Bukit Timah Tutor Mathematics

A connected Mathematics learning system from school foundations to examinations, applications and advanced study. Use the Mathematics Hub to move between levels, concepts, diagnosis, examinations, applications and world routes.

H2 Mathematics 9758 | Probability and Conditional Probability

Three students in school uniforms work through open books at a classroom table, with textbooks and stationery nearby and study notes on the whiteboard behind them.

H2 Mathematics Probability is a canonical JC topic guide for Singapore-Cambridge H2 Mathematics 9758, aligned to the 2027 SEAB syllabus. It is designed to sit between the stable mathematical object in BTT’s Knowledge Warehouse and the actual H2 examination demand.

Probability begins with a sample space and a model. H2 students lose marks when they apply formulas before deciding whether events are mutually exclusive, independent or conditional.

The 2027 syllabus includes probability addition/multiplication, mutually exclusive and independent events, tables, Venn diagrams, trees, permutation/combination techniques and simple conditional probability.

Mutually exclusive is not independent

Mutually exclusive events cannot occur together, so P(A∩B)=0. Independent events can occur together, but learning that one happened does not change the probability of the other, so P(A∩B)=P(A)P(B). Non-zero mutually exclusive events are therefore not independent.

Conditional probability

P(A|B)=P(A∩B)/P(B) restricts the universe to outcomes in B. A tree diagram often makes this conditional structure visible: branch probabilities after information has changed must reflect the new state.

Choose the representation

  • Venn diagram: overlapping categories.
  • Tree: sequential/conditional events.
  • Outcome table: two-variable finite sample spaces.
  • Counting: equally likely combinatorial outcomes.

Common traps

  • Adding probabilities for events that overlap.
  • Multiplying probabilities without independence/conditional justification.
  • Using unconditional probabilities after sampling without replacement.
  • Confusing P(A|B) with P(B|A).
  • Counting outcomes as equally likely when they are not.

H2 examination control

  1. State the mathematical model or condition before calculating.
  2. Keep exact values until the question requires approximation.
  3. Show enough working to expose the method; unsupported incorrect answers earn no marks.
  4. Use the approved graphing calculator as a tool, not as a substitute for interpretation.
  5. Re-read the answer in the context of the original question.

Return to JC Mathematics. Stable conceptual owner: Mathematics Knowledge Warehouse. Examination execution: Mathematics Examination Craft.


Syllabus check: SEAB H2 Mathematics 9758 for examination in 2027; checked 26 September 2026.

World Mathematics route: return to the World Mathematics Atlas to connect this JC topic with its prerequisites, international equivalents, examination routes and university Mathematics.

H2 Mathematics 9758 probability: build events before formulas

Define the sample space

Probability statements refer to events inside a model of possible outcomes. Clarify what one trial or outcome is before manipulating symbols.

Mutually exclusive is not independent

Mutually exclusive events cannot occur together. Independent events can occur together, but knowledge of one does not change the probability of the other under the model. Confusing these definitions causes systematic errors.

Conditional probability changes the reference set

P(A|B) asks for the probability of A given that B is known to have occurred. Trees, tables and Venn diagrams can make the restricted sample space visible.

Multiplication rules follow event structure

For sequential events, multiply conditional probabilities along a path. Independence is the special case where the conditional probability remains the original probability.

Addition requires overlap control

For A or B, account for the intersection unless the events are mutually exclusive. A Venn diagram can prevent double counting.

Bayesian reversal needs careful conditioning

When evidence updates the probability of a source or cause, organise prior probabilities and conditional evidence before reversing the condition.

Counting and probability interact

If outcomes are equally likely, combinatorial counting can determine event probabilities. Verify that the equally-likely assumption is justified.

Common errors

Reversing P(A|B), treating ‘or’ as always simple addition, assuming independence without evidence and failing to update denominators in without-replacement processes.

Reasonableness checks

Probabilities lie between zero and one; adding information can change a conditional probability; complementary events should sum to one. Use these invariants to catch calculator or setup errors.

Transfer

Mix trees, Venn diagrams, counting and verbal scenarios so the learner must construct the event model rather than identify a formula from a keyword.

Build the Sample Space Before Choosing a Probability Formula

Probability errors often begin before the first calculation. The learner sees a phrase such as “at least one”, “without replacement”, “given that”, “arranged”, or “chosen” and reaches immediately for a familiar formula. Sometimes the formula is appropriate. Sometimes it silently assumes the wrong sample space.

The durable probability habit is the reverse:

Define the experiment → define one outcome → decide whether outcomes are equally likely → decide whether order matters → decide whether replacement or dependence changes later stages → only then calculate.

1. What is the experiment?

A probability model begins with an experiment or random process. “Choose two cards” is incomplete. Are the cards chosen simultaneously or sequentially? With replacement or without replacement? Are order and identity recorded? Are cards equally likely to be selected? The wording determines the sample space.

2. What counts as one outcome?

Roll a fair die twice. If order matters, one outcome is an ordered pair such as (2,5). The sample space contains 36 equally likely ordered outcomes.

If only the sum is recorded, the possible recorded sums are 2 through 12, but those sums are not equally likely. Treating the eleven sums as eleven equally likely outcomes gives the wrong model.

3. Equally likely labels are not guaranteed

A spinner with three labelled regions A, B and C does not imply probability 1/3 for each label unless the regions have equal probability weight. Geometry, construction or stated probabilities determine the model.

4. One die, two different sample spaces

Experiment A: roll one fair six-sided die and record the face. Sample space: {1,2,3,4,5,6}, equally likely.

Experiment B: roll two fair dice and record the sum. Recorded values: {2,3,…,12}, not equally likely.

For sum 7 there are six ordered pairs: (1,6),(2,5),(3,4),(4,3),(5,2),(6,1). For sum 2 there is only one: (1,1). The recorded number alone hides multiplicity.

5. Build before dividing

The familiar probability fraction:

P(event) = favourable outcomes / total outcomes

works directly when the counted elementary outcomes are equally likely. If the outcomes are not equally likely, simple counting is insufficient.

6. Order matters when positions carry meaning

Choose two different letters from A, B and C and place them in first and second position.

Ordered outcomes:

AB, AC, BA, BC, CA, CB.

There are six.

If the task instead asks for an unordered two-letter group, AB and BA represent the same selection. The possible selections are AB, AC and BC: three outcomes.

The words “choose” and “arrange” often signal different sample spaces.

7. Replacement changes the second stage

A bag contains 3 red counters and 2 blue counters.

Choose one counter, replace it, then choose again.

Because the first counter is returned, the composition of the bag before the second draw is unchanged. The probability of red on the second draw remains 3/5 regardless of the first result.

Without replacement, the second-stage probabilities depend on what happened first.

8. Without replacement creates conditional structure

Same bag: 3 red, 2 blue. Draw twice without replacement.

P(first red) = 3/5.

If the first was red, the bag then contains 2 red and 2 blue, so:

P(second red | first red) = 2/4 = 1/2.

Therefore:

P(two reds) = (3/5)(1/2) = 3/10.

The second probability is conditional because the population changed.

