JC2 · H2 Mathematics · Notes
Choose a binomial model, translate tails and conditions, solve unknown parameters and sample sizes, and count packets, inspections and modes.
Binomial Distribution summary and formula sheet: every method on 5 pages
Mr Ian Ang’s summary of the whole binomial distribution chapter on five pages.
Page 1 of 5 Setting up and the calculator Part 1: formulas and question types
H2 Math summary notes
Binomial Distribution formulas and methods, section by section
Page 1 of 5Setting up and the calculatorPart 1: formulas and question types
B(n, p): conditions and setting up X
- Properties of a binomial distribution: a fixed number, , of independent trials; the probability of success, , is the same for each trial; each outcome is either a success or a failure.
- The formula is in MF27, with as the fixed range.
- Example: a spinner lands on red with probability 0.25 and is spun 12 times. The trial is a spin, so ; the success is red, so .
- Write: Let be the number of spins, out of 12, that land on red. Then .
- Warning: binomial means 'out of trials', and is upper case.
Assumptions in context: the format of the answer
- Format of the answer: (1) A [trial], [with its success condition], is independent of another [trial]. (2) The probability of a [trial], [with its success condition], is constant at ___ for all [trials]. Fill each [ ] and the blank from the question.
- Example: 8% of the bulbs a factory makes are faulty, and is the number of faulty bulbs in a box of 30. State two assumptions, in context, for to be binomial.
- (1) A bulb, being faulty, is independent of another bulb. (2) The probability of a bulb, being faulty, is constant at 0.08 for all bulbs.
- Warning: 'two outcomes' and 'a fixed number of 30 bulbs' are not assumptions, because they are facts in the question. Do not write 'the probability of a faulty bulb is independent of ...'.
- Not binomial: boxes picked until one has no faulty bulb, because the number of boxes is not fixed.
P(X = r) and P(X ≤ r) on the GC: binompdf, binomcdf and the words
| In words | Write as | |
|---|---|---|
| fewer than 3 | 0.391 | |
| at least 3 | 0.609 | |
| more than 3 () | 0.351 | |
| between 2 and 5 inclusive | 0.787 | |
| more than half (, so ) | 0.0143 |
- Press 2nd VARS (DISTR). A is binompdf( for (pdf: equals) and B is binomcdf( for (cdf: at most). Enter trials: 12 for , p: 0.25 for and the x value: 3, then Paste.
- The GC does only = and , so rewrite every probability as . With , the values up to 2 are and the values from 3 are ; convert the second group by subtracting from 1.
- For : fewer than 3 is ; at least 3 is ; more than 3 () is .
- Between 2 and 5 inclusive is ; more than half (, so ) is .
Conditional probability, binomial: 'given' and the x values in both conditions
- 'Given' means conditional, so use .
- Example: ; find . This is .
- The values in both conditions are 2, 3, 4, 5, so the top is .
- Then .
- Warning: wrong is ; the goes inside one P( ).
Page 2 of 5Binomials inside binomials
A binomial inside a binomial: carry the probability into a new B( , )
- Example: seeds fail with probability 0.15. A packet of 20 is good if at most 2 fail. Find the probability that at least 6 of the 12 packets in a box are good.
- Let be the number of seeds, out of 20, that fail: , and (5 s.f.). The new trial is a packet and the success is good. Let be the number of packets, out of 12, that are good: . Then .
- (b) Five packets: two with exactly 2 failed, three with exactly 3. The probability is . Like a tree, counts the orders.
- (c) Every packet has at least 2 failed: .
- At least 24 failed in the whole box: and . This is larger, as the total also counts boxes where some packet has fewer than 2.
- Warning: carry 5 s.f. into the new B( , ); give the answer to 3 s.f.
Splitting up conditions: the kth trial is the rth success
- When the th trial is the th success, split the trials into independent blocks.
- Example: each shopper uses a voucher with probability 0.4. Find the probability that more than 7 of the first 18 shoppers use a voucher and the 6th shopper is the 3rd to use one. The blocks are: (1) 2 out of to , (2) is the 3rd, (3) at least 5 out of to .
- Let be the number, out of the first 5, that use a voucher: . Let be the number, out of the last 12, that use a voucher: . The required probability is .
- Warning: more than 7 in all, with 3 by the 6th, means more than 4 in the last 12, so .
- The 8th and last variant: of the first 15, the 13th is the 8th and last. The blocks are 7 out of to , then , then both no. With , the probability is .
Mean, variance and the most probable value: E(X) = np, Var(X) = npq and the GC table
| 17 | 0.1074 |
|---|---|
| 18 | 0.1156 (largest) |
| 19 | 0.1144 |
- : find , and the most probable value of . The formulas are in MF27.
- and , where . On average 18.5 out of 50; is the spread.
- For the most probable value, press Y= and enter . Press 2nd WINDOW and set TblStart = 17, near . Press 2nd GRAPH for the table.
