Did you know? A randomly shuffled list of a hundred names has, on average, exactly one name remaining in its original position. No independence assumption between positions is needed. The result follows from counting individual possibilities and adding their expectations rather than attempting to enumerate a hundred factorial arrangements.
For JC2 learners exploring a JC2 H3 Mathematics tutor in Bukit Timah, the principle behind that surprising fact is linearity of expectation. It connects H2 probability and counting to H3’s non-routine mathematical reasoning, proof and problem solving. Indicator variables are an excellent enrichment technique for making complicated counting questions manageable; they are not separately named additional-content items in syllabus 9820.
Quick answer: If X₁,…,Xₙ have finite expectations, then E[X₁+⋯+Xₙ]=E[X₁]+⋯+E[Xₙ] even if those variables are dependent. For an event A, its indicator I_A equals one when A occurs and zero otherwise, so E[I_A]=P(A). Define a count as a sum of indicators, then add its individual event probabilities.
One-minute diagnostic: six dice
Roll six fair six-sided dice. What is the expected number showing six? Define an indicator for each die: it is one if the die shows six and zero otherwise. Each indicator has expectation 1/6. Their sum counts how many sixes appear, so its expectation is six times 1/6, or one.
Now imagine the dice outcomes are dependent but every individual die still has probability 1/6 of showing six. The expected count remains one. What changes under dependence is the distribution and possibly its variance—not linearity of the expected sum.
What a strong answer must distinguish
Expected value is a weighted average of possible outcomes, not necessarily a likely outcome or a value that appears in any one trial. A count with expectation 1.2 can be an integer in every actual experiment. This is entirely consistent.
The technique requires identifying exactly what is counted. Indicator variables are most useful when the total outcome is the sum of yes-or-no events. The events may overlap, but overlapping does not prevent their indicators from being added.
Every application should specify the random experiment, the indicator event, its probability and the requested expected total. A striking numerical result without those definitions is not a complete mathematical explanation.
Why linearity works without independence
Expected value is a weighted average over a probability distribution. For discrete random variables X and Y with finite expectations, distribute the weighted sum of X+Y over the outcomes: E[X+Y]=E[X]+E[Y]. No step requires the events or variables to be independent.
For example, suppose X is the result of rolling a fair die and Y=7−X. The two random variables are completely dependent: knowing X determines Y. Yet E[X]=3.5 and E[Y]=3.5, so E[X+Y]=7, as it must because X+Y is always seven.
A common misconception is that dependence prevents adding expected values. Independence is important for certain product and variance formulas, but it is not a condition for linearity of expectation. State this distinction explicitly whenever the random quantities interact.
If X₁,…,Xₙ are indicator variables, their sum counts the number of corresponding events that occur. This remains true even when one event affects another. Consequently E[ΣIᵢ]=ΣP(Aᵢ) by linearity. The problem shifts from enumerating whole outcomes to calculating simple marginal probabilities.
For a Bernoulli event A, I_A is one if A occurs and zero otherwise. Its expectation is 1×P(A)+0×P(not A)=P(A). This is a short proof that should appear before using the technique on unfamiliar counting problems.
The conclusion has a precise meaning. It gives a mean across repeated trials of the stated random experiment, not a prediction that the next individual experiment will exhibit exactly that many successes.
Worked problem: fixed points in a random permutation
Suppose n distinct cards are arranged in a uniformly random permutation. A fixed point is a position holding the same card that originally occupied it. Let Iᵢ indicate that the card at position i is correct.
For a fixed position i, the chance its original card appears there is 1/n. Each original card is equally likely to occupy any of the n positions in a uniform random permutation. Therefore E[Iᵢ]=1/n.
Let X be the number of fixed points. Then X=I₁+⋯+Iₙ, so E[X]=n×(1/n)=1. The result is independent of n as long as n is a positive integer and all permutations are equally likely.
The indicator events are not independent. In a permutation of two elements, if one card is in the correct place then the other must be correct too. That dependence does not affect the expectation sum.
For n=2, the two permutations have zero or two fixed points, each with probability one-half, so the expectation is (0+2)/2=1. The value one need not occur in either individual outcome.
This demonstrates the best use of indicators: a global combinatorial distribution may be hard to list, but each local event has a simple probability. The proof applies to arbitrarily large n without computing n! terms.
