Elementary Probability (Elementare Wahrscheinlichkeitsrechnung)

45 min · 12 drills

The tasks in this topic describe a random experiment in a few lines: a draw from a container, a roll of a die, a test that reacts or does not, a game with a stake and a payout. Five claims about probabilities follow, and each one is decided on its own. There is more calculation than in logic, but it is rarely long. Most claims fail on a mix-up: "and" with "or", "exactly one" with "at least one", P(A∣B) with P(B∣A). Once you know where these mix-ups sit, one line of calculation often settles the claim.

Of the twelve formulas on the formula sheet, exactly one helps here: the binomial coefficient (nk)=n!k!(n−k)!, which counts selections. No rule of probability is printed there, neither the addition rule nor the path rules, no formula for conditional probabilities (German: bedingte Wahrscheinlichkeit) and none for the expected value either. If a task needs the expected value (German: Erwartungswert), it prints the formula in the task itself; everything else you bring along. This article therefore derives the rules from a single idea, probability as a proportion, so that you can rebuild them yourself in the exam.

A probability is a proportion when all outcomes are equally likely

Outcome

An outcome is one possible result of a random experiment. For a roll of a die these are the numbers 1 to 6; if you draw a card from a deck, every single card is an outcome. The set of all outcomes is called the sample space.

Event

An event (German: Ereignis) groups outcomes together; it is a subset of the sample space. "Even number" is the event {2,4,6} for a die. An event occurs when one of its outcomes occurs.

Laplace experiment

A random experiment is called a Laplace experiment if all outcomes have the same probability. Only then does the following hold:

P(A)=number of outcomes favourable to Anumber of all possible outcomes

The result depends on this condition. A bag holds 10 chips: 6 white, 3 black and one gold. There are three colours, but what is equally likely is not the colours, it is the chips. Gold therefore has probability 110 and not 13, white has 610. Whoever counts colours is counting outcomes that differ in probability.

The same trap sits in every sum. Roll two tetrahedral dice, each showing one of the numbers 1 to 4. The 4⋅4=16 ordered pairs are equally likely, and (1,3) and (3,1) are two different outcomes. The sums 2 to 8 are not equally likely, because they arise from different numbers of pairs.

SumPairsProbability
2(1,1)116
3(1,2), (2,1)216
4(1,3), (2,2), (3,1)316
5(1,4), (2,3), (3,2), (4,1)416
6(2,4), (3,3), (4,2)316
7(3,4), (4,3)216
8(4,4)116

The middle sum is four times as likely as each of the two extreme ones. When you have to decide whether two events are equally likely, count pairs. A double, meaning the same number twice, comes from 4 pairs, the sum 6 from only 3: the double is more likely, although it sounds like the stricter condition.

Complementary event

The complementary event (German: Gegenereignis) Aˉ occurs exactly when A does not occur. Since one of the two cases always occurs and never both at once:

P(Aˉ)=1−P(A)

The complement is the shorter route as soon as a claim contains "not" or "at least". That at least one four appears on the tetrahedral dice could be counted off the pairs. The opposite is faster: no four means that both dice show one of the three other numbers, which gives 3⋅3=9 pairs. So P(at least one four)=1−916=716.

Your turn1 of 12

Two fair coins are tossed. With what probability does exactly one of them show heads?

Mutually exclusive and independent are two different notions

Mutually exclusive events

Two events A and B are called mutually exclusive (German: unvereinbar) or disjoint if they cannot occur together. Their intersection A∩B is empty and has probability zero. For a die, "number 1" and "even number" are mutually exclusive.

For the probability that A or B occurs, add both probabilities and subtract what you counted twice: the outcomes belonging to both events are contained in P(A) and once more in P(B).

P(A∪B)=P(A)+P(B)−P(A∩B)

For mutually exclusive events the subtracted part is zero and the plain sum remains: for mutually exclusive events with P(A)=0.15 and P(B)=0.45, P(A∪B)=0.6. The same formula gives a quick check in the opposite direction. Since no probability can exceed 1, two events with P(A)+P(B)>1 must overlap. With 0.65 and 0.45 the intersection is at least 0.1, and the two cannot be mutually exclusive.