9. Independence is a relationship, not a feeling

Events A and B are independent when learning that one occurred does not change the probability of the other:

P(A | B) = P(A)

when the conditional probability is defined.

Equivalently:

P(A ∩ B) = P(A)P(B).

Do not infer independence merely because two events “sound unrelated”. Verify the model.

10. Mutually exclusive is not independent

Two non-zero-probability events that cannot happen together are mutually exclusive. If A and B are mutually exclusive, then:

P(A ∩ B) = 0.

If both have positive probability, independence would require:

P(A ∩ B) = P(A)P(B) > 0.

Therefore non-zero mutually exclusive events are not independent.

11. A tree diagram is a sample-space organiser

A tree diagram is not merely presentation. It makes conditional branch probabilities visible. Each path represents one sequential outcome. Multiplying along a path gives that path probability; adding appropriate mutually exclusive paths gives an event probability.

12. Tree example: two draws without replacement

Bag: 2 green, 1 yellow. Draw twice without replacement.

First green: probability 2/3.

Then green: 1/2; yellow: 1/2.

First yellow: probability 1/3.

Then green: 2/2 = 1.

Thus:

P(GG) = (2/3)(1/2) = 1/3.

P(GY) = 1/3.

P(YG) = 1/3.

The three possible colour sequences have equal probability here, but that equality comes from the branch products, not from simply counting labels.

13. “At least one” often invites a complement

Roll a fair die four times. Find the probability of at least one six.

Direct counting is possible but cumbersome. The complement event is no six on all four rolls:

P(no six) = (5/6)^4.

Therefore:

P(at least one six) = 1 – (5/6)^4.

The method choice follows from event structure.

14. Complement works only when it is exact

The complement of “at least one success” is “no successes”. The complement of “more than two” is “at most two”. The complement must cover every remaining possibility exactly once.

15. “Exactly one” is not the same as “at least one”

Two fair coin tosses:

Sample space: HH, HT, TH, TT.

Exactly one head: HT, TH → probability 2/4 = 1/2.

At least one head: HH, HT, TH → probability 3/4.

Words define events.

16. Conditional probability changes the denominator

Suppose a fair die is rolled and you are told the result is even. What is the probability the result is greater than 3?

Original sample space: {1,2,3,4,5,6}.

Given even, the conditional sample space becomes {2,4,6}.

Within it, values greater than 3 are {4,6}. Therefore:

P(X>3 | X even) = 2/3.

The information “given even” changes which outcomes remain possible.

17. Conditional probability is not reverse conditioning

P(A | B) and P(B | A) generally differ.

“Probability a student studied given that they passed” is not the same as “probability a student passed given that they studied”. The denominators represent different conditioned groups.

18. Two-way table example

Suppose 100 students are classified:

PassedDid not passTotal
Studied54660
Did not study202040
Total7426100

P(Passed | Studied) = 54/60 = 0.9.

P(Studied | Passed) = 54/74.

The same intersection count 54 appears, but the conditioning changes the denominator.

19. Sample-space first prevents Bayes confusion

Bayes-type problems become easier when the populations and branches are drawn before formula substitution. The formula is a compressed version of a sample-space rearrangement.

20. Counting model: permutations

If 5 different books are arranged in a row, there are:

5! = 120

orders.

Every position matters. Swapping two books creates a different arrangement.

21. Counting model: combinations

If 3 books are chosen from 5 and order does not matter, the number of selections is:

C(5,3) = 10.

A combination is appropriate because only membership in the selected set matters.

22. Formula choice follows outcome identity

Do not ask “Is this a permutation question?” Ask: “If I swap two selected objects, do I get a genuinely different outcome?” If yes, order matters. If no, combination structure may be appropriate.

23. Repetition changes counting

How many 4-digit PINs can be formed using digits 0–9 if repetition is allowed?

Each position has 10 choices:

10^4 = 10,000.

If repetition is forbidden, choices become 10 × 9 × 8 × 7.

The sample-space rule changes before any probability is computed.

24. Leading-zero rules matter

A four-digit PIN may begin with zero. A four-digit positive integer may not. Same symbols, different object definition, different count.

25. Circular arrangements change equivalence

When arranging distinct people around a round table, rotations may represent the same relative seating. Fix one person and arrange the remaining n−1 people, giving (n−1)! under the usual circular-equivalence model.

Again, the formula depends on what counts as a distinct outcome.

26. Identical objects change counting

Arrange letters of LEVEL.

There are 5 letters with L repeated twice and E repeated twice. Distinct arrangements:

5!/(2!2!) = 30.

Treating identical copies as distinguishable overcounts.

27. Restricted arrangements need structure

Suppose A and B must sit together among 6 people in a row. Treat AB as one block initially. Then there are 5! block arrangements and 2 internal orders for A and B, giving 2×5!.

This works because the restriction creates a natural compound object.

28. “Not together” often invites complement counting

Total arrangements minus arrangements with the restricted pair together can be easier than counting all allowed cases directly.

29. Inclusion-exclusion begins with overlap

If event A and event B can overlap:

P(A ∪ B) = P(A) + P(B) − P(A ∩ B).

The subtraction repairs double counting.

30. Do not add probabilities of overlapping events blindly

On a die, A = even = {2,4,6}; B = greater than 3 = {4,5,6}.

P(A)=3/6, P(B)=3/6.

Adding gives 1, but A and B overlap at {4,6}.

Correct union probability:

3/6 + 3/6 − 2/6 = 4/6 = 2/3.

31. Equal probability requires equal weighting of elementary outcomes

When rolling two fair dice, ordered pairs are equally likely. Sums are not. When drawing a card uniformly from a standard deck, individual cards are equally likely; ranks are not equally represented if the model conditions on another property without adjustment.

32. The unit-of-randomness question

Ask what is selected uniformly: a person, household, ticket, day, transaction, card, arrangement, integer, interval point? Different sampling units produce different probability models.

33. Geometric probability uses measure, not counting alone

If a point is chosen uniformly from an interval, there are infinitely many possible points. Probability can be a length ratio. In two dimensions, it may be an area ratio. The “favourable over total” idea survives, but count is replaced by measure.

34. Example: uniform point on [0,10]

Choose X uniformly from [0,10].

P(3 ≤ X ≤ 7) = length of [3,7] / length of [0,10] = 4/10 = 0.4.

35. Continuous exact points can have probability zero

For a continuous uniform model, P(X=5)=0 even though 5 is a possible value. Probability zero does not necessarily mean logical impossibility in a continuous model; it means zero measure.

36. Model before formula in continuous probability too

A density must integrate to 1 over its domain. Before integrating an event, confirm the support and density definition.

37. Conditional sampling without replacement

A box contains 4 white and 6 black balls. Two are drawn without replacement.

P(second white) can be found by branches:

P(W then W) + P(B then W)

= (4/10)(3/9) + (6/10)(4/9)

= 12/90 + 24/90 = 36/90 = 0.4.

Interestingly, this equals the original white proportion 4/10. The branch calculation proves it for this case.

38. Symmetry can simplify probability

In many random-order problems, symmetry can replace long counting if each object is equally likely to occupy each position. But symmetry must be justified by the random mechanism.

