behavior of balls noticeably drifts over time.
Moving average — a ball's frequency averaged over a window of the last N draws. You can see whether a ball is getting "hotter" or, conversely, cooling down.
Trend chart — a visualization that combines the moving averages of all balls. It helps you spot groups moving together.
Delta — the difference between neighboring draws. A number's average delta shows how regularly it appears.
Markov chains — a transition matrix from draw to draw: if 7 came up yesterday, with what probability will 12 come up tomorrow? In a fair lottery these probabilities should not differ from uniform ones, and that is exactly why the method is interesting: significant differences are either a drum artifact or random noise, which is also useful to see.
Time-based methods are especially vivid on lotteries with a fast rhythm — for example, Keno, where a draw runs every few minutes and quickly accumulates large samples. In rarer games like France Loto or US Powerball the trends build up over years.
Randomness tests and proprietary methods
This group of methods does not predict specific numbers — it answers two other questions. First: does the archive behave like a random stream at all? Second: what other angles on the same data yield new information? The answers matter, because if the archive is random on all tests, the hypothesis "this very number will come up soon" is suspect from the start. And if some method shows a stable deviation, it is worth including in the consensus.
Autocorrelation — whether a draw depends on previous ones. In a fair lottery autocorrelation is close to zero. Stable deviations are either a drum artifact or a systematic data error.
Runs test — checks how runs of even and odd, high and low numbers are distributed. Too even or too "clumpy" is a potential anomaly.
Benford's law — the distribution of the first digit. In lotteries with a limited range the law works nonlinearly, but the deviation chart itself is an informative structural marker.
Shannon entropy — a measure of the "chaos" in the frequency distribution. The higher the entropy, the closer the archive is to perfect uniformity. A drop is a signal to look at which category of balls stands out.
Positional statistics — how often a number comes up in a specific position within a draw. In some lotteries the order is fixed, and this gives an additional slice.
The Karnaukh method — a proprietary ranker that combines frequency, draws-between-hits, and the density of hits in recent windows into a single number.
The HarCHO method — an alternative proprietary algorithm with a different weighting formula. It provides an independent estimate to cross-check against Karnaukh.
The slicing method — splits the archive into equal segments and compares the behavior of balls within each. It helps you see whether a number's "behavior" has changed over time.
In the vast majority of draw-based lotteries, randomness tests show that the archive is simply noise. This is not a disappointment but an honest answer: it means attempts to "beat" the drum with statistics are useless, and a sensible strategy comes down to choosing a lottery with better odds and consciously managing your budget, rather than hunting for "hot" combinations. The proprietary methods (Karnaukh, HarCHO, slicing) are useful not because they predict numbers but because they show the archive from angles that basic frequencies miss — and help you avoid false conclusions from a single ranking.
When "hot" by frequency does not match "hot" by Z-score
This is the most common trap for novice analysts. You open the list of hot numbers and see the number 23 there. Then you check the Z-score — and 23 has a value of only 0.4. A contradiction? No, these are different questions.
Frequency answers: "does this number come up more often than others?" Z-score answers: "does it come up significantly more often, or is it random noise?" A number can rank first by frequency in the top list, but with a Z-score of 0.4 its "hotness" is no different from a random fluctuation. Conversely, a number with a modest frequency but a Z-score of 1.8 hints: the gap from the expected average is statistically noticeable.
A good rule: first sort by frequency, then filter by Z-score of 1.5 or more. This removes from the shortlist all numbers where "hotness" is just sampling luck. Supplement this with a filter by ball weight, and you are left with a short list that can already be checked through Pearson and Bernoulli.
In practice: a consensus of three or four methods
Below is a table on how to build a decision from different methods in practice. This is not a guarantee of winning — it is a way to avoid betting blindly.
Goal Methods for consensus How to decide Find "undervalued" numbers Frequency + ball weight + Z-score Look for a number that is simultaneously at the top by weight and has a Z-score of 1.5 or more Confirm "hot" numbers Hot numbers + Pearson + Bernoulli Take the top 10 hot numbers, but filtered by significance Determine whether there is a trend in the archive Moving average + trend chart + autocorrelation If all three agree on "there is a trend," it is more likely real Assess the "randomness" of the archive Pearson + entropy + runs test If the metrics consistently show noise, there is no classic anomaly Test the hypothesis "numbers move in pairs" Markov chains + positional statistics Examine transitions and positional correlations
First choose the goal, then the methods — not the other way around. If you simply open all twenty pages in a row and try to "see a pattern," your brain will inevitably find something, even when there is nothing there. This is called apophenia, and in lotteries it works especially ruthlessly. If you want a precise calculation of the probability of a specific combination based on your analysis, use the odds calculator.
A useful discipline is to record your decisions. Write down which numbers you chose and why after the analysis, then check against the draw. After 20-30 draws it will become clear which combinations of methods give meaningful signals in your lottery and which are just coincidences. For number lotteries like US Powerball or MillionDAY this approach works faster than waiting for a "lucky" draw: you learn to tell structural signals from random ones and stop betting on what "looks promising."
What follows from this
A single indicator decides nothing. Trust only a consensus of three different methods.
Start with the basics: frequency → draws-between-hits → ball weight. This is often enough for 70% of tasks.
Check significance with the Z-score — a number that landed among the "hot" ones by frequency should have a Z-score of 1.5 or more, otherwise it is just sampling noise.
Time-based methods (moving average, trends) are more useful on frequent lotteries like Keno than on rare ones like France Loto.
Randomness tests (autocorrelation, runs test, entropy) do not predict numbers — they assess whether it makes any sense at all to look for patterns in a particular archive.
Proprietary methods (Karnaukh, HarCHO, slicing) are a supplement to the basics, not a replacement. Use them as a sixth point of view, not as a "secret formula."
Choose the goal, then the methods. Twenty statistics pages opened one after another create the illusion of analysis, not analysis itself.