Worked problem: sampling without replacement
A bag contains four red counters and six blue counters. Draw three different counters uniformly at random without replacement. Let X be the number of red counters in the sample. The draws are dependent, since removing a red counter changes the composition of the bag.
Define Iⱼ for j=1,2,3 to be one when the jth draw is red. By symmetry of a uniformly random ordered sample, each draw position has marginal probability 4/10 of being red, even though later conditional probabilities depend on earlier outcomes.
Thus X=I₁+I₂+I₃ and E[X]=3×(4/10)=1.2. The result is not an impossible claim that 1.2 counters appear in a sample; it is the average red count over many hypothetical three-counter samples.
Another representation is to mark each of the four red counters with an indicator of whether it is selected. The probability that any specified red counter is among three drawn from ten is 3/10. Summing four such indicators gives 4×(3/10)=1.2 again.
These are two distinct, valid ways to count the same random total. One approach uses indicators for sample positions; the other uses indicators for red objects. Their agreement illustrates how choosing the right representation simplifies probability calculations.
Occupancy: expected empty and occupied boxes
Throw eight distinguishable balls independently into four labelled boxes, each box equally likely to receive any ball. What is the expected number of empty boxes? Directly enumerating all 4⁸ allocations is possible but unnecessary.
For each box j, let Iⱼ indicate that it remains empty. One particular ball avoids that box with probability 3/4. Because individual ball placements are independent, all eight avoid it with probability (3/4)⁸. Therefore E[Iⱼ]=(3/4)⁸.
The expected number of empty boxes is 4(3/4)⁸. The expected number of occupied boxes is four minus this amount: 4[1−(3/4)⁸]. Both statements follow by adding indicators, but the calculation of each emptiness probability uses the independence assumption.
The word ‘independence’ appears at different levels of the proof. Independence of ball placements allows multiplication of the eight avoidance probabilities. Independence between the four box-empty indicators is not needed—and in fact those indicators interact.
Changing the model changes the formula. If every box must receive at least one ball, the random experiment is no longer unrestricted independent placement. The emptiness indicators are all zero and the expected number of occupied boxes is exactly four.
Expected collisions and pair indicators
Suppose n people independently have birthdays uniformly distributed among 365 days, ignoring leap days. Let X count pairs of people sharing a birthday. For each unordered pair i,j, define Iᵢⱼ to be one if that pair shares a birthday.
There are C(n,2) unordered pairs, and under the stated birthday model a particular pair shares a birthday with probability 1/365. Thus E[X]=C(n,2)/365. For n=23, this is 253/365, approximately 0.693.
This expected number of matching pairs is not the same as the probability of at least one birthday match. A single event can contain multiple matching pairs, so a random variable can exceed one. Its expectation need not equal an event probability.
The modelling assumptions matter. Real birthdays are not perfectly uniform or independent, and leap days have been ignored. The exact formula describes the simplified mathematical experiment, not necessarily a population dataset.
The point is not to memorise a particular birthday number. It is to recognise that a complicated collision count becomes a sum of simple pair indicators, which can be averaged without enumerating every possible calendar assignment.
Linearity is not a variance rule
The variance of a sum is not generally the sum of the individual variances. For random variables X and Y with finite second moments, Var(X+Y)=Var(X)+Var(Y)+2Cov(X,Y). The covariance term disappears under suitable independence or zero-covariance conditions.
For example, if Y=−X, then X+Y=0 has variance zero. Yet Var(X)+Var(Y)=2Var(X), generally positive. A student who blindly adds variances has treated dependent quantities as independent.
By contrast, expected values remain additive: E[X+(−X)]=E[X]−E[X]=0. The mathematics is consistent; the rules govern different measures of the distribution.
A tutor can use this contrast as a diagnostic. Ask whether a question seeks a mean, a variance or the probability of an event, since the applicable assumptions and formulas differ.
Proof-first probability communication
The first sentence of an indicator proof should name the experiment and the counted random variable. The next should define the yes-or-no events. Then show that the count is exactly their sum and calculate each event probability under the stated distribution.
If an event probability requires independence or a symmetry assumption, state it at the point of use. Do not say ‘linearity requires independence’; do not omit dependence when multiplying marginal probabilities.
Finish with the expected count and its interpretation. For discrete objects an expected count may be fractional, and that is not an error. The mean is a property of the distribution, not a guarantee about one sample.