Independent events

Two events are called independent (German: unabhängig) if the occurrence of one does not change the probability of the other. In terms of calculation this is equivalent to the product rule:

P(A∩B)=P(A)⋅P(B)

Now the difference can be calculated. If A and B are mutually exclusive and both have a positive probability, the intersection is empty, P(A∩B)=0, while the product P(A)⋅P(B) is positive. Mutually exclusive events are therefore dependent, and as strongly as possible: whoever learns that B has occurred knows that A can no longer occur. With the numbers above, independence would require 0.15⋅0.45=0.0675, but the intersection is empty.

mutually exclusiveindependent
meaningcannot occur togetherdo not influence each other
intersectionP(A∩B)=0P(A∩B)=P(A)⋅P(B)
"or"P(A)+P(B)P(A)+P(B)−P(A)⋅P(B)
recognisedfrom the description of the experimentfrom the task or by calculation

Your turn2 of 12

For two mutually exclusive events, P(A)=0.3 and P(B)=0.4. Which statement is correct?

A two-way table arranges two characteristics and their proportions

When two characteristics are observed in the same people, the numbers belong in a table with four inner cells, a two-way table (German: Vierfeldertafel). Of 400 respondents, 220 have a streaming subscription (S), 120 are members of a gym (F), and 48 have both. That gives one margin total per characteristic and one inner cell. All other cells follow by completing rows and columns to their totals.

FFˉtotal
S48172220
Sˉ72108180
total120280400

Every phrasing in the exam corresponds to one cell or a sum of cells.

  • "S and F" is the inner cell 48, so 48400=0.12.
  • "S or F" includes the people with both: 48+172+72=292, so 0.73. The addition rule gives the same, 0.55+0.3−0.12=0.73.
  • "only S" means S without F: 172, so 0.43. The word "only" removes the intersection.
  • "neither S nor F" is the cell 108, so 0.27, and at the same time 1−0.73.
  • "exactly one of the two" gives 172+72=244, so 0.61.

Whether the two characteristics are independent is decided by comparing the intersection with the product of the margin proportions: 0.55⋅0.3=0.165, but the intersection is 0.12. The characteristics are dependent. They would be independent only with 0.165⋅400=66 people having both. An intersection that differs from the product is not a rounding detail; it already is the verdict.

Your turn3 of 12

Of 500 students, 300 speak English fluently (E), 150 speak French (F), and 90 speak both. What share speaks exactly one of the two languages?

In a tree diagram you multiply along a path and add across paths

If an experiment runs in stages, draw a branching for each stage and write on each branch the probability with which it is taken, given what has happened up to that point. The branches leaving one point add up to 1.

Path rules

The probability of a path is the product of the probabilities along that path. The probability of an event is the sum of the probabilities of all paths that belong to it.

A drawer holds 11 batteries, 4 of which are flat. You take out two without putting the first one back. At the first stage a flat battery comes up with 411. At the second stage it depends on what is already missing: if the first was flat, 3 flat ones remain among 10, otherwise 4 flat ones among 10.

PathCalculationProbability
flat, flat411⋅31012110
flat, full411⋅71028110
full, flat711⋅41028110
full, full711⋅61042110

The four paths add up to 110110=1, and that is the first check of every calculation. Both being flat has probability 12110=655.

One question regularly causes confusion here: how likely is the second battery to be flat? The second stage shows 310 and 410, and both are conditional probabilities that take the first draw as known. Without that knowledge you add the two paths ending in "flat": 12110+28110=40110=411. This is the same number as for the first battery, and it has to be. Before anyone looks, every battery in the drawer has the same chance of being drawn second as of being drawn first.

Drawing with and without replacement

When drawing without replacement (German: ohne Zurücklegen), what is drawn stays out. The probabilities of later stages change with the result of earlier ones, so the draws are dependent. When drawing with replacement, what is drawn goes back, every stage sees the same contents, and the draws are independent.

With replacement both batteries would be flat with 411⋅411=16121≈0.132, without replacement only with 655≈0.109. Drawing the same kind twice becomes rarer without replacement, because the first draw has already shrunk the stock of that kind. The calculation 411⋅411 for the case without replacement is the most frequent mistake of this section.

Your turn4 of 12

A box holds 6 tickets, 2 of them winners. Two people draw one ticket each, one after the other, without replacement. With what probability does the second person win, as long as nobody knows the first ticket?

