15 Common Mistakes When Studying Probability (And How to Fix Them) | LearnByTeaching.ai
Probability is deceptively hard because human intuition about randomness is reliably wrong. The birthday paradox, the Monty Hall problem, and base rate neglect all demonstrate that our natural sense of likelihood leads us astray. Succeeding in probability requires replacing intuition with rigorous calculation and visual representation.
Confusing P(A|B) with P(B|A)
Conditional probability is the most common stumbling block. Students reverse the conditioning — treating the probability of A given B as if it were the probability of B given A — which can give wildly different values.
A student is told that 95% of people with a disease test positive. When asked for the probability that someone who tested positive has the disease, they answer 95%, ignoring that with a rare disease and false positives, the actual probability could be much lower (the base rate fallacy).
How to fix it
Always write out the conditioning explicitly: P(disease|positive) is NOT P(positive|disease). Use Bayes' theorem to convert between them. Draw a tree diagram or a 2x2 frequency table with actual numbers to see how base rates affect the answer.
Assuming events are independent without verification
Students default to multiplying probabilities as if events are independent, even when they are not. Independence is a specific mathematical property that must be verified, not assumed.
A student calculates the probability that two cards drawn from a deck are both aces as (4/52)(4/52) = 1/169, treating the draws as independent. Without replacement, the correct answer is (4/52)(3/51) = 1/221.
How to fix it
Before multiplying probabilities, explicitly ask: does the outcome of the first event change the probability of the second? If the answer is yes (or if you are not sure), use the multiplication rule for dependent events: P(A and B) = P(A) * P(B|A). Independence means P(B|A) = P(B).
Confusing permutations and combinations
Students struggle with whether order matters in a counting problem. Using permutations when combinations are appropriate (or vice versa) gives an answer that is off by a factor of n!.
A student asked to count the number of 5-card poker hands from a 52-card deck uses P(52,5) = 311,875,200 instead of C(52,5) = 2,598,960, overcounting by a factor of 5! = 120 because they treated the order of cards as meaningful.
How to fix it
Always ask: does the order of selection matter? If selecting a committee (order doesn't matter), use combinations. If arranging people in a line (order matters), use permutations. When in doubt, describe what constitutes a 'different' outcome — if rearranging the same items gives the same outcome, use combinations.
Not using tree diagrams for conditional probability
Students try to solve conditional probability problems purely algebraically, without the visual aid that tree diagrams provide. This leads to errors in identifying the correct sample space and computing the right probabilities.
A student attempts a two-stage probability problem (rain/no rain, then drive/bus) algebraically and confuses the joint and conditional probabilities. A tree diagram would show each path with its probability and make the calculation clear.
How to fix it
Draw a tree diagram for every problem involving sequential events or conditional probabilities. Label each branch with its probability, multiply along paths for joint probabilities, and add across relevant paths for total probability. This visual approach prevents the algebraic confusion that plagues conditional probability.
Misunderstanding the law of large numbers
Students believe that the law of large numbers means random processes must 'even out' in the short run — that a coin showing five heads in a row is 'due' for tails. This is the gambler's fallacy, and it contradicts what the law actually states.
A student argues that after flipping 10 heads in a row, the next flip is more likely to be tails 'because the probability should average to 50%.' The coin has no memory; each flip is independent with P(tails) = 0.5.
How to fix it
Understand what the law of large numbers actually says: the proportion of heads approaches 0.5 as the number of flips approaches infinity. It says nothing about short-run corrections. Future flips do not compensate for past flips. The law is about long-run averages, not individual outcomes.
Double-counting in probability calculations
When events overlap, adding their individual probabilities counts the intersection twice. Students who forget to subtract the overlap using inclusion-exclusion get probabilities that are too high — sometimes exceeding 1.
A student calculates P(drawing a heart or a king) as P(heart) + P(king) = 13/52 + 4/52 = 17/52, forgetting to subtract the king of hearts (counted in both). The correct answer is 13/52 + 4/52 - 1/52 = 16/52.
How to fix it
Always use the inclusion-exclusion principle for 'or' probabilities: P(A or B) = P(A) + P(B) - P(A and B). Draw a Venn diagram to visualize the overlap. If the events are mutually exclusive (no overlap), then P(A and B) = 0 and you can simply add.
Not defining the sample space clearly
Students jump into calculations without explicitly defining the sample space — the set of all possible outcomes. Without a clear sample space, it is impossible to determine whether outcomes are equally likely or to count correctly.
A student rolls two dice and asks for the probability of a sum of 4. They list outcomes as sums (2, 3, 4, ... 12) and claim P(sum=4) = 1/11. But the correct sample space has 36 equally likely ordered pairs, and sum=4 occurs in 3 of them: P = 3/36 = 1/12.
How to fix it
Before any calculation, write out the sample space (or describe it precisely if too large to list). Verify that outcomes are equally likely — if they are not, you cannot use the classical probability formula P = favorable/total. For dice, cards, and coins, the sample space structure determines everything.
Confusing 'at least one' with 'exactly one'
Problems asking for 'at least one success' trip up students who calculate the probability of exactly one success. The complement approach is much easier and avoids this confusion.
A student asked for the probability of getting at least one head in three coin flips calculates the probability of exactly one head: C(3,1)*(1/2)^3 = 3/8. The correct answer using the complement is 1 - P(no heads) = 1 - 1/8 = 7/8.
How to fix it
For 'at least one' problems, always use the complement: P(at least one) = 1 - P(none). This converts a complex problem (one success or two successes or three successes...) into a single, simple calculation. This technique works for any number of trials.