- The table gives : 0.1074; : 0.1156; : 0.1144. The most probable is 18.
- Warning: this is not a 'hence'. Use the GC, look around and take the largest .
Page 3 of 5Finding n, a and pPart 2: the long worked layouts
Finding n or a: the GC table
| 44 | 0.9016 (tick) |
|---|---|
| 45 | 0.8953 |
- When the unknown is an integer ( or ), use the GC table and put where the unknown is.
- Left: ; find the smallest such that . Enter with the unknown at the x value, then 2nd GRAPH for the table: gives 0.8868 and gives 0.9578 , a tick. The smallest is 8.
- Right: each phone case is faulty with probability 0.04; ; find the largest . Let be the number faulty out of . Then .
- Enter with the unknown at the trials. Set Tbl = 10: gives 0.9252 and gives 0.8609. Then set Tbl = 1 from 40: gives 0.9016 , a tick, and gives 0.8953 . The largest is 44.
- Warning: turns round to . Show the row on each side: the least is the first row that works, the largest the last.
Finding p: the GC graph, or Var(X)
- When is any real value, use the GC graph (a table is for integers; a graph is for any real value).
- Example: , where , and . Write down an equation in and find . gives , that is , entered as .
- Set WINDOW (since ), then 2nd TRACE and choose zero. The graph crosses the axis at 0.0265 and 0.142. So (rejected as ) or .
- (b) Not forced to form an equation: , . Use and , then 2nd TRACE intersect: .
- (c) From the variance: , , . Then , and (0.75 rejected).
- Warning: use the formula only when the question makes you write the equation; otherwise use the GC.
Page 4 of 5Most probable value and the mode
Most probable value from the ratio P(X = k + 1) / P(X = k)
- Example: . By considering , find the most probable value of .
- The ratio is , cancelling with .
- Step 1, start here: gives , so and : . So .
- Step 2: gives , , so .
- Step 3: is the most probable.
- Warning: start at even when the question does not guide you.
Mode given: the range of p
- One mode, : the bar at is the only peak, so and , and these give a range. Two modes (rare): .
- Example: eggs come in boxes of 8. The probability that an egg has a double yolk is . The modal number of double yolked eggs in a box is 3. Find exactly the range of values of .
- , with , , .
- (1) : , so . (2) : , so . Hence .
- Warning: divide by the common factor (positive), make the subject, then intersect the two.
Mode given: the values of n, without a calculator
- Example: and the mode of is 5. Without a calculator, find the possible values of .
- : . Write each as factorials, , and cancel and the common powers.
- This leaves , that is , so .
- Similarly gives , so .
- So ; is an integer, so . Warning: cancel before multiplying out: and .
Page 5 of 5Many samples and inspection of batches
Many samples: the Central Limit Theorem
- For : and .
- With samples, large (), by the Central Limit Theorem and , approximately.
- Example: . 50 independent observations of are taken. Find the probability that their mean exceeds 6.5.
- and . By the CLT, since 50 is large, approximately, and .
- Warning: write 'approximately' and the reason: the number of samples is large.
Inspection of batches: the tree, P(pass) and the expected number checked
| 15 | 30 | |
|---|---|---|
| 0.16910 |
- Example: each item is defective with probability 0.06. A sample of 15 is checked. At most 1 defective: the batch passes. Exactly 2: a second sample of 15 is checked, and the batch passes only if it has none. Otherwise it fails. Find , , the expected number of items checked and the expected number of batches that fail in 365 days.
- Let be the number of items, out of 15, that are defective: . The tree has branches (pass, 0.774, 15 checked), (0.169) and (fail, 0.0571, 15 checked). After : (pass, 0.395, 30 checked) and (fail, 0.605, 30 checked).
- . Then .
- Let be the number of items checked per batch: or 30, not binomial. and . . The expected number of failed batches in 365 days is .
- In terms of : a sample of 10, pass if none; if exactly 1, a second sample of 10, pass if none. .
- Warning: only for a binomial; the number checked is not binomial (15 or 30, not 0 to ).
Worked examples
Binomial Distribution worked examples
Mr Ian’s worked examples for the chapter, with the full working.
Count successes in a fixed number of trials
Name one trial and its success condition. If there are n trials, each has the same success probability p, and all trials are independent, the success count is . Success is just the event being counted; a defect can be a “success” in this mathematical sense.
State the assumptions in context
Count flavoured chocolates in one packet
A large production run contains 20% mint chocolates. A packet contains 20 randomly selected chocolates. Let X be its number of mint chocolates. State the model assumptions, then find the probability of exactly four mint chocolates.
Assume that whether each chocolate is mint is independent of the others, and that the probability of being mint is constant at 0.2 for every selection. A small sample from a very large mixed run makes these assumptions plausible; a shared batch fault could invalidate independence.