Independent events are multiplied, and "at least one" goes through the opposite

Two bus lines run late independently of each other, line 1 with probability 0.4 and line 2 with 0.25. The table shows the four possible cases. Every row is a product because the events are independent.

Line 1Line 2Probability
latelate0.4⋅0.25=0.1
lateon time0.4⋅0.75=0.3
on timelate0.6⋅0.25=0.15
on timeon time0.6⋅0.75=0.45
  • Both late: 0.1.
  • Exactly one late: two rows, 0.3+0.15=0.45.
  • At least one late: everything except the last row, 1−0.45=0.55. That exceeds "exactly one" by the case with two delays.
  • The sum 0.4+0.25=0.65 answers none of these questions. It counts the case with two delays twice, and with more events or larger probabilities it even exceeds 1.

Pay attention to what the opposite of an event really is. The opposite of "both late" is "at least one on time", not "both on time". The opposite of "none late" is "at least one late", not "all late". Two events are complementary only if together they cover every case and none twice.

With more than two events the advantage of the opposite becomes clearer. Three flights are each late with 0.3, independently of each other. That none is late has probability 0.73=0.343, so at least one is late with 1−0.343=0.657. Exactly one, on the other hand, needs three paths, depending on which flight is late: 3⋅0.3⋅0.72=0.441.

Gambler's fallacy

The gambler's fallacy is the belief that a random device evens out earlier results, for instance that after a run of heads, tails is now "due". With independent throws the device has no memory. Every throw has the same probabilities as the first.

After four throws without a six, the fifth throw of a fair die shows a six with 16, just like any other throw. The impression that something is due mixes up two questions. A particular sequence of three coin tosses, such as tails, tails, tails, has probability (12)3=18, and the mixed-looking sequence heads, tails, heads also has 18. A difference only appears once you ask about the count. "Two heads" has three sequences (HHT, HTH, THH), hence 38, while "three heads" has only one.

Over many throws, too, chance promises no exact balancing. With very many throws the proportion of heads lies close to 12, but the count rarely hits exactly half. With 1 000 throws, exactly 500 heads is the most likely count and still unlikely, about 2.5%, because the probability is spread over many neighbouring counts.

Your turn5 of 12

Three applications are each accepted with probability 0.2, independently of each other. How likely is at least one acceptance?

With a conditional probability the reference group changes

Conditional probability

The conditional probability P(A∣B), read "A given B", is the probability of A when it is known that B has occurred. Counting now happens only inside B:

P(A∣B)=P(A∩B)P(B)

In the two-way table this means that the denominator is no longer the grand total but the row or column of the condition. Among the 120 gym members, 48 have a streaming subscription, so P(S∣F)=48120=0.4. Among the 220 subscribers, 48 are members, so P(F∣S)=48220≈0.22. Same intersection, two denominators, two different numbers. The complement works as a condition too: among the 280 people without a membership, 172 have a subscription, P(S∣Fˉ)=172280≈0.61.

Solving the definition for the intersection gives the product rule of the tree diagram in its general form: P(A∩B)=P(B)⋅P(A∣B). The product of the unconditional probabilities P(A)⋅P(B) is the special case of independent events, for which P(A∣B)=P(A) holds.

An abstract example shows how much follows from three pieces of information. Let P(A)=0.25, P(B)=0.4 and P(A∩B)=0.15.

  • P(A∣B)=0.150.4=0.375. Knowing B lifts A from 0.25 to 0.375.
  • P(B∣A)=0.150.25=0.6. The reversed condition has a different denominator and a different value.
  • P(A∪B)=0.25+0.4−0.15=0.5.
  • A and B are not independent, because 0.25⋅0.4=0.1 differs from 0.15. Equivalently, P(A∣B) differs from P(A). Whether the conditional probability turns out larger or smaller makes no difference: any change means dependence.

Your turn6 of 12

Of 50 customers, 30 pay cash and 20 by card. 12 of the cash payers and 4 of the card payers buy a special offer. With what probability does a buyer of the offer pay cash?

