Did you know? Sometimes a binomial question does not tell you how many trials there are. Instead, it gives you the expected number of successes and the variance, then asks you to reconstruct the model. You can recover both parameters from just those two pieces of information—provided the numbers describe a valid binomial distribution.
For students and parents researching a JC1 H1 Mathematics tutor in Bukit Timah, this is a useful example of mathematical reasoning. Memorising B(n,p) is one skill. Working backwards to find n and p, checking that the number of trials is a positive integer, and deciding whether the model is appropriate are deeper skills. This guide makes each decision explicit.
Quick answer: for X ~ B(n,p), the mean m = E(X) = np and the variance v = Var(X) = np(1 − p). Divide the second relationship by the first when m > 0 to get v/m = 1 − p. Thus p = 1 − v/m, and then n = m/p. Finally check 0 < p ≤ 1, n is a positive integer, and all stated information is consistent.
Reading route: definitions · first solution · invalid parameters · new probabilities · practice.
Start with what n and p mean
The parameter n counts the fixed number of trials. A trial could be the independent inspection of a manufactured item or a repeated yes-or-no experiment in an invented classroom story. The parameter p is the probability of whichever outcome the problem defines as a success.
A ‘success’ is simply an event label. The event might be a defective component, a correct answer or a person selecting a category. Its everyday desirability is irrelevant. If the random variable X counts the number of that chosen event, then n and p describe how many opportunities there are and how likely that event is on each opportunity.
A binomial model needs a fixed number of trials, only two relevant classifications per trial, independence between trials and the same p each time. Those conditions are part of the question, not consequences of the two formulas. Even plausible values of n and p do not justify the model if the process violates its assumptions.
- n: positive whole-number count of trials; not an estimated continuous time.
- p: probability of the event labelled success; must be between 0 and 1 inclusive.
- q = 1 − p: probability of the complementary event within a trial.
- X: number of successes among the n trials, so it takes integer values 0 through n.
- E(X) = np: long-run mean count across repeated sets of n trials.
- Var(X) = np(1 − p): theoretical variability of that count.
- SD(X) = √[np(1 − p)]: standard deviation, which is not the variance.
The official 2027 H1 Mathematics 8865 syllabus includes the binomial distribution as a probability model, conditions under which it is suitable, and its mean and variance without proof. The SEAB MF27 formula list shows the binomial probability, mean and variance formulas. Reconstructing n and p is a practical way to connect those listed relationships.
Why the two formulas are enough
Suppose the mean and variance are both stated. At first, there are two unknowns: n and p. The equation m = np says that the product of trial count and success chance is fixed. That alone does not identify n and p uniquely; many pairs can produce the same mean.
The variance adds a second relationship. Since v = np(1 − p), it equals the mean multiplied by the failure probability 1 − p. Dividing v by m removes np and leaves 1 − p. That is the useful observation.
If m is positive, the algebra is simple: v/m = 1 − p, so p = 1 − v/m. Once p is known and positive, n = m/p. Do not skip the final validity checks. We have solved two equations, but valid probability parameters have further restrictions.
If a problem gives zero mean, this division step is not permitted. A zero mean can arise from p = 0 under a positive trial count, and it does not uniquely identify n. A thoughtful student recognises this edge case rather than blindly dividing by zero.
Worked example one: recover twenty trials
An illustrative random variable counts successes from an independent fixed-trial experiment. We are told its mean is 6 and its variance is 4.2. The question says it follows a binomial distribution; find n and p.
- Write E(X) = np = 6.
- Write Var(X) = np(1 − p) = 4.2.
- Divide variance by mean: 4.2/6 = 0.7 = 1 − p.
- Therefore p = 0.3.
- Use np = 6 to obtain n = 6/0.3 = 20.
- Check that n = 20 is a positive integer and 0 < 0.3 < 1.
The recovered model is X ~ B(20, 0.3). Reinsert its parameters: the mean is 20 × 0.3 = 6 and variance is 20 × 0.3 × 0.7 = 4.2. Both values agree with the original statements.
