Key takeaways
- Every probability sits between 0 and 1. Zero is impossible, 1 is certain, and 0.25 is the same as 25% or “1 in 4.”
- Independent events multiply. Two dice both showing six is 1/6 × 1/6 = 1/36, about 2.78%.
- Unions need the overlap removed. P(A or B) = P(A) + P(B) − P(A and B).
- “At least one” is a complement problem. Compute the chance of none, then subtract it from 1.
- Mutually exclusive and independent are different ideas. Disjoint events with nonzero probability are dependent.
What a probability is
A probability is a number between 0 and 1 describing how often an event would happen across an indefinitely long run of identical trials. For equally likely outcomes, P(A) = (outcomes in A) ÷ (total outcomes). A fair die has six outcomes, so P(even) = 3/6 = 0.5 and P(rolling a 4) = 1/6 ≈ 0.1667.
The same quantity appears in three formats: 0.25, 25%, and “1 in 4.” They are interchangeable, and mixing formats inside one calculation is a reliable way to produce nonsense. Probability describes the process, not the next trial: ten tosses producing six heads is unremarkable, while ten thousand will land near 0.5.
The rules in one place
| Rule | Formula | Use it when |
|---|---|---|
| Complement | P(not A) = 1 − P(A) | You want “none” or “at least one” |
| Addition | P(A or B) = P(A) + P(B) − P(A and B) | Events can overlap |
| Addition, disjoint | P(A or B) = P(A) + P(B) | Events cannot both happen |
| Multiplication | P(A and B) = P(A) × P(B) | Events are independent |
| Conditional | P(A | B) = P(A and B) ÷ P(B) | You know B happened |
Each rule is bookkeeping for counting outcomes; when a result looks wrong, return to the sample space and count.
Complements: the “at least one” shortcut
“At least one” is usually the hardest phrase to compute directly and the easiest to compute backwards. Rolling at least one six in four rolls means exactly one six, two, three, or four — four cases. Its complement is one case: no sixes at all. Each roll misses a six with probability 5/6, so four independent misses have probability (5/6)⁴ = 625/1296 ≈ 0.4823. Subtract from 1: 1 − 0.4823 = 0.5177, or about 51.8%. Four rolls is barely better than a coin flip, which surprises most people.
The same pattern covers redundancy: two alarms that each detect a fire with probability 0.9, independently, both fail with probability 0.01, so at least one sounds 99% of the time. Adding 0.9 + 0.9 = 1.8 would be meaningless.
Independent events multiply
Two events are independent when the occurrence of one leaves the probability of the other unchanged. Then P(A and B) = P(A) × P(B). Two dice: P(both six) = 1/6 × 1/6 = 1/36 ≈ 2.78%. Three coin flips all heads: 1/2 × 1/2 × 1/2 = 1/8 = 12.5%.
Independence is an assumption to check, not a default: sampling with replacement keeps draws independent, sampling without does not. A pump and controller working with probabilities 0.98 and 0.97 both work with probability 0.9506, so the pair fails 4.94% of the time — close to the naive 5%, but the gap matters when a service contract is priced on it.
Unions add, then subtract the overlap
For events that can overlap, P(A or B) = P(A) + P(B) − P(A and B). Draw one card from a standard deck. Hearts are 13/52 and face cards are 12/52, but three face cards are also hearts, so P(heart or face) = 13/52 + 12/52 − 3/52 = 22/52 ≈ 0.423, or 42.3%. Skip the subtraction and those three cards are counted twice, inflating the answer to 25/52.
When events cannot happen together the overlap is zero, so P(A or B) = P(A) + P(B). Rolling a 2 or a 5 on one die is 1/6 + 1/6 = 1/3. The test for disjointness is P(A and B) = 0, not whether the events feel unrelated.
Mutually exclusive is not independent
These terms get swapped constantly. Mutually exclusive means the events cannot occur together; independent means one occurring does not change the other’s probability. For events with nonzero probability the two are opposites: if A and B are disjoint and P(A) > 0, then P(A | B) = 0, so knowing B happened rules A out entirely.