39. “Random” does not automatically mean uniform

A random variable can follow any specified distribution. A random selection procedure may favour some outcomes. Always distinguish randomness from equal likelihood.

40. Expected value is a weighted average over the model

For discrete X:

E(X) = Σ x P(X=x).

The weights are probabilities. Expected value is not necessarily an attainable outcome. A fair die has expected value 3.5 although no face shows 3.5.

41. Expected value depends on the sample-space model

If outcome probabilities are misassigned, the expectation is wrong even if the arithmetic is flawless.

42. Rare-event intuition can fail

Events with small individual probabilities can become likely across many opportunities. “At least one” over repeated trials often needs complement reasoning.

43. Example: at least one success in n independent trials

If success probability is p each independent trial, then:

P(at least one success) = 1 − (1−p)^n.

This formula is not a starting point. It is a consequence of the complement event and independence.

44. Dependence can destroy repeated-trial formulas

If probabilities change after each trial, (1−p)^n may no longer describe “no successes”. Without replacement is a common cause.

45. Sampling with replacement versus without replacement

A deck contains 4 aces among 52 cards.

Two draws with replacement:

P(no ace) = (48/52)^2.

Two draws without replacement:

P(no ace) = (48/52)(47/51).

The words “with replacement” change the model, not merely one number.

46. Independence can be tested numerically

If P(A)=0.4, P(B)=0.5 and P(A∩B)=0.2, then:

P(A)P(B)=0.2.

The equality supports independence within the stated model.

47. Pairwise independence is not always full independence

At advanced levels, several events can be pairwise independent without being jointly independent. This is another reason to state the exact independence condition required rather than using the word loosely.

48. Venn diagrams show set relationships

Venn diagrams are useful when events are defined by overlapping properties. Trees are useful for sequential processes. Tables are useful for two categorical variables. Choose the representation that exposes the event structure.

49. The wrong representation can create wrong assumptions

A tree for a non-sequential problem may be cumbersome but still possible. A Venn diagram for a changing-without-replacement process can hide conditional stages. Representation choice is part of modelling.

50. Sample-space casebook: coin tosses

Three fair coin tosses produce eight equally likely ordered sequences:

HHH, HHT, HTH, HTT, THH, THT, TTH, TTT.

Exactly two heads occurs in HHT, HTH, THH → 3/8.

At least two heads occurs in those three plus HHH → 4/8 = 1/2.

51. Sample-space casebook: dice

Two dice, sum 8. Favourable ordered pairs:

(2,6),(3,5),(4,4),(5,3),(6,2).

Probability = 5/36.

Do not write 1/11 because 8 is one of eleven possible sums.

52. Sample-space casebook: cards

One card from a standard 52-card deck. Probability of a heart or a king:

13/52 + 4/52 − 1/52 = 16/52 = 4/13.

The king of hearts is in both categories and must not be double-counted.

53. Sample-space casebook: committee

Choose a committee of 3 from 8 people. Order does not matter:

C(8,3)=56.

If one specific person must be included, choose the remaining 2 from the other 7:

C(7,2)=21.

54. Sample-space casebook: positions

Choose president, secretary and treasurer from 8 people. Roles differ, so order/position matters:

8×7×6=336.

Same people, different outcome definition, different count.

55. Sample-space casebook: repeated digits

Form a 3-digit code from digits 0–9 with repetition allowed. There are 10^3 = 1000 codes.

Form a 3-digit positive integer with no repeated digits. First digit: 9 choices (1–9). Second: 9 remaining choices including 0. Third: 8. Total 648.

56. Sample-space casebook: conditional table

If 30 of 50 members are adults and 18 adults attend an event, while 10 of 20 non-adults attend, then:

P(attend | adult)=18/30=0.6.

P(adult | attend)=18/(18+10)=18/28.

Conditioning selects a denominator.

57. Sample-space casebook: medical-style base rates

Suppose a synthetic screening model has 1% prevalence, 90% sensitivity and 95% specificity in a population of 10,000.

Expected affected: 100. True positives: 90.

Unaffected: 9,900. False positives at 5%: 495.

Among positive results, 90 of 585 are true positives.

This illustrates why reverse conditional probability depends strongly on base rates. It is a mathematical example, not medical advice or a claim about any real test.

58. Sampling bias is not repaired by probability arithmetic

If a “random sample” is actually taken from a non-representative frame, correct formulas can still produce misleading population conclusions. Probability models depend on how the sample was generated.

59. Simulation can test a model, not replace its definition

A computer can simulate coin tosses or card draws. Before running the simulation, the learner should state the intended random mechanism. Otherwise the code may faithfully simulate the wrong process.

60. Simulation frequency approaches model probability with variability

In repeated simulations, observed proportions fluctuate. Larger runs often stabilise around the model probability, but one finite run does not define the theoretical probability.

61. Model validation through small cases

For a complicated counting formula, test a smaller version where all outcomes can be listed. This can reveal double counting or missing cases.

62. Small-case verification example

Suppose a formula claims the number of 2-person committees from n people is n(n−1). Test n=3. Direct listing gives AB, AC, BC: 3 committees. Formula n(n−1)=6 overcounts by a factor of 2 because order was counted. Correct formula C(n,2)=n(n−1)/2.

63. A probability formula is compressed reasoning

Permutation, combination, conditional probability, complement and binomial formulas are not isolated buttons. Each compresses a set of assumptions about outcomes and relationships.

64. Binomial model assumptions

A binomial model requires a fixed number of trials, two outcome categories for the model, constant success probability and independence between trials under the model.

If sampling without replacement from a small finite population changes success probability materially, a binomial model may not be appropriate.

65. Binomial example

Five independent items each have defect probability 0.1. Probability of exactly two defects:

C(5,2)(0.1)^2(0.9)^3.

The combination chooses which two trials are defects; the powers assign probabilities to the selected pattern; independence allows multiplication.

66. Hypergeometric structure without naming it

If five items are drawn without replacement from a small batch containing a fixed number of defective items, the probabilities change as items are removed. The model is different from independent constant-p trials.

67. Conditional probability can reveal dependence

If P(A|B) differs from P(A), then B changes the probability of A, so the events are not independent.

68. Independence does not imply no relationship in language

In probability, “independent” has a technical definition. Everyday notions of unrelated causes or physical separation do not define it.

69. Random variables compress outcomes numerically

A random variable assigns a number to each elementary outcome. Different elementary outcomes can map to the same random-variable value. This is why values of a random variable are not automatically equally likely even when elementary outcomes are.

70. Die-sum example revisited

The random variable S = sum of two fair dice maps 36 equally likely ordered pairs to only 11 possible sums. Multiplicities differ, so P(S=s) differs across s.

71. Probability distributions are weighted sample spaces

A distribution records the probability attached to each random-variable value. Before using expected value or variance, confirm the probabilities sum to 1 and match the intended model.

72. Probability language audit

Translate key words explicitly:

  • and → intersection;
  • or → union, usually inclusive unless context says otherwise;
  • given → conditional probability;
  • at least → lower bound including equality;
  • at most → upper bound including equality;
  • exactly → one precise count/value;
  • none → complement of at least one in repeated-trial contexts;
  • without replacement → changing composition;
  • independent → multiplication rule with unchanged conditional probability.