This last verification is not optional decoration. A student might have reversed the ratio and accidentally found p = 0.7, which would give variance 4.2 only if the mean happened to be 6 with a different relationship. Substitution into both formulas catches the mistake.
What does n = 20 mean in the story?
It means twenty trials belong to one repetition of this experiment. It does not mean there are twenty successes. The expected number of successes is six; the realised X could be any whole number from zero through twenty.
In a school question, the trial may be a tested component or an independently answered item. The random variable’s definition should be restated so the student does not confuse the trial count n with the average count np.
Worked example two: start from a mean of 4.8
A second made-up model is said to be binomial with mean m = 4.8 and variance v = 3.84. Recover its parameters without guessing a trial count from a calculator list.
The variance-to-mean ratio is 3.84/4.8 = 0.8. Thus 1 − p = 0.8 and p = 0.2. Next n = 4.8/0.2 = 24. The model is X ~ B(24,0.2).
Check: np = 24 × 0.2 = 4.8 and np(1 − p) = 4.8 × 0.8 = 3.84. Both original quantities are recovered.
The non-integer mean 4.8 is not a problem. Expected value is a weighted long-run average; it does not have to be one of the integer counts X can take in one experiment. But n itself must be a positive integer, which is why the checks on these two quantities differ.
Worked example three: mean 3 and variance 2.25
Let X have a binomial distribution with E(X) = 3 and Var(X) = 2.25. Calculate p = 1 − 2.25/3 = 1 − 0.75 = 0.25. Then n = 3/0.25 = 12.
So X ~ B(12,0.25). Check its variance: 12 × 0.25 × 0.75 = 2.25. This worked example is intentionally smaller, making it easy for learners to test their arithmetic without relying on an electronic solver.
The example also makes a relationship visible: if p = 0.25, the variance is 75% of the mean. That happens because Var(X)/E(X) = 1 − p for a non-degenerate binomial model.
Why a binomial variance cannot exceed its mean
Since v = m(1 − p) and 0 ≤ 1 − p ≤ 1, the variance of a binomial count cannot exceed its mean. In the usual case 0 < p < 1 and m > 0, the variance is strictly smaller than the mean.
For instance, an alleged binomial model with mean 5 and variance 7 cannot satisfy the formulas. If we try p = 1 − 7/5, we get p = −0.4. Negative probabilities are invalid. The correct response is not to report a negative success chance; it is to recognise that the two quantities are inconsistent with a binomial distribution.
This is an excellent pre-calculation test. Before dividing numbers, compare mean and variance. If the variance is larger than a positive mean, pause immediately. A teacher can use this to test whether the pupil understands model constraints, not just algebra.
When the data cannot describe a binomial model
A non-integer trial count
Suppose a hypothetical problem states E(X) = 5 and Var(X) = 3. The ratio gives p = 1 − 3/5 = 0.4. Then n = 5/0.4 = 12.5. The arithmetic is correct but the result cannot be the exact trial count of a binomial model, because a trial count must be a whole number.
If these are exact stated values and the model claims a fixed integer number of trials, there is no such B(n,p). If the mean and variance were rounded estimates from real data, they might be close to an integer-parameter model, but that is a different statistical estimation question requiring careful treatment. Do not silently round n from 12.5 to 13 and claim the equations are exact.
A variance greater than the mean
As noted, E(X) = 5 and Var(X) = 7 yields an invalid negative p. A correct statement is ‘These cannot both be the theoretical mean and variance of a binomial random variable.’
A variance equal to zero
For n > 0, variance np(1 − p) is zero if p = 0 or p = 1. In those cases the outcome count is deterministic: always zero or always n. The ratio method may be usable when m is positive and v = 0, giving p = 1 and n = m, provided that n is an integer.
When the mean and variance are both zero, p = 0 for any positive n, so n cannot be uniquely identified from those two figures. This is a useful mathematical exception to the routine formula.