The official H3 syllabus assumes H2 probability and counting and explicitly values combinatorial arguments and non-routine proof. Indicator variables are a useful enrichment method for applying these capabilities, not a prediction that a specific indicator-variables problem will appear.
Thirty-six original expectation and indicator clinics
1. One fair coin
Worked answer: The expected heads indicator of one fair toss is 1/2.
Proof check: It is an event probability expressed as the expectation of a zero-one variable. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
2. Ten fair tosses
Worked answer: The expected number of heads in ten fair tosses is 10×1/2=5.
Proof check: Linearity counts successes without listing 2¹⁰ complete outcomes. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
3. Dependent tosses
Worked answer: Suppose ten toss indicators are dependent but each has marginal heads probability 1/2.
Proof check: Their expected total is still five; independence is not required to add expectations. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
4. Six fair dice
Worked answer: The expected number showing face six is 6×(1/6)=1.
Proof check: A simple marginal probability can replace an enormous joint sample space. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
5. Wrong expectation interpretation
Worked answer: An expected count of one does not mean every experiment contains one success.
Proof check: The fixed-point permutation of two cards gives either zero or two, but averages one. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
6. Two-card permutation
Worked answer: A uniformly random two-card permutation has expected fixed points one.
Proof check: Each of two positions has fixed probability one-half. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
7. Ten-card permutation
Worked answer: A uniformly random ten-card permutation has expected fixed points one.
Proof check: Ten position indicators each contribute one-tenth. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
8. One-hundred-card permutation
Worked answer: A uniformly random permutation of one hundred cards still has expected fixed points one.
Proof check: The count of possible arrangements changes, but the sum of position probabilities remains one. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
9. Independence trap
Worked answer: Fixed-point indicators in a permutation are not independent.
Proof check: Knowing one position’s status changes information about other positions, yet linearity remains valid. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
10. Four red, six blue
Worked answer: Three draws without replacement have expected red count 3×4/10=1.2.
Proof check: Marginal symmetry handles dependence between draws. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
11. Mark red objects
Worked answer: Each of four red counters has probability three-tenths of selection in a size-three sample.
Proof check: Summing object indicators gives the same 1.2 expected red count. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
12. Five draws
Worked answer: Five draws from the same four-red, six-blue bag give expected red count two.
Proof check: Multiply five by 4/10 even though the successive draws are dependent. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
13. All red selection impossible
Worked answer: Drawing five objects cannot produce five red if the bag holds only four.
Proof check: An expected value calculation does not establish that every count is possible. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
14. Four boxes, eight balls
Worked answer: The expected count in one specified box is 8×1/4=2.
Proof check: Object indicators count how many of eight independent placements choose that box. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
15. Empty box
Worked answer: The probability a specified box stays empty after eight independent uniform placements into four boxes is (3/4)⁸.
Proof check: Independence matters when multiplying the eight avoidance probabilities. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
16. Expected empty boxes
Worked answer: With four boxes, expected empty count is 4(3/4)⁸.
Proof check: The different box-empty indicators need not be independent to sum their expectations. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
17. Expected occupied boxes
Worked answer: Expected occupied count is 4[1−(3/4)⁸].
Proof check: Occupied and empty indicators complement one another for each box. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
18. Five coupon types
Worked answer: After ten independent uniform draws from five types, expected distinct types is 5[1−(4/5)¹⁰].
Proof check: Count each type’s appearance indicator rather than every possible coupon sequence. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
19. Birthday pair
Worked answer: Under independent uniform 365-day birthdays, a specified pair matches with probability 1/365.
Proof check: The event model must state its uniform and independence assumptions. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
20. Birthday count
Worked answer: Among twenty-three people, expected matching pairs is C(23,2)/365=253/365.
Proof check: Count unordered pairs, not ordered pairs, to avoid doubling. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
21. Birthday probability
Worked answer: Expected collision pairs 253/365 does not equal the probability of at least one matching pair.
Proof check: Multiple pairs can match in one outcome, so the count is not a zero-one indicator. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
22. Five tosses adjacent equals
Worked answer: For five independent fair coin tosses, the expected number of adjacent equal pairs is 4×1/2=2.
Proof check: There are four adjacent comparisons; each equality event has probability one-half. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
23. Adjacent dependence
Worked answer: Adjacent-equality indicators overlap through shared tosses.