73. “Or” can be misunderstood

In mathematics, A or B usually includes outcomes where both occur. If the events can overlap, subtract the intersection when using the addition rule.

74. “Either…or” may be exclusive in ordinary language

Read context carefully. Formal probability notation removes this ambiguity.

75. Ordered versus unordered pairs

If two students are selected for different roles, (A,B) and (B,A) differ. If they are selected simply as a pair, they do not. State the outcome object in words before counting.

76. Replacement versus independence

Replacement often creates independence in simple uniform drawing models because composition resets. But independence is the mathematical property; replacement is one mechanism that can produce it. Do not treat the words as definitions of each other.

77. Dependence can arise without physical removal

Events can be dependent because of shared information, selection conditions or common constraints even if nothing is literally removed from a container.

78. Probability model versus real-world frequency

A theoretical model may assume a fair coin or equally likely outcomes. Real devices can deviate. The model answers a mathematical question under stated assumptions; empirical validation is a separate task.

79. Model assumptions should be visible in the answer

Useful phrases:

  • Assuming the die is fair…
  • Because the draws are without replacement…
  • Given that the first card is an ace…
  • Treating ordered pairs as equally likely…
  • Since order does not matter…

This language exposes the reasoning that justifies the formula.

80. Fresh mixed case 1

A bag has 5 red and 3 blue counters. Draw two without replacement. Find P(one of each colour).

Ordered routes:

RB: (5/8)(3/7)

BR: (3/8)(5/7)

Add:

30/56 = 15/28.

81. Fresh mixed case 2

Same bag, but replace the first counter before the second draw. P(one of each):

(5/8)(3/8) + (3/8)(5/8) = 30/64 = 15/32.

Same colours, different replacement rule, different answer.

82. Fresh mixed case 3

Six students compete for gold, silver and bronze. Number of podium outcomes:

6×5×4=120.

Order matters because medals differ.

83. Fresh mixed case 4

Choose 3 students from 6 for a committee:

C(6,3)=20.

Same group size, no roles, order irrelevant.

84. Fresh mixed case 5

Two fair dice. Find probability the sum is at least 10.

Favourable ordered pairs:

sum 10: 3; sum 11: 2; sum 12: 1. Total 6.

Probability = 6/36 = 1/6.

85. Fresh mixed case 6

Given that the sum of two fair dice is at least 10, find probability the sum is 12.

Conditional sample space has 6 equally likely ordered pairs corresponding to sums 10,11,12. Only (6,6) gives 12. Probability = 1/6.

86. Fresh mixed case 7

Three independent fair coin tosses. Find probability of at least one tail.

Complement: all heads = 1/8. Therefore at least one tail = 7/8.

87. Fresh mixed case 8

A 4-character code uses A, B, C, D with repetition allowed. There are 4^4 = 256 codes.

If repetition is forbidden, there are 4! = 24.

88. Fresh mixed case 9

A class contains 12 students, 7 of whom take Mathematics Club. Choose one student uniformly. P(Math Club)=7/12.

Choose a club uniformly from two clubs and then a student uniformly within that club: this is a different sampling mechanism and does not generally produce the same individual probabilities.

89. Fresh mixed case 10

An event has probability 0.2 on each independent trial. Over 5 trials, probability of no occurrences = 0.8^5. At least one = 1−0.8^5.

90. Diagnose before calculating

For every fresh problem, write four labels before the formula:

  • Outcome: what is one elementary outcome?
  • Weight: are elementary outcomes equally likely?
  • Order: does sequence/role matter?
  • Dependence: do earlier outcomes change later probabilities?

91. The first invalid probability step

If an answer is wrong, find the first model error rather than only the arithmetic error. Common first failures include:

  • wrong elementary outcome;
  • assuming equal likelihood;
  • ignoring order;
  • using replacement when there is none;
  • treating dependent stages as independent;
  • using wrong conditional denominator;
  • double-counting overlapping cases;
  • counting identical objects as distinct.

92. Calculator use after model definition

Technology can evaluate combinations, binomial probabilities and numerical expressions efficiently. The calculator should enter after the learner has stated the model. Otherwise a fast numerical answer can hide a wrong sample space.

93. Verification with an alternate representation

Check a counting answer by listing a smaller case. Check a tree result with combinations when appropriate. Check a conditional probability with a two-way table. Check an “at least one” calculation by small-case enumeration. Independent representations expose different failure modes.

94. Probability and counting are linked but not identical

Counting tells how many outcomes exist under a definition. Probability also needs weights. Counting is sufficient for direct ratios only when the counted elementary outcomes are equally likely.

95. Sample-space transfer test

A learner is transfer-ready when they can face an unfamiliar story problem and define the random experiment without a topic label. The strongest evidence is not recall of a formula name. It is correct construction of outcomes and conditions.

96. Final mixed transfer set

  1. Three cards are drawn without replacement. Explain how the sample space changes after each draw.
  2. Two identical prizes are given to two of eight students. Does order matter?
  3. President and vice-president are chosen from eight students. Does order matter?
  4. A fair die is rolled until a six occurs. Is the number of rolls fixed?
  5. A point is chosen uniformly from a line segment. Should outcomes be counted as a finite list?
  6. A survey samples households, not people. What is the sampling unit?
  7. A table gives P(A), P(B), and P(A∩B). How can independence be tested?
  8. A bag draw is repeated with replacement. What condition tends to stay constant?
  9. A test asks P(A|B). Which group defines the denominator?
  10. A code allows repeated symbols. Which counting assumption changes?

97. Final self-check

  • I can say what one elementary outcome is.
  • I know whether my elementary outcomes are equally likely.
  • I know whether order matters.
  • I know whether repetition/replacement is allowed.
  • I know whether later probabilities depend on earlier outcomes.
  • I can distinguish mutually exclusive from independent.
  • I can identify the conditional denominator.
  • I can justify addition, multiplication, complement or counting choices.
  • I can test a complicated formula on a small listable case.
  • I state assumptions before using calculator output.

98. Continue through BTT Mathematics

Use H2 Mathematics 9758 | Permutations and Combinations for counting structures and Statistics and Data for the wider data-and-probability system. Use How to Check Maths Answers when the problem is verification rather than model selection.

99. The durable target

Probability becomes reliable when the formula is the last compression of reasoning, not the first guess.

Build the sample space. State what changes. Decide what is equally likely. Then calculate.

Probability Model Clinic: Ten More Ways the Sample Space Changes the Answer

1. Ordered draws

A bag contains A, B and C. Two letters are drawn without replacement and their order is recorded. Outcomes AB and BA are different. If the question records only the unordered pair, they become one outcome. Define the recorded object first.

2. Equal labels, unequal weights

A spinner has three labelled sectors but angles 180°, 120° and 60°. Probabilities are 1/2, 1/3 and 1/6, not 1/3 each. Count labels only after confirming equal likelihood.

3. Replacement resets composition

From a bag with 1 red and 3 blue, two draws with replacement give P(RR)=(1/4)^2. Without replacement, RR is impossible because there is only one red counter. One phrase changes the entire model.

