Try
Flip four independent fair coins and count heads. Instead of tracking the exact sequence, record X, the number of heads. Several sequences produce X=2: HHTT and HTHT are different outcomes with the same count. Predict why multiplying one sequence probability by a counting factor is necessary.
Predict
Would the same binomial model describe drawing four cards without replacement from a small deck, or tossing until the first head? In the first case probabilities can change with earlier results; in the second the trial count was not fixed in advance. Binary outcomes alone do not establish a binomial model.
Experiment
There are six choices for the two head positions among four coins. Each specified sequence has probability(1/2)⁴=1/16, so P(X=2)=6/16=3/8. Enumerate the six sequences to verify the coefficient rather than accepting it as decoration.
For three independent trials with success probability1/5, calculate at least one success through its complement:1−(4/5)³=61/125. Compare the exact answer with the incorrect shortcut3·(1/5), which double counts outcomes with multiple successes when used as a union probability. That product is instead the expected success count.
Understand
A binomial variable X counts successes in n fixed trials with two categorized outcomes per trial, common success probability p and independence. The labels success and failure are mathematical categories, not judgments about desirability. For k successes, P(X=k)=C(n,k)p^k(1−p)^(n−k), where C(n,k) counts the choices of k success positions among n trials.
The coefficient is n!/[k!(n−k)!] for integer0≤k≤n. Every sequence with k successes has the same product probability under the common-p independent model. Different sequences are disjoint, so their probabilities add. This reasoning explains both the coefficient and the powers. At p=0 or p=1, use the corresponding deterministic distribution directly rather than getting distracted by ambiguous-looking zero powers in a calculator.
Mean is np and variance is np(1−p); standard deviation is√[np(1−p)]. For n=20 and p=0.3, mean6 and variance4.2 describe the sampling distribution of the count, not a guarantee of six successes. The sample proportion X/n has the same information on a different scale when n is positive.
Sampling without replacement from a small finite population violates the independent-trial model. With three red and two blue balls, two reds without replacement has probability(3/5)(2/4)=0.3, versus0.36 under independent replacement. When the sample is a small fraction of a large population, independence can sometimes be a useful approximation, but label it and check its adequacy.
If independent customers have different purchase probabilities, the count need not be binomial. Its expectation still equals the sum of those probabilities. Shared circumstances can create dependence as well. Diagnose which assumption fails instead of using the formula merely because the outcome is a count.
Apply
Choose a proposed trial sequence and write four sentences checking fixed n, two outcomes, independence and constant p. For a school survey, distinguish a fixed number of randomly selected respondents from stopping after obtaining a desired number of favorable replies. The stopping rule changes what random variable is being modeled.
Estimate a binomial probability from simulation evidence
The exact binomial formula and a simulation estimate answer the same event question in different ways. To simulate X for four independent fair trials, use four fresh fair coin tosses and record the number of heads, then repeat the entire four-toss experiment. Do not reuse a single toss four times: that makes the outcomes dependent. Before running, fix the event X≥3 and the number of repetitions. Count each four-toss experiment once in the denominator.
For practice, suppose a simulation report supplies the following illustrative synthetic output, not an actual classroom run: counts for X=0,1,2,3,4 are respectively 6,24,36,26,8 in 100 repetitions. The empirical distribution is 0.06,0.24,0.36,0.26,0.08, and the estimated probability of X≥3 is (26+8)/100=0.34. The independent-trial model instead gives the exact probability (4+1)/16=0.3125. Their difference 0.0275 is Monte Carlo discrepancy, not proof that the probability formula failed. Repeated simulation can reduce typical sampling noise, but cannot repair a wrong independence model or ensure monotone improvement on every run.
Changed-case written check. A second illustrative report gives counts 14,49,76,47,14 in 200 repetitions. Construct its empirical distribution, estimate P(X≥3), and compare with the same exact model. Explain how the target changes if you stop each four-toss block at its first head. Then perform and retain your own fixed-size coin simulation, recording its actual counts rather than relabeling these supplied counts as your experiment.
Feedback
The relative frequencies are 0.07,0.245,0.38,0.235,0.07; the upper event estimate is 61/200=0.305. It differs from 0.3125 by −0.0075. Stopping at the first head changes the experiment and no longer observes the original four-trial head count. A reviewable response includes the random mechanism, fixed block size, repetition count, event numerator and denominator, complete empirical distribution, exact comparison and limits. A supplied report assesses interpretation, while carrying out a new simulation supplies additional execution evidence; neither is independent certification.
This action addresses current AP Statistics identifier 2.10.D (Fall 2026 CED), checked 30 October 2026: official framework.
Report binomial standard deviation, not variance in disguise
For X counting successes in 25 independent trials with common success probability 0.36, E(X)=25·0.36=9 successes and Var(X)=25·0.36·0.64=5.76 squared successes. Therefore SD(X)=√5.76=2.4 successes. Across repetitions of the complete 25-trial experiment, nine is the mean count and 2.4 is the root-mean-square deviation about that mean. Neither says every run has exactly nine successes or lies within 2.4 of nine.
Changed-case check. Forty independent components each fail with probability 0.10. Report the mean, variance and standard deviation of the failed-component count, including units. Explain which number would be wrongly reported if the square root were omitted.
Feedback
The mean is 4 failed components, the variance is 40·0.10·0.90=3.6 squared components, and the standard deviation is √3.6≈1.897 components. Reporting 3.6 components as the SD confuses variance with its square root. The model describes variation across batches with the stated fixed size, common probability and independence; it is not a guarantee for a particular batch. A shared failure cause would require revisiting independence before any of these binomial interpretations.
Continue
Next treat the sample mean or proportion itself as a random variable. Repeated-sample distributions connect these probability calculations to statistical uncertainty.