Selecting chocolates until the first mint one appears is a different experiment: the number of selections is not fixed, so that waiting count is not binomial. Sampling a substantial fraction of a finite bag without replacement also changes the success probability.
Translate the integer boundaries before calculating
At least three; more than four but not more than eight
For the same packet, find and . A cumulative binomial calculation returns .
The second event contains 5, 6, 7 and 8. Subtracting would wrongly include four mint chocolates.
Intersect the condition with the requested event
Fewer mint than plain, given at least three mint
Find the probability that a packet has fewer mint than plain chocolates, given that it has at least three mint chocolates.
Together with the given condition, the numerator event is .
Here q is the requested conditional probability. The denominator is the probability of the given event alone. The strict inequality excludes the 10–10 tie.
Unknown parameters: form equations from what is given
Recover n and p from the mean and variance
A binomial variable has mean 6 and variance 4.2. Find n and p, then the probability of no successes.
Check that n is a positive integer and . An impossible parameter value means the proposed binomial model or the input equations must be reconsidered.
An unknown probability inside a mixed pack
A pack contains two toys with independent fault probability 0.02 and three keychains with independent fault probability p. All five fault events are independent. Given at most one faulty item in a pack, the probability that its only fault is a toy is 0.035. Find p.
Let A count faulty toys and K count faulty keychains. The denominator contains only the mutually exclusive cases :
Write the numerator N and denominator D separately to keep the conditional equation clear:
The conditioning event has positive probability, so . Divide by , then solve:
A single binomial model for all five items would be invalid because toys and keychains do not have a common fault probability.
Count packets or count individual chocolates?
Find the least sample size
Each chocolate is independently mint with probability 0.2. Find the least n such that the probability of fewer than three mint chocolates in n selections is at most 0.3.
Use a table with n restricted to integers. The boundary values are:
| n | P(X ≤ 2) | Meets condition? |
|---|---|---|
| 17 | 0.309622 | No |
| 18 | 0.271342 | Yes |
Thus the least n is 18. As n increases, adding an independent trial cannot lower the success count, so decreases. The preceding integer verifies minimality.
Ten packets: every packet versus the total
A box contains ten independent packets of 20 chocolates. Find the probability every packet has at least three mint chocolates, then the probability the box contains at least 30 mint chocolates.
For one packet, . Every packet qualifies only when all ten independent packet events occur:
For the second event count all 200 individual chocolates. Their success probability is still 0.2:
Every packet having three mint chocolates guarantees at least 30 in total. The converse fails: packets may compensate for each other. This explains the larger second probability.
If asked instead for exactly eight qualifying packets, define :
The 10 trials are now packets, and their success probability is q, not 0.2.
Split off a specified trial or inspection stage
The eighth selection is the sixth blue
Balls are selected with replacement, independently, with blue probability 0.6. Find the probability that selection 8 is the sixth blue selection.
Exactly five of the first seven must be blue, and selection 8 must itself be blue. Its position is fixed, so the combination applies only to the first seven:
If the question additionally says the sixth and last blue in ten selections, selections 9 and 10 must both be non-blue: multiply this result by .
An inspection may require a second sample
Each independent sample contains 10 items with defect probability 0.1 per item. Accept immediately if the first sample has at most one defect; reject if it has at least three. With exactly two, take a second independent sample and accept only if it has at most one defect. Find the acceptance probability, then the chance a second sample was needed given acceptance.
Here and . The branch that triggers retesting is not itself acceptance. It must be followed by a successful second inspection.
Most probable means the largest probability
Find a range of p for a unique mode
For , find the range of p for which 2 is the unique most probable count. Compare its two neighbours, using .
Hence . Binomial probabilities increase up to a mode and then decrease; the adjacent ratios decrease with k, so these two comparisons establish the global maximum.
At , counts 1 and 2 tie. At , counts 2 and 3 tie. If “2 is a mode” allows a tie, include both endpoints; “unique” excludes them.
Watch the boundary cases produce a tie
p = 0.40: mode 2; highest probability 0.3456.
Applying the P(X = k) formula
For , with , write . Show that the ratio of consecutive probabilities is
for integers k from 0 to n − 1. Start with the binomial formula. After the common powers cancel, expand the two combinations:
Thus means ; means the probabilities have begun to decrease. The ratios decrease as k increases. If a ratio equals 1 at the changeover, the two adjacent values tie as modes.
Find the most probable count when n and p are known
Find the most probable number of successes when .
The probabilities increase through the step from 76 to 77 and decrease from 77 onwards. Check both neighbours:
Hence 77 is the unique most probable count. There is no tied pair because the changeover value is not an integer.
A total over many independent samples
If independently, each has mean 4 and variance 3.2. For 50 packets, the total has mean 200 and variance 160; the mean has mean 4 and variance . An exact combined count is also available: . Use the Sampling guide to work through the CLT and a probability for a sample mean.