4. “At most” needs the correct complement

For a binomial-style count X, the complement of X≤2 is X≥3, not X=3. Translate the event before using a formula.

5. Conditioning can make old probabilities obsolete

If a card is known to be a face card, the relevant sample space is the 12 face cards, not the full 52-card deck. Every conditional problem should identify the reduced universe explicitly.

6. Different sampling units

Selecting one student uniformly from a school differs from selecting one class uniformly and then one student uniformly inside that class when class sizes differ. “Random student” and “random class then student” are different mechanisms.

7. Independence needs a test

If P(A)=0.3, P(B)=0.4 and P(A∩B)=0.12, the multiplication condition holds. If the intersection probability is 0.18, the events are not independent under that model.

8. Small-case enumeration checks counting formulas

Before trusting a general arrangement formula, test n=3 or n=4 by listing outcomes. If a formula predicts six unordered pairs from three objects, the listing immediately reveals the overcount.

9. Simulation checks implementation

A simulation should reproduce the stated probability mechanism. If a without-replacement experiment is coded with independent constant probabilities, a large number of trials only makes the wrong model more stable.

10. Interpretation completes the calculation

A probability of 0.2 means one-fifth under the model; it does not guarantee one success in every five real trials. Probability describes uncertainty across the model, not a schedule of outcomes.

Extended mixed transfer set

  1. A committee of four is selected from ten people. Explain why order does not matter.
  2. Four ranked prizes are awarded among ten people. Explain why order/role matters.
  3. Two cards are drawn without replacement. Explain why the second draw is conditional on the first.
  4. A die is rolled until a six appears. Identify why a fixed-n binomial model is not the immediate model for the stopping time.
  5. A point is chosen uniformly from a square. Express the probability of landing in a region as an area ratio.
  6. Given P(A)=0.5, P(B)=0.4 and P(A∩B)=0.2, test independence.
  7. Two events are mutually exclusive with positive probabilities. Explain why they cannot be independent.
  8. A code may begin with zero. Explain how this differs from counting positive integers with the same number of digits.
  9. A sample is chosen from households but a conclusion is written about individuals. Identify the unit-of-analysis problem.
  10. An AI tool says two possible sums of dice are equally likely because both are single numbers. Repair the sample space.

Model sentence before arithmetic

Require the learner to write one line such as “There are 36 equally likely ordered pairs”, “The second draw is conditional because there is no replacement”, or “Order does not matter because only committee membership is recorded”. That line often reveals whether the formula is justified before calculation begins.

Fresh retest without a formula label

A box contains 5 white and 4 black counters. Three counters are drawn without replacement. Find the probability of exactly two white counters. The learner may use combinations or a tree, but must first state the sample-space mechanism and justify the route.

Combination route:

favourable selections = C(5,2)C(4,1); total selections = C(9,3).

Therefore probability = [10×4]/84 = 10/21.

The formula is earned by the unordered equally likely three-counter selection model.

Final rule

When probability work goes wrong, inspect the model before the calculator. A correct formula applied to the wrong sample space is still a wrong solution.

Release Check: State the Probability Model Before the Number

Before a probability answer is complete, the learner should be able to state the model in one sentence.

Examples:

  • “The 36 ordered pairs from two fair dice are equally likely.”
  • “The second draw is conditional on the first because sampling is without replacement.”
  • “Order does not matter because only committee membership is recorded.”
  • “I use the complement because ‘at least one’ is the complement of ‘none’.”
  • “These events overlap, so I subtract the intersection after adding their probabilities.”

That sentence is a compact proof that the chosen formula belongs to the sample space.

Five final contrasts

With replacement / without replacement: decide whether composition resets.

Ordered / unordered: decide whether roles or sequence create distinct outcomes.

Mutually exclusive / independent: decide whether events can occur together and whether conditioning changes probability.

Elementary outcomes / recorded values: decide whether recorded values have equal multiplicity.

Unconditional / conditional: decide which universe supplies the denominator.

Final fresh task

A bag contains 4 red, 3 blue and 2 green counters. Two are drawn without replacement. Find the probability that the colours are different.

One route uses the complement:

P(different)=1−P(same).

P(same)=P(RR)+P(BB)+P(GG)

=(4/9)(3/8)+(3/9)(2/8)+(2/9)(1/8)

=12/72+6/72+2/72=20/72=5/18.

Therefore:

P(different)=13/18.

The important step is not the subtraction. It is recognising the without-replacement branch probabilities and the mutually exclusive same-colour routes.

Final self-audit

  • My elementary outcomes are defined.
  • Equal likelihood is justified, not assumed.
  • Order and replacement are explicit.
  • Conditional denominators are correct.
  • Overlapping cases are not double-counted.
  • The answer is interpreted in the experiment that generated it.

Probability becomes trustworthy when the model is visible before the arithmetic.

Closure: Probability Is a Model of Possibilities, Not a Formula Hunt

The final transfer test is verbal before it is numerical. Give the learner a new probability problem and ask for five sentences before any calculation:

  1. What is the random experiment?
  2. What is one elementary outcome?
  3. Are elementary outcomes equally likely?
  4. Does order, replacement or conditioning matter?
  5. Which event is being measured?

If those sentences are accurate, the calculation is usually much easier to trust.

If they are inaccurate, a correct-looking numerical answer may simply be the output of a formula attached to the wrong model.

For revision, mix problems deliberately: with and without replacement, ordered and unordered selection, overlapping events, conditional tables, “at least one” complements, unequal outcome weights and small counting cases that can be listed by hand.

The learner should be able to explain why the same-looking numbers produce different probabilities when the sampling mechanism changes.

The release standard is therefore not “knows permutations, combinations and conditional probability”. It is “can construct the sample space and justify the operation before using those tools”.

Final Note

When two probability solutions disagree, compare their sample spaces before comparing their arithmetic. One solver may be counting ordered outcomes while the other is counting unordered selections; one may assume replacement while the other removes objects; one may condition on new information while the other keeps the original denominator. Those differences are mathematical, not stylistic.

The cleanest habit is to annotate the solution with the model assumption that licenses each operation. “Multiply because this is the branch probability after the first draw.” “Add because these paths are mutually exclusive.” “Subtract from one because the event is the complement.” “Use combinations because only membership matters.” When the justification is visible, errors become easier to locate and the same reasoning transfers to unfamiliar problems.

Probability Model Construction Lab: Build the Possibilities Before the Formula

Probability is easiest to misuse when the learner starts with a formula instead of the experiment. The same words—choose, arrange, at least one, given that, without replacement—can describe very different sample spaces depending on order, dependence, weighting and what is actually recorded.

The examples below use original synthetic scenarios. Their purpose is to train model construction rather than formula recognition.

Lab 1: what exactly counts as an outcome?

Roll two fair dice.

If the experiment records both faces, one outcome is an ordered pair such as (2,5). There are 36 equally likely outcomes.

If the experiment records only the sum, the recorded values are 2 through 12, but these are not equally likely because different sums have different numbers of underlying ordered pairs.