Proof check: Linearity applies even if the indicators interact, provided each marginal probability is correct. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
24. Random graph
Worked answer: With n vertices and each possible edge present with probability p, expected edges is C(n,2)p.
Proof check: Count unordered vertex pairs and their edge indicators. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
25. Ten-vertex graph
Worked answer: With n=10 and edge probability 0.2, expected edges is C(10,2)×0.2=9.
Proof check: An expected edge count may be an integer without being the actual number in every graph. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
26. Biased coin
Worked answer: If a coin has heads probability 0.3, ten tosses have expected heads count three.
Proof check: Each indicator has expectation 0.3; the assumption that probabilities stay fixed matters. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
27. Different probabilities
Worked answer: If three events have probabilities 0.2, 0.5 and 0.9, expected number occurring is 1.6.
Proof check: Add marginal probabilities even when the events overlap. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
28. Zero-one bound
Worked answer: For any event A, its indicator has expectation P(A) between zero and one.
Proof check: The range follows directly from the probability axioms. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
29. Counting disjoint events
Worked answer: If A and B are mutually exclusive, I_A+I_B indicates A or B.
Proof check: The indicators cannot both be one, so their sum remains zero or one. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
30. Counting overlapping events
Worked answer: If A and B overlap, I_A+I_B can equal two.
Proof check: The sum counts how many listed events occur, not merely whether any event occurs. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
31. Indicator union
Worked answer: For arbitrary A and B, I_(A union B)=I_A+I_B−I_(A intersection B).
Proof check: Taking expectations recovers the familiar inclusion–exclusion probability identity. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
32. Variance of opposites
Worked answer: If Y=−X, then X+Y=0 has variance zero.
Proof check: Variance does not generally add without considering covariance. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
33. Mean of opposites
Worked answer: E[X−X]=0, regardless of the distribution of X when its expectation is finite.
Proof check: Linearity handles dependent copies of the same variable. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
34. Probability versus expected count
Worked answer: A probability cannot exceed one, but an expected count may exceed one.
Proof check: Do not treat an arbitrary count of successes as a single indicator. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
35. Limited sample
Worked answer: A single experiment yielding zero success does not refute a positive expected value.
Proof check: Expected value describes the distribution across all outcomes. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
36. Model assumption
Worked answer: Nonuniform birthday probabilities change each pair’s matching probability.
Proof check: A correct symbolic method still depends on a credible probability model. Define the counted event precisely and explain which probability assumptions enter before applying linearity of expectation.
A seven-session learning plan for H3-level probability
Session one: Begin with the expectation of one indicator. Ask the learner to derive E[I_A]=P(A), then express the number of successes in a short set of events as a sum of such indicators. This shifts attention from a huge outcome table to simpler event probabilities.
Session two: Compare expected counts under independent and dependent random variables. Use the example X+Y=7 on paired die values to prove that dependence does not prevent the expectations from adding. Then contrast this with the more restrictive rules governing products and variances.
Session three: Solve fixed points in a uniformly random permutation of three, ten and a hundred items. The student should be able to explain why each position is correct with probability 1/n and why the expected number remains one.
Session four: Use sampling without replacement. Define indicators by draw position and by marked red objects, then show both approaches give the same expected count. This highlights how a good representation can simplify dependent sampling.
Session five: Model occupancy of labelled boxes. Explain the probability a specified box is empty, why multiplying avoidance probabilities requires independent placements, and why adding different box indicators does not require their independence.
Session six: Introduce counting collisions using pairs. Separate the expected number of matching pairs from the probability of at least one match. Ask the student to specify whether pairs are ordered or unordered before counting.
Session seven: Solve an unfamiliar expected-count question and require a complete write-up: random experiment, indicator definition, sum identity, marginal probability, expected total and interpretation. The tutor should correct the first unsupported independence claim, not just the final arithmetic.
A parent can assess progress by asking what the random experiment is, what counts as a success and why a fractional expectation is meaningful. These questions avoid the need to solve the entire probability problem while encouraging the student to explain the mathematics.
The immutable eduKateSG three-student Mathematics reference near Sixth Avenue MRT illustrates a Secondary 1 teaching format and does not confirm an H3 class. Enquire directly about the current JC2 H3 teaching scope, timetable and availability.
Common errors in expectation proofs
The most frequent error is assuming linearity requires independence. It does not. The expectation of a sum equals the sum of expectations whenever the relevant expectations exist. Independence is needed in some probability-product calculations, not in the linearity identity itself.