Total probability is a weighted average

At a university, 75% study full-time and 25% part-time. Of the full-time students, 4% drop out in the first year, of the part-time students 12%. In the tree, two paths lead to "drop-out":

P(drop-out)=0.75⋅0.04+0.25⋅0.12=0.03+0.03=0.06
Law of total probability

If the experiment splits into cases B1,B2,… that exclude each other and together cover everything, then P(A) is the sum of the paths P(Bi)⋅P(A∣Bi) over all cases. This is an average of the conditional probabilities, weighted by the shares of the cases.

The simple average of 4% and 12%, that is 8%, would only be right if both groups were the same size. Because the full-time group is three times as large, it pulls the overall value towards its own rate. Also watch what a rate refers to: the 12% apply within the part-time group. Relative to all students, part-time students who drop out are 0.25⋅0.12=0.03, that is 3%.

Your turn7 of 12

An insurer has 80% careful customers with an accident rate of 1% and 20% risk-taking customers with an accident rate of 9%. What share of all customers has an accident?

A condition is most safely reversed with natural frequencies

A screening system flags suspicious card payments. 1% of all payments are fraud. The system flags 98% of the fraudulent payments and, by mistake, 3% of the genuine ones. The question is how likely a flagged payment really is fraud, that is P(fraud∣flagged). The information supplies the reverse direction, P(flagged∣fraud).

Instead of using formulas, calculate with an imagined 10 000 payments. Of these, 100 are fraud and 9 900 are genuine. Of the 100 the system flags 98, of the 9 900 it flags 297.

flaggednot flaggedtotal
fraud982100
genuine2979 6039 900
total3959 60510 000

Of the 395 flagged payments, 98 are fraud: P(fraud∣flagged)=98395≈0.25. The 98% describe the row "fraud", the probability asked for describes the column "flagged", and the two numbers lie far apart. The reason is that fraud is rare: 3% of very many genuine payments is more than 98% of a few fraudulent ones. Medical tests follow the same pattern, with people who are ill in place of fraud cases.

Your turn8 of 12

A disease affects 2% of a group. A test detects 90% of those who are ill and falsely reacts in 5% of those who are healthy. How likely is a person with a positive test actually ill?

When counting, it matters whether the order counts

For Laplace probabilities you have to count outcomes, and with selections from a group there are soon too many to write down. From 12 participants in a seminar, 4 are drawn by lot for an interview.

Factorial

n!, read "n factorial", is the product 1⋅2⋅…⋅n, and in addition 0!=1 is defined. It counts the orders into which n different things can be arranged: for 4 people that is 4!=24.

If order matters, for instance because the four interviews take place on four different days, there are 12 possibilities for the first slot, 11 for the second, and so on:

12⋅11⋅10⋅9=12!8!=11 880

In general there are n!(n−k)! ordered selections of k from n. If only who is in matters, each group of four appears 4!=24 times in this count, once for each order. Dividing by 24 gives 495 groups, and that is exactly what the binomial coefficient from the formula sheet delivers:

(124)=12!4!⋅8!=11 88024=495
Binomial coefficient

(nk), read "n choose k", is the number of ways to choose k of n different things without regard to order. It holds that (nk)=(nn−k), because every choice of k uniquely determines the n−k not chosen.

For the seminar the symmetry means that there are as many ways to choose 4 of 12 as to leave 8 of 12 behind. The number 124=20 736, by contrast, counts something else, namely selections in which the same person may appear more than once and order counts. For a group of people it is never right.

SituationNumberIn the example
with order, without repetitionn!(n−k)!11 880
without order, without repetition(nk)495
with order, with repetitionnk20 736

The same coefficient also counts sequences of coin tosses. If exactly three of five tosses are to show heads, you choose which 3 of the 5 toss numbers carry heads: that gives (53)=10 sequences, each with probability (12)5, so 1032 altogether.

How likely is a particular person, say Ms Berger, to be among those drawn? The groups containing her consist of her and 3 of the remaining 11, which makes (113)=165 groups, and 165495=13. It is faster via the places: 4 places for 12 people treated alike give each individual 412. The reciprocal 1495 would be the probability of one particular complete group, not of one particular person.