A variance with no sensible units
If a problem supplies a mean of six successes and a variance of 4.2, variance describes squared success-count units. In a count model those numbers are often presented numerically without emphasising the squared unit. Be aware that comparing a standard deviation of 4.2 to a mean of six would be a different claim; do not treat variance and standard deviation as interchangeable.
What if standard deviation is given instead?
Suppose X is binomial with mean 6 and standard deviation √4.2, approximately 2.0494. Its variance is the square of that standard deviation: 4.2. Then the first worked example applies, giving p = 0.3 and n = 20.
A common error is to place SD = 2.0494 directly into v = np(1 − p), even though v denotes variance. The correct preparation line should read ‘mean = 6; variance = 4.2; SD = √4.2’.
If the problem gives an SD rounded to two decimal places, squaring it will produce an approximate variance. Use enough precision and interpret a nearly integer recovered n carefully rather than claiming exact equality to a rounded observation.
Reverse-solving by substitution instead of a shortcut
Some pupils remember p = 1 − v/m but cannot explain where it comes from. Ask them to derive it directly from the original pair: np = m and np(1 − p) = v. Replace np in the second equation with m, giving m(1 − p) = v.
Then divide by m, provided m is positive, to obtain 1 − p = v/m. This derivation is just two lines, but it reconnects the shortcut with the model.
A good tutor will teach the derivation before the shortcut becomes automatic. That way the pupil can reconstruct it during an unfamiliar question without relying on a formula they fear they have forgotten.
From recovered parameters to probabilities
Once the model is X ~ B(20,0.3), an exam question may ask for a specific event. Define the event in words first. To find no successes, use P(X = 0) = 0.7²⁰, approximately 0.000798. To find at least one success, use the complement 1 − 0.7²⁰, approximately 0.999202.
The result is large because twenty independent opportunities with 30% success chance make at least one success very likely under this mathematical model. These are hypothetical probabilities, not measured reliability figures for any real school or organisation.
If the question asks for exactly six successes, use P(X = 6) = 20C6 × 0.3⁶ × 0.7¹⁴. The combination counts which six trial positions are successes, and the powers describe the independent outcomes.
The important lesson is that the recovered n and p do not automatically determine the event. The problem’s wording still controls whether the calculation is exact, cumulative, at most or at least.
A word problem where p is unknown but n is known
Suppose twenty independent trials are made and the expected number of successes is six. Then np = 6 with n = 20, so p = 6/20 = 0.3. No variance information is necessary in this particular question.
But if the same twenty-trial problem gives a mean of six and a variance of seven, it is internally inconsistent with a binomial model. Do not assume that once n is supplied, the remaining information must automatically be compatible.
This distinction shows why students should recognise the minimum information needed for each calculation and independently check extra conditions when given.
What if only a variance is supplied?
A binomial variance alone generally does not identify both n and p. For example, v = 2.25 can arise from B(12,0.25), whose mean is 3, but many other combinations of n and p may produce the same variance.
To recover both parameters uniquely, the question needs another independent relationship, such as a mean, a known n, a known p or another stated probability condition that fixes the remaining uncertainty.
A student’s instinct should be to count the unknown parameters and independent pieces of information before attempting algebra. One equation with two unknowns is usually not enough.
The difference between probability p and probability P(X = r)
The trial-level success probability p is a parameter between zero and one. The count-event probability P(X = r) is the probability that exactly r successes occur in the whole set of n trials.
For X ~ B(20,0.3), p = 0.3 does not mean P(X = 6) is 0.3. The count-event probability is computed from the binomial distribution formula and the number of possible success positions.
This matters because a learner may recover p correctly and then mistake that parameter for the answer to a later multi-trial question. Write both symbols with their descriptions before calculating.
A condition test with no replacement
Suppose twenty components are selected from a small batch without replacement, and the batch contains a limited number of defects. The probability that a selected component is defective generally changes after each selection. Even if an approximate mean or variance matches binomial-looking values, those facts alone do not prove the selections are independent with a fixed p.
The standard binomial model requires the right trial mechanism. Our with-and-without-replacement guide explains how dependence changes sequential probabilities. A student should distinguish ‘the numbers fit two formulas’ from ‘the experiment actually fits the distribution’.