Misapplying the multiplication rule to 'or' situations
Students confuse when to add probabilities (or) and when to multiply (and). They multiply when they should add, or add when they should multiply, because the verbal cues are ambiguous.
A student calculates the probability of rolling a 3 or a 5 on one die as (1/6)(1/6) = 1/36, multiplying instead of adding. The correct answer is 1/6 + 1/6 = 2/6 = 1/3, because these are mutually exclusive events combined with 'or.'
How to fix it
Multiplication is for 'and' (both events occur, often in sequence). Addition is for 'or' (at least one event occurs, typically mutually exclusive). When you see 'or,' think addition (with inclusion-exclusion if events overlap). When you see 'and,' think multiplication (with conditional probability if events are dependent).
Not simulating probability problems to build intuition
Students compute probabilities abstractly without ever running simulations to check their answers. Simulation is the single best way to build intuition for probability and catch calculation errors.
A student computes the probability for the Monty Hall problem and gets 1/2 (the wrong, intuitive answer) but never simulates the game to see that switching wins 2/3 of the time, which would have revealed the error immediately.
How to fix it
Simulate problems before or after solving them analytically. Use Python, a spreadsheet, or even physical coins and dice. Run 1000 trials and compare the simulated frequency to your calculated probability. If they disagree, your calculation is wrong. This practice builds the probabilistic intuition that pure calculation cannot.
Computing expected value incorrectly
Students add possible outcomes without weighting by probability, or weight by probability but forget to include all outcomes (especially zero-value or negative outcomes).
A lottery ticket costs $2 and pays $100 with probability 1/100. A student computes E = (1/100)(100) = $1, forgetting to account for the 99/100 probability of winning nothing. The correct expected value is (1/100)(100) + (99/100)(0) - 2 = -$1.
How to fix it
Write out every outcome with its probability and value. Expected value = sum of (probability x value) for ALL outcomes. Include the cost or losing outcomes. Always check: the expected value should be between the minimum and maximum possible values.
Struggling with counting problems involving restrictions
Counting problems with restrictions (e.g., 'at least two girls,' 'no repeated digits,' 'adjacent seats') require breaking the problem into cases or using complements. Students who try a single formula often miscount.
A student counts the number of 3-digit numbers with no repeated digits by computing 10 * 9 * 8 = 720, forgetting that the first digit cannot be zero, so the correct count is 9 * 9 * 8 = 648.
How to fix it
For restricted counting problems, handle the most restrictive condition first. For digit problems, always address the leading digit (cannot be zero) before other positions. For 'at least' restrictions, consider the complement (total minus none). Break complex restrictions into non-overlapping cases.
Not understanding probability distributions before jumping to formulas
Students plug into binomial, Poisson, or normal distribution formulas without understanding what each distribution models. This leads to using the wrong distribution for a given scenario.
A student applies the binomial distribution to model the number of customers arriving at a store per hour, when the Poisson distribution is appropriate because arrivals are independent events in a continuous time interval, not a fixed number of yes/no trials.
How to fix it
For each distribution, learn the scenario it models: binomial = fixed number of independent yes/no trials; Poisson = events in a fixed interval of time or space; geometric = trials until first success; normal = sum of many independent factors. Match the scenario first, then apply the formula.
Cramming probability before exams instead of practicing consistently
Probability requires building problem-solving patterns over time. Students who cram find that the problems look different from what they practiced because probability questions can be framed in endlessly varied ways.
A student studies the formulas the night before the exam but encounters a word problem on the test that requires recognizing a conditional probability situation embedded in a medical testing scenario, and they cannot identify the structure.
How to fix it
Practice probability problems daily, mixing problem types. Focus on identifying the structure (conditional, independent, permutation/combination, distribution) before calculating. The variety of probability problem framings means you need extensive practice to recognize patterns across different contexts.
Not checking answers against basic probability rules
Students submit answers outside the range [0, 1] or answers that violate basic rules (P(A or B) > P(A) + P(B) for mutually exclusive events) without noticing the error.
A student calculates a probability of 1.3 and does not recognize this as impossible, or computes P(A and B) > P(A), which violates the fact that the intersection cannot be more likely than either event alone.
How to fix it
Always check: Is my answer between 0 and 1? Is P(A and B) <= min(P(A), P(B))? Is P(A or B) >= max(P(A), P(B))? Does the sum of probabilities in a complete partition equal 1? These sanity checks catch arithmetic errors and conceptual mistakes before you submit.
Quick Self-Check
- Can you explain why P(disease|positive test) is not the same as P(positive test|disease) and use Bayes' theorem to compute the first from the second?
- Can you determine whether a counting problem requires permutations or combinations by asking whether order matters?
- Can you solve an 'at least one' probability problem using the complement method?
- Can you draw a tree diagram for a two-stage conditional probability problem and correctly compute joint and marginal probabilities?
- Can you explain why the gambler's fallacy is wrong, even though the law of large numbers guarantees long-run convergence?
Pro Tips
- ✓Draw tree diagrams for every conditional probability problem — visual representation prevents the P(A|B) vs. P(B|A) confusion that is the most common probability error.
- ✓For 'at least one' problems, always use the complement: P(at least one) = 1 - P(none). This shortcut is cleaner, faster, and less error-prone than computing each case.
- ✓Simulate probability problems in code before or after solving them analytically — if your simulation gives a different answer from your formula, the formula is wrong.
- ✓Before choosing a formula, identify the structure: is this conditional? Independent? A counting problem? Matching the scenario to the method is more important than computation speed.
- ✓Always sanity-check your answer: probability must be between 0 and 1, and P(A and B) can never exceed P(A) or P(B).