The first modelling question is therefore not “Which formula?” but “What is one elementary outcome in this experiment?”

Lab 2: labels do not guarantee equal probability

A spinner has regions labelled A, B and C. If the regions are unequal in area, the three labels need not have probability 1/3 each.

Counting labels works only when the elementary outcomes being counted are equally likely.

Lab 3: order changes the sample space

Choose two letters from A, B, C.

If order matters, possible outcomes are AB, AC, BA, BC, CA, CB: six.

If order does not matter, possible selections are AB, AC, BC: three.

The mathematical distinction between arrangement and selection begins before factorial notation appears.

Lab 4: replacement changes later probabilities

A bag contains 4 red and 1 blue counter.

With replacement, the probability of red on the second draw remains 4/5 regardless of the first result.

Without replacement, a first red changes the bag to 3 red and 1 blue, so the second red probability becomes 3/4.

The word “replacement” changes the branch structure of the sample space.

Lab 5: independence should be checked, not guessed

Two events are independent when learning that one occurred does not change the probability of the other.

For two rolls of a fair die, the event “first roll is 6” is independent of “second roll is even”.

For two draws without replacement, “first card is red” and “second card is red” are generally not independent because the population changes after the first draw.

Lab 6: mutually exclusive is not independent

On one fair die roll, A = “result is 1” and B = “result is 2” are mutually exclusive. They cannot occur together.

Since both have positive probability, they are not independent: if A occurs, the probability of B becomes zero.

This is a common vocabulary trap because “unrelated” in everyday language is not the mathematical definition of independence.

Lab 7: complement as a model shortcut

Roll a fair die five times. Find the probability of at least one six.

Directly counting one six, two sixes, three sixes, four sixes and five sixes is possible but inefficient.

The exact complement is no six at all:

P(no six)=(5/6)^5.

Therefore:

P(at least one six)=1-(5/6)^5.

The formula follows from event structure.

Lab 8: the complement must be exact

The complement of “at least two successes” is not “no successes”. It is “zero or one success”.

Before using a complement, state the event in words and write what remains when it does not occur.

Lab 9: exactly one versus at least one

Two fair coin tosses have outcomes HH, HT, TH, TT.

Exactly one head: HT, TH → probability 1/2.

At least one head: HH, HT, TH → probability 3/4.

A one-word difference changes the event set.

Lab 10: conditioning changes the universe

A fair die is rolled and you are told the result is greater than 2. Find the probability it is even.

The conditional sample space is {3,4,5,6}. The favourable outcomes are {4,6}. Therefore:

P(even | result>2)=2/4=1/2.

The denominator changed because new information removed possibilities.

Lab 11: reverse conditioning is a different question

P(A|B) and P(B|A) generally differ because they condition on different populations.

“Probability of passing given that a student studied” is not the same as “probability a student studied given that they passed”.

The two fractions can share the same intersection numerator while using different denominators.

Lab 12: two-way table construction

PassedDid not passTotal
Studied54660
Did not study202040
Total7426100

P(pass|studied)=54/60=0.9.

P(studied|pass)=54/74≈0.730.

The conditioning statement identifies the denominator.

Lab 13: sample space by tree

A bag contains 2 green and 1 yellow counter. Draw twice without replacement.

Paths:

  • G then G: probability (2/3)(1/2)=1/3.
  • G then Y: probability (2/3)(1/2)=1/3.
  • Y then G: probability (1/3)(1)=1/3.

The path probabilities happen to be equal. That equality emerges from the branch model rather than from counting three labels blindly.

Lab 14: tree diagrams are not mandatory when symmetry is clearer

If a problem is naturally counted through combinations, a tree can become cumbersome. The representation should reveal the sample space, not satisfy a ritual.

Probability modelling includes choosing the representation that makes assumptions easiest to inspect.

Lab 15: combinations need equally weighted selections before simple counting works

If 3 objects are selected uniformly from a set of 10 distinct objects, all 3-element subsets are equally likely under the stated selection mechanism.

If the mechanism favours some objects, the same combination count no longer gives event probabilities directly.

Lab 16: permutations and combinations answer different sample-space questions

Choosing a president and vice-president from 8 people gives ordered role assignments: 8×7.

Choosing any two representatives gives unordered pairs: C(8,2).

Same people, different recorded outcome.

Lab 17: repeated objects change counting

The word LEVEL contains repeated letters. Treating all five letters as distinct would overcount arrangements.

Distinct arrangements are:

5!/(2!2!)=30,

because L and E each repeat twice.

The sample space must respect indistinguishable outcomes.

Lab 18: “at least” can be easier through complement in counting too

Suppose a committee of 4 is chosen from 7 people, including Alice. Probability Alice is included:

Complement: Alice excluded.

Total committees: C(7,4).

Committees without Alice: C(6,4).

Therefore:

P(Alice included)=1-C(6,4)/C(7,4).

Lab 19: sampling without replacement changes independence

From 5 items, 2 are defective. Select two without replacement.

P(first defective)=2/5.

P(second defective | first defective)=1/4.

The second-stage probability depends on the first outcome.

Lab 20: sampling with replacement restores identical stage probabilities

With replacement:

P(defective on each draw)=2/5.

The draws become independent under the usual random-replacement model.

Lab 21: Bayes thinking starts from a joint model

Do not memorise Bayes’ theorem as a symbol swap. Build the joint probabilities first.

If a condition occurs in 10% of a population, a test detects it 90% of the time, and has a 5% false-positive rate, draw the population branches. The probability of a positive test comes from both true-positive and false-positive paths.

The conditional probability of the condition given a positive test is then a ratio within the positive-test population.

Lab 22: base rates matter

Using the synthetic numbers above:

True positives: 0.10×0.90=0.09.

False positives: 0.90×0.05=0.045.

Total positive probability: 0.135.

Therefore:

P(condition|positive)=0.09/0.135=2/3.

A 90% detection rate does not imply a 90% posterior probability after a positive result.

Lab 23: continuous probability is area, not point counting

If a continuous random variable has density f(x)=2x on 0≤x≤1, then:

P(X>0.8)=∫[0.8 to 1] 2x dx = 1-0.64=0.36.

The sample space is continuous. Individual exact values have probability zero even though intervals have positive probability.

Lab 24: density height is not probability by itself

A density value may exceed 1. Probability is the area under the density over an interval, not the vertical height alone.

This is a major model shift from discrete counting.

Lab 25: expected value is not a guaranteed outcome

A game may have expected payoff $2 while no single play ever pays exactly $2.

Expectation is a probability-weighted average over the model. It should not be interpreted as a prediction that each trial produces the expected value.

Lab 26: expected value requires a complete probability model

If one possible outcome is omitted or probabilities do not sum to 1, the expected value calculation is incomplete.

Before computing expectation, audit the sample space and weights.

Lab 27: conditional independence can differ from marginal independence

At more advanced levels, two variables may appear associated overall but become independent within a conditioning group, or vice versa.

The important lesson for H2 learners is simpler: independence always belongs to a specified probability model. It is not an intrinsic everyday label.

Lab 28: event algebra should match words

“A or B” in probability is usually inclusive unless the problem states exactly one or the events are mutually exclusive.