If a question explicitly states independent trials with constant success probability, use that supplied condition. If it asks whether a model is suitable, discuss the process rather than assuming it.
A worked example that uses a rounded mean
An organisation reports an estimated average success count of approximately 4.8 and an estimated variance of approximately 3.84 over repeated trials. The clean algebra suggests n = 24 and p = 0.2 if those figures are treated as exact theoretical values.
However, estimates from a finite real dataset are subject to sampling uncertainty, and rounded values need not correspond exactly to any single theoretical B(n,p). A student should report the algebra required by the school question and avoid inventing greater empirical certainty than the statements support.
This is especially important when an exam question is framed as ‘a binomial random variable has mean…’ rather than ‘a sample of observations suggests a mean…’. The former supplies theoretical model information; the latter describes evidence from which a model might be estimated.
Why the formula list is a starting point, not an answer
The official MF27 list of formulae and results includes binomial B(n,p) with mean np and variance np(1 − p). It is a shared booklet used for several A-Level Mathematics syllabuses, and not every formula printed in it is examinable in H1.
A pupil can find the formulas quickly but must still decide whether the event is binomial, what the parameters mean and whether the given values can be valid. The booklet helps with retrieval; it does not interpret the story or perform the reasoning.
For more detailed probability events and use of approved calculator functions, read our introductory binomial distribution tutorial and graphing calculator accuracy guide.
The seven-step reconstruction checklist
- State that X counts successes and that a binomial model is given or appropriately justified.
- Write n as a positive integer and p as a probability between zero and one.
- Translate the theoretical mean into m = np.
- Translate the theoretical variance into v = np(1 − p); if SD was given, square it first.
- For m > 0, calculate p = 1 − v/m, then n = m/p.
- Verify n and p meet the mathematical restrictions and reproduce both m and v.
- If the question asks for a subsequent event, define it independently before using the binomial probability formula or calculator.
A strong response needs no grand performance. It is a series of small, meaningful decisions. Many weak solutions fail at the fifth or sixth step by accepting a non-integer n or a negative p without checking.
A short 90-minute tutorial model
Begin with ten minutes reviewing how B(n,p) differs from a normal model. Ask the learner to identify the fixed trial count and success event from two contrasting stories. Avoid letting them begin with a formula if they do not know what is counted.
Use twenty minutes to derive the relationship from mean and variance and solve the 6-and-4.2 example. Change the values for another twenty minutes and require the student to reconstruct the parameters without prompts.
Use fifteen minutes for invalid data—variance greater than mean, non-integer n and zero-mean degeneracy. Then set two probability event questions that use the recovered parameters. Finish with an independent changed problem and a brief error log.
The sequence is illustrative and should follow the college’s topic progress. Its purpose is to move the learner from copying formulas toward understanding which models and values are possible.
Small-group teaching in Bukit Timah
The immutable eduKateSG three-pupil tutorial reference describes a 90-minute weekly small-group format near Sixth Avenue MRT at 8 Fourth Avenue, Singapore 268674. That page is a reference for the teaching model, not confirmation that a particular JC1 H1 Mathematics group is currently open.
Within a group of three, one learner might recover p but forget to check n, another might confuse standard deviation with variance, and a third might use the recovered p directly as a count-event probability. The tutor should give each student a distinct follow-up problem.
Families should confirm tutor subject experience, timetable, fees and current availability before travelling or enrolling. A marked worksheet showing the first wrong decision is more informative than a broad statement that the student ‘is weak in statistics’. See consultation details.
Five self-check problems
Practice one: six and 4.2
A binomial count has mean 6 and variance 4.2. Find n and p. Answer: p = 0.3, n = 20. Verify 20 × 0.3 × 0.7 = 4.2.
Practice two: three and 2.25
A binomial count has mean 3 and variance 2.25. Find n and p. Answer: p = 0.25, n = 12. The trial count is an integer.