A card cannot be a heart and a spade, so those events are mutually exclusive. A card being a heart says nothing about whether it is a face card, so those events are independent. Confusing the two over-counts unions or multiplies probabilities that should have been added, and the error survives any amount of careful arithmetic.
Conditional probability and draws without replacement
Conditional probability rescales the sample space to the cases where B is known true: P(A | B) = P(A and B) ÷ P(B). If 50% of days are cloudy, 30% are rainy, and 25% are both, then P(rain | cloudy) = 0.25 ÷ 0.50 = 0.50. Rain is twice as likely on a cloudy day as on a random day.
Drawing without replacement is the standard dependent sequence. The chance both cards are aces is 4/52 × 3/51 = 12/2652 = 1/221 ≈ 0.452%. With replacement it is 4/52 × 4/52 = 1/169 ≈ 0.592%. The first card changes the deck, so the second draw is conditioned on the first.
Odds, risk, and “1 in 4”
Odds and probability are different formats for related information. If an event has probability p, the odds in favor are p ÷ (1 − p): a 25% chance is odds of 1:3, spoken “one to three,” not 1:4. Converting back, probability = odds ÷ (odds + 1).
The distinction matters in medicine and insurance, where “1 in 4 users reported a side effect” is a probability, while “four times more likely” is a ratio of two probabilities and can sound far larger than the underlying risk.
Common mistakes
- Adding probabilities of overlapping events without subtracting the intersection, which can push the total above 1.
- Multiplying events that are not independent, such as two cards from one deck or two failures sharing a power supply.
- Treating “at least one” as a single trial instead of a complement calculation.
- Assuming past results change independent future ones. A coin that has landed heads five times still has a 50% chance of heads on the next toss.
- Ignoring the base rate when reading a test result, which turns a 95%-sensitive test for a 1% condition into a 16% post-test probability.
- Rounding intermediate probabilities and then multiplying the rounded values, so the error compounds across several factors.
Practical caveats: base rates and small samples
Base rates dominate screening results. Suppose 1% of a population has a condition, the test detects 95% of true cases, and 5% of healthy people test positive. In 10,000 people, 100 have the condition and 95 of them test positive; 9,900 are healthy and 495 test positive. The share of positives that are real is 95 ÷ (95 + 495) ≈ 16.1%. A positive result is still more likely to be a false alarm than a true case.
Independence is also often false in the real world: weather, markets, and equipment failures cluster because they share causes. When independence cannot be defended, conditional probabilities or a simulation are the honest tools. Small samples mislead too — with 100 fair trials the standard deviation of the count is √(100 × 0.5 × 0.5) = 5, so 60 heads is only two standard deviations out.
Frequently asked questions
Can a probability be greater than 1?
No. Probability is bounded by 0 and 1, inclusive. A result of 1.8 means overlapping events were double-counted, or a union rule was applied to a joint event that should have been multiplied.
When do I add and when do I multiply?
Add for “or” when the events cannot happen together, and use the general addition rule with a subtraction when they can. Multiply for “and” when the events are independent. If the events are dependent, use a conditional probability for the second factor.
Are mutually exclusive events independent?
No. If two events cannot both happen and each has a nonzero probability, then one occurring makes the other impossible, which is dependence in its strongest form.
What does “at least one” mean in a probability question?
One or more occurrences out of a set of trials. Compute it as 1 minus the probability of zero occurrences, which is almost always shorter than adding the probabilities of one, two, three, and so on.
What is the difference between probability and odds?
Probability is favorable outcomes divided by all outcomes. Odds are favorable outcomes divided by unfavorable outcomes. A 25% probability is 1:3 odds, while 1:4 odds correspond to a 20% probability.
If something has a 1% chance per try, what is the chance in 100 tries?
About 63.4%, because 1 − 0.99¹⁰⁰ ≈ 0.634. The expected number of occurrences is 1, but “expected” is an average, not a guarantee, and the chance of at least one is well short of 100%.
Further reading
Keep exploring
Decision path