Use:

P(A∪B)=P(A)+P(B)-P(A∩B).

The intersection is subtracted because it was counted twice.

Lab 29: “either but not both” is different

The event exactly one of A or B occurs is:

(A∩Bᶜ) ∪ (Aᶜ∩B).

Language precision determines event structure.

Lab 30: Venn diagrams can expose double counting

When event overlap is the main issue, a Venn diagram may make the model more transparent than a formula list.

Representation is part of probability reasoning.

Lab 31: sample-space completeness test

After listing outcomes, ask:

  • Can every possible experiment result be placed somewhere?
  • Did any outcome get counted twice?
  • Are supposedly equal-weight outcomes actually equal?
  • Does order matter?
  • Does one stage change the next?

Lab 32: sample-space granularity

A model can be too coarse. Recording only die sums hides the ordered-pair multiplicities. Recording every microscopic detail can be unnecessarily fine.

Choose a level at which elementary outcomes can be assigned probabilities correctly and events can be defined without ambiguity.

Lab 33: model before formula in card problems

For a standard deck problem, first decide:

  • how many cards are drawn;
  • whether replacement occurs;
  • whether order matters;
  • what event is being asked;
  • whether all elementary selections are equally likely.

Only then choose combinations, permutations, conditional probability or a tree.

Lab 34: model before formula in arrangement problems

Ask whether objects are distinct, whether positions have restrictions, whether circular arrangements identify rotations, and whether repeated objects create indistinguishable outcomes.

Factorials alone do not decide the sample space.

Lab 35: model before formula in repeated trials

Binomial structure requires more than “something happens many times”. Typical conditions include a fixed number of trials, two outcome categories per trial, constant success probability and independence under the model.

If the success probability changes after each draw without replacement, a simple binomial model may be inappropriate.

Lab 36: probability of zero is not always impossibility in continuous models

For a continuous distribution, P(X=a)=0 for an exact value can coexist with that value being possible in the sense of lying in the support.

This is another reason discrete intuition cannot simply be copied into continuous probability.

Lab 37: simulation as evidence

Simulation can check whether a theoretical probability is plausible. It does not replace a derivation when the probability can be established mathematically.

Small simulations also contain sampling noise. Compare observed frequency with theory over increasing numbers of trials rather than expecting exact equality.

Lab 38: simulation should match the model

If a simulation samples with replacement while the original experiment is without replacement, the code is simulating the wrong probability space even if it runs perfectly.

Model audit precedes software audit.

Lab 39: probability trees need branch sums of 1

At each node, the outgoing branch probabilities should sum to 1 if they cover all next-stage possibilities.

This is a quick structural check for missing or duplicated branches.

Lab 40: final probability model checklist

  • What is the experiment?
  • What is one elementary outcome?
  • Are the elementary outcomes equally likely?
  • Does order matter?
  • Is there replacement?
  • Are stages independent?
  • What information is conditioned on?
  • Does the event wording mean exactly, at least, at most or complement?
  • Is counting appropriate, or are unequal weights present?
  • Does the chosen representation expose all branches?

Transfer task 1: without replacement

A bag contains 5 black and 3 white counters. Two are drawn without replacement. Find the probability of exactly one white.

One route uses combinations:

C(3,1)C(5,1)/C(8,2)=15/28.

A sequential route gives:

(3/8)(5/7)+(5/8)(3/7)=30/56=15/28.

Two representations confirm the same model.

Transfer task 2: with replacement

Same bag, but replace after the first draw. Probability of exactly one white:

(3/8)(5/8)+(5/8)(3/8)=30/64=15/32.

The words changed only slightly; the probability changed because the sample space changed.

Transfer task 3: conditional denominator

A fair die is rolled. Given that the result is prime, find the probability the result is odd.

Prime outcomes are {2,3,5}. Odd among them: {3,5}. Probability = 2/3.

Transfer task 4: order matters

Three runners A, B, C finish in a random order with no ties. Probability A finishes before B?

By symmetry, among all six orders, A precedes B in three. Probability = 1/2.

The sample space is permutations, but the event can be solved more elegantly by symmetry.

Transfer task 5: at least one

A biased coin has P(H)=0.4. Toss it 4 times independently. Probability of at least one head:

1-(0.6)^4.

Transfer task 6: independence test

A card is drawn from a standard deck. A = “card is a heart”; B = “card is a queen”.

P(A)=13/52=1/4.

P(B)=4/52=1/13.

P(A∩B)=1/52.

Since (1/4)(1/13)=1/52, A and B are independent under this model.

Transfer task 7: mutually exclusive test

A = “card is a heart”; B = “card is a spade”. They are mutually exclusive on one draw. Since both have positive probability, they are not independent.

Transfer task 8: expected value

A synthetic game pays $10 with probability 0.2 and loses $1 with probability 0.8.

Expected payoff:

10(0.2)+(-1)(0.8)=2-0.8=$1.20.

This does not mean every play returns $1.20. It is the probability-weighted long-run average under the model.

Transfer task 9: unequal outcome weights

A spinner has region A covering half the circle, B one quarter and C one quarter. Recording only labels gives three outcomes, but they are not equally likely.

P(A)=1/2, P(B)=P(C)=1/4.

Transfer task 10: model audit before calculation

A problem says: “Two students are chosen from a class.” Before calculating, write down:

  • class size;
  • whether selection is uniform;
  • whether order matters;
  • whether the students must satisfy categories;
  • whether the selection is simultaneous or sequential.

If any required information is missing, the probability may be under-specified.

Closing principle

Probability formulas do not create the model. They operate inside one.

Build the possibilities first. State what one outcome means. Decide how outcomes are weighted. Track order, replacement, dependence and conditioning. Then choose the calculation that matches the structure.

The strongest probability student can explain why the denominator is what it is before pressing a calculator key.

Final Probability Model Clinic: Ten Decisions Before the Answer

1. Define the elementary outcome

Before using any formula, state what one outcome looks like. Is it a die face, an ordered pair, an unordered selection, a full path through a tree, or an interval of a continuous variable?

If the elementary outcome is unclear, later counting is fragile.

2. Check whether outcomes are equally likely

The familiar fraction “favourable over total” requires equally likely elementary outcomes when used by direct counting. Sums of two dice, uneven spinner regions and weighted selections are common situations where labels are not equally likely.

3. Decide whether order matters

If swapping positions creates a new outcome, use an ordered model. If the selection remains the same, order may not matter.

Role assignments, rankings and sequences usually care about order. Committees and unordered subsets often do not.

4. Decide whether the population changes

Without replacement, later probabilities can depend on earlier outcomes. With replacement, the original composition is restored.

This one condition can change independence, denominators and the correct tree structure.

5. Decide what is conditioned on

“Given that B occurred” means the new relevant universe is B. Every conditional probability should be read as a probability inside that restricted world.

Write the denominator group in words before calculation.

6. Translate event language exactly

At least, at most, exactly, none, one or more, either, both and not both define different event sets.

A small wording change can alter the entire probability model while leaving the same symbols available.

7. Look for a complement

At least one success is often easier as one minus no successes. More than two can be easier as one minus at most two.