Binomial coefficients also count compositions. Of the 12 participants, 7 are students and 5 are in employment. Exactly 2 students among the 4 drawn requires 2 of the 7 students and 2 of the 5 employed. Every choice of one part can be combined with every choice of the other, which is why you multiply:

P(exactly 2 students)=(72)⋅(52)(124)=21⋅10495=210495≈0.42

Only students are drawn with (74)495=35495=799. The tree confirms this: 712⋅611⋅510⋅49=84011 880=799. Wrong would be (712)4≈0.12, because that calculation draws with replacement, as if the same person could be drawn twice. For each single place, however, the share does apply: each of the four people drawn is, taken on their own, a student with 712, for the same reason as with the second battery.

Your turn9 of 12

From 15 members of a club, 5 are drawn by lot for a committee. With what probability are two particular members both on it?

The expected value is the average over many repetitions

Random variable

A random variable X assigns a number to every outcome of a random experiment: a payout, a count, a profit. P(X=x) is the probability that it takes the value x. The probabilities of all possible values add up to 1, because every result delivers exactly one of these values.

Expected value

The expected value E(X) is the sum of all possible values, each weighted by its probability:

E(X)=∑xx⋅P(X=x)=x1⋅P(X=x1)+x2⋅P(X=x2)+…

This formula is not on the formula sheet. Tasks that need it print it in the task, usually in the short form with the summation sign. Read the summation sign as a work instruction: write down every possible value, multiply it by its probability, add everything up. The value zero contributes nothing, but it still counts in the check that the probabilities add up to 1.

An online shop counts how many add-on items an order contains. The random variable X has this distribution:

x0124
P(X=x)0.30.40.20.1
x⋅P(X=x)00.40.40.4

The probabilities add up to 1, and the sum of the last row is E(X)=1.2. From this you can read what the value means and what it does not.

  • Scaling up: with 50 orders, 50⋅1.2=60 add-on items are expected on average.
  • Not a possible value: no single order contains 1.2 items. The expected value is an average, not a result.
  • Not the simple mean of the values: 0+1+2+44=1.75 ignores that the values differ in probability.
  • A different question: at least one add-on item is included with 1−0.3=0.7. That is a probability, not an expected value, calculated via the complement and not by weighting.

Your turn10 of 12

A ticket wins exactly 20 euros with probability 0.05, exactly 5 euros with 0.15, and nothing otherwise. How much does a ticket pay out on average?

A fair stake balances the expected value, not the single game

A wheel of fortune has 10 equal sectors. One sector pays out 20 euros, three sectors pay 5 euros each, the other six pay nothing. A spin costs a stake of 4 euros.

E(payout)=20⋅0.1+5⋅0.3+0⋅0.6=2+1.5=3.5

On average, then, 3.50 euros flow back, and the player's expected profit is 3.5−4=−0.5, a loss of 0.50 euros per spin, which the operator books as income on average.

Fair game

A game is called fair if the expected profit is zero, that is, if the stake equals the expected payout exactly. "Roughly equal" is not enough: a small difference per game adds up to a large amount over many games.

The wheel would be fair with a stake of 3.50 euros. The expected value can also be shifted through the probabilities. If the wheel had only 8 instead of 10 sectors with the same winning sectors, the expected payout would rise to 20⋅18+5⋅38=2.5+1.875=4.375 euros, and with a stake of 4 euros the player would come out ahead on average.

Because the expected value is an average, it does not decide every choice. A small business needs a component and has two offers. Supplier S delivers for certain and costs 5 200 euros. Supplier R costs 4 000 euros but fails with probability 0.1, and then the replacement costs an extra 10 000 euros.

  • The expected cost with R is 4 000+0.1⋅10 000=5 000 euros, that is 200 euros less than with S.
  • In fact the business never pays 5 000 euros with R, but either 4 000 or 14 000 euros. Here too the expected value is not a possible result.
  • Whoever places many such orders saves with R in the long run. Whoever orders only once and could not bear 14 000 euros has good reason to choose S, paying 200 euros on average for certainty. The expected value measures the middle of the results, not how far apart they lie.

Your turn11 of 12

A game costs a stake of 3 euros. With probability 0.1 it pays out 15 euros, with probability 0.25 it pays out 4 euros, otherwise nothing. Which statement is correct?