A second error is using one count when the question asks for a probability. For birthdays, the number of matching pairs can exceed one in a single outcome. Its expectation must not be mistaken for the probability that at least one match occurs.
Third, students may use incorrect marginal event probabilities. In a model where placements are not uniform, the probability that a particular box remains empty is not automatically (3/4)⁸. State the distribution and its assumptions before calculating.
Fourth, a probability calculation may count ordered pairs when the object is an unordered pair. If n people have birthdays, there are C(n,2) possible pairs. Counting n(n−1) would count each unordered pair twice.
Fifth, a student may claim an expected count must be a possible observed count. A sample of red counters contains whole counters, but a distribution can have expectation 1.2. That is a weighted average, not an impossible physical measurement.
Sixth, adding variances as though they were expectations can be wrong. When the variables share outcomes, covariance can contribute. Identify the requested statistical quantity before choosing an algebraic rule.
Frequently asked questions
Is linearity of expectation explicitly named in the H3 9820 additional syllabus?
The official 2027 syllabus assumes H2 probability and counting, and explicitly lists combinatorial arguments and proofs. The indicator-variable problems here are a useful extension that develops those capabilities, rather than a separately named additional-content chapter or a prediction of a future examination question.
Does linearity require independent events?
No. If X and Y have finite expectations, E[X+Y]=E[X]+E[Y] even when the two variables are dependent. Independence might simplify a particular event probability or a variance, but it is not required for the addition of expectations.
What is an indicator variable?
An indicator variable is one when a specified event occurs and zero when it does not. Its expectation equals the probability of that event. A count of successes can often be represented as a sum of indicators, making its expected value straightforward.
Why is the expected number of fixed points in a random permutation one?
In a uniform permutation of n distinct objects, each original position is occupied correctly with probability 1/n. Add these n probabilities using indicator variables. The result is n×(1/n)=1 for every positive integer n.
Does an expected count of one mean one success is likely?
Not necessarily. A two-card permutation has either zero or two fixed points, each with probability one-half, yet its mean number of fixed points is one. An expectation can even be a value that never appears as an actual count.
Why does sampling without replacement still allow an easy expectation?
Although successive draws are dependent, a given draw position has a well-defined marginal probability of being red. In the four-red, six-blue example, each of three positions has marginal probability 4/10, giving expected total red count 1.2.
What independence is required in the empty-box calculation?
The placements of different balls must be independent to multiply their separate avoidance probabilities as (3/4)⁸. The indicators for whether the different boxes are empty do not need to be independent for their expected values to add.
Is the expected number of birthday collisions a collision probability?
No. The expected number of matching pairs counts every unordered pair that shares a birthday, potentially more than one pair in one outcome. The probability of at least one matching pair is the expectation of a different zero-one indicator.
Can the expected number of objects be fractional?
Yes. Counts are integers in individual experiments, but their expectation is a weighted average across outcomes. It may be fractional, as with an expected red-counter count of 1.2 from a three-item sample.
When do variances add?
Variance of a sum includes covariance terms. If the variables are independent, those covariance terms vanish and variances add. Without that condition, blindly copying the expectation addition rule into variance calculations can produce wrong answers.
What makes a convincing H3-style indicator proof?
Define the experiment and random count, express the count as a sum of precisely defined indicators, justify every marginal probability and then apply linearity. Finish by interpreting what the expected value represents under the stated distribution.
Official references and further study
The 2027 Singapore–Cambridge H3 Mathematics (9820) syllabus documents the assumed H2 probability content, combinatorial reasoning and formal proof expectations. For adjacent skills, read conditional probability and counting arguments, inclusion–exclusion, and double counting and combinatorial identities.
The practical takeaway is that the right indicator representation turns a difficult global count into a sum of understandable local probabilities. A strong mathematical solution explains those local events and assumptions before writing the final expectation.
A final one-minute indicator audit
For a new problem, identify the objects being counted and assign one indicator to each candidate object. Then check whether summing those indicators reproduces exactly the requested random count. Compute each marginal probability from the experiment, not from an assumed independence rule that the question never supplied. Add those probabilities, label the expected quantity and interpret its value. If the result seems surprising, test one small version of the experiment to identify whether you confused a count, a probability or a conditional distribution. A reliable proof explains the relationship between the original random process and the indicator representation.