The complement is useful only when it matches the event exactly.

8. Choose a representation that exposes structure

Use a tree for sequential dependence, a table for two categorical variables, a Venn diagram for overlapping events, combinations for uniform unordered selections, permutations for ordered arrangements and integration for continuous densities.

Representation is part of the reasoning, not decoration.

9. Verify structural bounds

Every probability must lie between 0 and 1. Complementary probabilities should sum to 1. Branch probabilities from a complete node should sum to 1. Mutually exclusive path probabilities can be added.

These checks catch many model errors quickly.

10. State the model before the number

A strong final solution can often be summarised in one sentence before calculation:

“The draws are without replacement, so the second-stage probabilities are conditional.”

“The committee is unordered, so combinations are appropriate.”

“The event is at least one success, so I will use the complement of zero successes.”

The sentence proves that the learner knows what probability space the formula is operating inside.

Model comparison task 1: same bag, different sampling rule

A bag contains 3 red and 2 blue counters.

With replacement:

P(two reds)=(3/5)(3/5)=9/25.

Without replacement:

P(two reds)=(3/5)(2/4)=3/10.

The event label is identical. The sampling model changes the answer.

Model comparison task 2: same people, different outcome definition

From 6 students:

Select 2 representatives → C(6,2)=15.

Select a captain and deputy → 6×5=30.

The selected people may be the same pair, but role order doubles the outcome count.

Model comparison task 3: same observed value, different elementary space

Rolling two dice and observing a sum of 6 can arise from:

(1,5), (2,4), (3,3), (4,2), (5,1).

The sum is one recorded value but five elementary ordered outcomes.

This explains why sums cannot be treated as equally likely labels.

Model comparison task 4: same percentage, different conditioning

Suppose 30 of 40 students who studied passed, while 10 of 60 who did not study passed.

P(pass|studied)=30/40=0.75.

But:

P(studied|pass)=30/40=0.75 only coincidentally here because total passes also equal 40. In another table the values can differ sharply.

Never infer reverse conditioning from one conditional probability.

Model comparison task 5: mutually exclusive versus independent

On one die roll:

A = “odd”; B = “even”. They are mutually exclusive and complementary.

They are not independent because knowing A occurred makes B impossible.

Mathematical vocabulary should be checked from definitions.

Model comparison task 6: continuous versus discrete

For a discrete die, exact outcomes have positive probability.

For a continuous density, an exact point can have probability zero while intervals have positive probability.

The same word “probability” operates on different sample-space structures.

Model comparison task 7: independent trials versus changing success probability

A binomial model assumes a constant success probability across independent trials.

Drawing without replacement from a small finite population generally changes the success probability after each draw.

Recognising when binomial conditions fail is more important than remembering the binomial formula.

Model comparison task 8: tree versus combination check

For a uniform unordered selection, solve once with combinations and once through sequential probabilities when practical. Agreement between two representations is strong verification.

Model comparison task 9: event algebra check

If:

P(A)=0.6, P(B)=0.5, P(A∩B)=0.3,

then:

P(A∪B)=0.6+0.5-0.3=0.8.

The subtraction removes the intersection that was counted twice.

Model comparison task 10: probability release check

Before finalising any answer, ask:

  • Could another reasonable reader interpret one outcome differently?
  • Did I assume equal likelihood without evidence?
  • Did I ignore order or replacement?
  • Did conditioning change the denominator?
  • Did I confuse mutually exclusive with independent?
  • Does the final value lie in [0,1]?
  • Can I explain the model in one sentence?

Parent and tutor signal

Probability progress is visible when the learner stops saying “I used this formula because the question looked similar” and starts saying “I used this structure because order does not matter”, “the draws are dependent”, or “the condition restricts the sample space”.

That change is the movement from formula matching to probability modelling.

Release standard

The learner is ready when they can construct the sample space, assign or justify weights, identify dependence and conditioning, select a representation, calculate the event, and test the result against structural bounds without heavy prompting.

Probability becomes reliable when the model is built before the arithmetic begins.

Probability Last Mile: From Correct Calculation to Controlled Model

A probability answer is not fully controlled until the learner can explain what changed when the model changes. This is the final release test because many formula errors are really model errors.

Change one condition

Take a completed problem and alter exactly one condition:

  • with replacement → without replacement;
  • ordered → unordered;
  • independent → dependent;
  • at least one → exactly one;
  • unconditional → given that;
  • discrete → continuous.

Then ask which parts of the solution must change and which parts stay valid.

Explain the denominator

For every probability fraction, say what the denominator represents. In conditional probability, name the conditioned group. In combinations, explain what the total selections count. In continuous probability, explain why area replaces simple outcome counting.

If the denominator cannot be described in words, the model is not yet secure.

Explain the independence assumption

When multiplying stage probabilities, state whether independence is being used or whether the second factor is conditional. This one sentence distinguishes a model from blind multiplication.

Explain why order matters—or does not

Ask whether AB and BA are genuinely different outcomes. If yes, an ordered structure is needed. If no, counting both separately overcounts the space.

Explain why the answer is plausible

Check:

  • result lies between 0 and 1;
  • more restrictive events do not receive larger probability without reason;
  • complements add to 1;
  • symmetry is respected where the experiment is symmetric;
  • branch probabilities remain coherent.

Final transfer challenge

A bag contains 4 red and 2 blue counters. Two are drawn.

First solve the probability of exactly one blue without replacement. Then solve it with replacement. Do not begin with a formula. Write the sample space logic first.

Without replacement:

(2/6)(4/5)+(4/6)(2/5)=16/30=8/15.

With replacement:

(2/6)(4/6)+(4/6)(2/6)=16/36=4/9.

The event wording stayed the same. The sampling rule changed the model and therefore the probability.

Release sentence

The strongest probability solution ends with more than a number. It can say: “These are the elementary outcomes, this is how they are weighted, this condition changes the relevant sample space, and this calculation matches that structure.”

That is the point at which probability stops being a formula hunt and becomes mathematical modelling of uncertainty.

A final habit is to rebuild one completed probability question after changing a single condition. Replace “with replacement” by “without replacement”, change an unordered selection into ordered roles, or add a conditioning statement. If the learner can explain exactly which branches, denominators or counting rules must change—and which parts remain unchanged—the model is genuinely understood rather than memorised. Probability competence is visible when the student can reconstruct the sample space under changed assumptions before reaching for a familiar formula.

The same discipline should survive outside textbook probability. Whenever uncertainty is described with percentages, rates or likelihoods, ask what population or experiment generated the number, what outcomes were possible, how those outcomes were weighted, and what information was conditioned on. This prevents a familiar-looking percentage from being mistaken for a fully specified probability model. The calculation becomes trustworthy only when the underlying possibilities and assumptions are visible enough to audit.

A mature probability solution should also survive a change in wording. If the numbers remain the same but “with replacement” becomes “without replacement”, “at least one” becomes “exactly one”, or an unconditional question becomes conditional, the learner should rebuild the sample space rather than reuse the old formula by reflex. That flexibility is the real release standard: probability is understood when the student can reconstruct the model under changed assumptions and explain why the calculation must change.