Practice three: mean 5 and variance 7
Can these be the theoretical mean and variance of a non-degenerate binomial distribution? No. Variance exceeds the mean, forcing an invalid negative p.
Practice four: mean 5 and variance 3
The equations give p = 0.4 and n = 12.5. If the figures are exact and X is claimed to be a binomial count with integer n, no such model exists.
Practice five: mean 4.8 and variance 3.84
Answer: p = 0.2, n = 24. Verify using both np and np(1 − p).
An independent question that changes the conditions
A teacher may instead give a fixed n = 25 and a mean of five. Then p = 5/25 = 0.2 and the variance is 25 × 0.2 × 0.8 = 4. The structure is familiar, but the information supplied is different.
Ask the learner to explain why only one equation was needed to find p in this case, and why the variance is a consequence rather than the second equation needed for solving. This builds adaptability rather than formula memorisation.
Why mean and variance belong together
Expected value and variance answer two different questions about the same model. The first says where the count is centred across repeated sets of trials; the second says how spread out the possible counts are. Neither statistic tells you exactly what the next experiment will produce. When a pupil reconstructs a binomial model from both figures, they are combining information about centre and spread.
Take B(20,0.3). The expected count of six is not a guarantee of six successes next time. Some repetitions may have four or five successes, and others may have eight or more. The variance 4.2 describes the scale of this uncertainty. It does not mean there are always 4.2 extra successes or that an observed count can be non-integer.
A different binomial distribution can have the same mean but a different spread. For example, B(10,0.6) also has mean six, but its variance is 10(0.6)(0.4) = 2.4. B(20,0.3) has variance 4.2. The centre is identical, yet the outcomes are distributed differently. That is exactly why the mean alone cannot determine both parameters.
Now compare B(30,0.2). Its mean is again six, while its variance is 30(0.2)(0.8) = 4.8. Three binomial models, all with mean six, have distinct variances: 2.4, 4.2 and 4.8. A learner who can explain these differences has gone beyond repeating a formula; they understand how p affects the spread for a fixed expected count.
Our expected value, variance and independent totals guide develops the centre-versus-spread distinction for transformed and combined random variables. A good JC1 tutoring check is to give all three distributions above and ask: which one has the largest variance, and why should mean six not be treated as the only possible result?
Frequently asked questions
How do I find p if only binomial mean and variance are given?
For positive mean m and variance v, use p = 1 − v/m. Then calculate n = m/p and check the parameter restrictions.
Why must n be an integer?
Because n counts the fixed number of trials. A model with 12.5 trials is not a standard binomial distribution, even if the algebra produces that number.
Why can a binomial variance not exceed its mean?
Var(X) = E(X)(1 − p), and the factor 1 − p lies between zero and one. With positive mean, the variance cannot be larger.
What if the variance is zero?
The count may be deterministic with p equal to zero or one. If the mean is also zero, n cannot be identified uniquely from mean and variance alone.
Do I use standard deviation instead of variance if it is given?
Square the standard deviation first to obtain variance. Do not put the SD directly into np(1 − p).
Do matching formulas prove that real trial data are binomial?
No. The sampling mechanism must also satisfy fixed trials, two outcomes, independence and constant success probability.
Is solving backwards part of H1 Mathematics?
The 2027 H1 syllabus includes binomial modelling and the mean and variance of B(n,p), without proof. Solving unknown parameters from those relations is an application of the stated formulas.
Will a graphing calculator find n and p automatically?
It may solve equations or evaluate binomial probabilities, depending on the approved model, but the learner still must set up and validate the parameters and trial assumptions.
The big idea to remember
Do not let a neat algebraic answer override what n and p mean. A positive integer trial count and a valid probability are part of the model. When the given mean and variance do not support those conditions, recognising the incompatibility is the correct mathematical outcome.
Use the 2027 official H1 syllabus for the subject scope, the MF27 booklet for supplied formulas, and our Bukit Timah H1 Mathematics overview for the wider pathway. Confirm the official syllabus for the learner’s actual eventual examination year rather than assuming all cohorts sit the 2027 paper.