Five claims under time pressure

Each of the five claims is decided on its own, and the task does not reveal how many are true. Scoring uses partial points, as described in the article on logic: if more than one claim is correct, every correct tick earns a share and every wrong tick subtracts a share, without the question falling below zero; if only one is correct, only exactly that one tick counts, and any further tick sets the question to zero. A tick on a hunch can therefore cost points you had already secured; tick what you can justify. A procedure that works:

  1. Break down the experiment. Which outcomes are equally likely? Is the drawing with or without replacement? Are events stated as independent or as mutually exclusive?
  2. Set up a table or a tree. Two characteristics belong in a two-way table, stages in a tree. Either takes a minute and usually answers all five claims at once.
  3. Translate the wording into cells. "and", "or", "only", "exactly one", "at least one", "neither nor": each phrase means a particular cell or a particular sum of paths. For "at least", 1−P(none) is almost always the shortest route.
  4. Find the reference group. If "among", "of the" or "if" appears in the claim, the denominator changes. Ask of every percentage what it refers to.
  5. Check the justification too. If a claim contains a "because", it is only true if the justification holds as well.

A complete example. Let A and B be two independent events with P(A)=0.6 and P(B)=0.2.

  • P(A∩B)=0.12. True. For independent events the intersection is the product, 0.6⋅0.2=0.12.
  • P(A∪B)=0.8. False. The sum counts the intersection twice; correct is 0.6+0.2−0.12=0.68.
  • A and B are mutually exclusive. False. That would require an empty intersection, but it has probability 0.12.
  • P(A∣B)=0.6. True. Independence means that B does not change the probability of A; as a check, 0.120.2=0.6.
  • Neither A nor B occurs with probability 0.32. True. This is the complement of "A or B", so 1−0.68=0.32. Likewise 0.4⋅0.8=0.32, because together with A and B their complements are independent too.

All five decisions hinge on the word "independent" in the assumption. Had it said "mutually exclusive", the intersection would be empty and the union 0.8: then the second and third claims would be true and the other three false.

Your turn12 of 12

Of 200 applicants, 120 pass an entrance test. 90 of those who passed and 20 of the rest attended a preparation course beforehand. B stands for passing, K for attending the course, and an applicant is picked at random. Which claim is true?

Common mistakes

  • Counting outcomes that are not equally likely. Colours, sums of dice or numbers of heads are not Laplace outcomes. What you count are single chips, ordered pairs, single sequences.
  • Confusing mutually exclusive with independent. Mutually exclusive events with positive probability are dependent, because one rules out the other.
  • Not subtracting the intersection for "or". The plain sum holds only for mutually exclusive events.
  • Using constant probabilities without replacement. The second stage carries conditional probabilities that depend on the first draw.
  • Choosing the wrong complement. The opposite of "none" is "at least one", not "all".
  • Swapping P(A∣B) and P(B∣A). The hit rate of a test is not the probability that a hit is correct.
  • Averaging without weights. Rates from groups of different sizes are weighted by the group shares.
  • Reading a rate without its reference group. A rate that applies within a group becomes a share of everyone only after multiplying by the group share.
  • Confusing the chance of winning with the expected value. How often a game pays out says nothing about its advantage without the amounts. The stake is subtracted only from the expected payout.
  • Taking the expected value for the result of a single trial. It describes the average over many repetitions; a single trial usually lands somewhere else.

Key takeaways

  • Laplace: favourable over possible outcomes, but only if all outcomes are equally likely. The complement has 1−P(A).
  • Or: P(A∪B)=P(A)+P(B)−P(A∩B). The subtraction drops out only for mutually exclusive events.
  • And: P(A∩B)=P(B)⋅P(A∣B), for independent events P(A)⋅P(B).
  • Tree: multiply along a path, add across paths. Without replacement the branches of later stages change.
  • Conditional: P(A∣B)=P(A∩B)P(B), the condition sits in the denominator. Reverse it with natural frequencies.
  • Counting: n!(n−k)! with order, (nk) without. A particular person is among k drawn from n with kn.
  • Expected value: each value times its probability, all added up. It is the average over many repetitions, and fair means expected profit zero.

Now practise it

Elementary Probability (Elementare Wahrscheinlichkeitsrechnung)

Trainer · Mathematics

Practise
Next: Binomial Distribution (Binomialverteilung)
0/12

Was this article helpful?