Mea­sur­ing the uncer­tain

Will it rain today? Will our school win the match? Nobody knows for cer­tain. Still, some out­comes are clearly more likely than oth­ers, and prob­a­bil­ity is how we mea­sure that like­li­hood, on a scale from 00 to 11. In this les­son we will say what ran­dom­ness means and set up the scale. Then we look at the two hon­est ways of putting a num­ber on like­li­hood. One is to count what hap­pened in repeated tri­als. The other is to rea­son about out­comes that are equally likely.

Ran­dom­ness

A ran­dom exper­i­ment is some­thing you can repeat, such as toss­ing a coin, rolling a die or draw­ing a slip from a box. You know all the pos­si­ble results in advance, but you do not know the result of any one trial. You can say what could hap­pen, but never what will.

That unpre­dictabil­ity is actu­ally use­ful. Think of the toss before a cricket match. It decides who bats first because nei­ther cap­tain can influ­ence it. It is fair pre­cisely because it is ran­dom.

Rain is ran­dom in the same way. It depends on so many fac­tors, and so del­i­cately, that no one can be cer­tain about a par­tic­u­lar day. Even so, the like­li­hood can be esti­mated from pat­terns.

The prob­a­bil­ity scale

Prob­a­bil­ity runs from 00, which means impos­si­ble, to 11, which means cer­tain. In the mid­dle, 0.50.5 means an even chance.

Prob­a­bil­ity Mean­ing Exam­ple
00 impos­si­ble rolling a 77 on an ordi­nary die
near 00 unlikely rolling a 33 (one face in six)
0.50.5 even chance a fair coin show­ing heads
near 11 likely draw­ing a num­ber card (22 to 1010) from a pack: 3636 of 5252
11 cer­tain draw­ing a red sweet from a bag of only red sweets
Probability scale from 0 (impossible) through 0.5 (even chance) to 1 (certain), marking rolling a 7 at 0, rolling a 3 at 1/6, heads at 0.5, a number card at 36/52 and a red sweet at 1.
The exam­ples from the table placed on the scale from 0 to 1.

You can say the same thing with per­cent­ages, so 0.750.75 is a 75%75\% chance. And remem­ber, a prob­a­bil­ity can never be neg­a­tive, and it can never be more than 11.

Exper­i­men­tal prob­a­bil­ity

The first way is sim­ple: repeat an exper­i­ment many times, and count.

Experimental probability of an event=number of trials in which it occurredtotal number of trials\displaystyle \text{Experimental probability of an event} = \frac{\text{number of trials in which it occurred}}{\text{total number of trials}}

This frac­tion is also called the rel­a­tive fre­quency of the event.

Exam­ple 1. A die is rolled 6060 times and shows a 55 on 99 of them. So the exper­i­men­tal prob­a­bil­ity of a 55 is 960=0.15\displaystyle \dfrac{9}{60} = 0.15.

Exam­ple 2. A draw­ing pin is tossed 200200 times and lands point-up 7676 times. Our esti­mate of the prob­a­bil­ity of point-up is 76200=0.38\displaystyle \dfrac{76}{200} = 0.38. Could we have worked this out just by think­ing? No. The pin is not sym­met­ric, so exper­i­ment is the only way.

The­o­ret­i­cal prob­a­bil­ity

The sec­ond way needs no exper­i­ment at all. It works when every out­come is equally likely, as with a fair coin, a fair die or a well-shuf­fled pack.

P(event)=number of favourable outcomesnumber of possible outcomes\displaystyle P(\text{event}) = \frac{\text{number of favourable outcomes}}{\text{number of possible outcomes}}

Exam­ple 3. For a fair die, P(5)=16≈0.167\displaystyle P(5) = \dfrac{1}{6} \approx 0.167, and P(even)=36=12\displaystyle P(\text{even}) = \dfrac{3}{6} = \dfrac{1}{2}.

Exam­ple 4. A let­ter is picked at ran­dom from MATH­E­MAT­ICS. There are eleven let­ters, and AA appears twice, as does TT. So P(A)=211\displaystyle P(A) = \dfrac{2}{11}, P(vowel)=411\displaystyle P(\text{vowel}) = \dfrac{4}{11} (A, E, A, I).

Exam­ple 5. A bag has 44 red and 99 blue mar­bles. P(red)=413\displaystyle P(\text{red}) = \dfrac{4}{13}. Notice that the out­comes "red" and "blue" are not equally likely. It is the thir­teen mar­bles that are equally likely, and that is what we count.

Exper­i­ment ver­sus the­ory

Exam­ple 6. A die is rolled 1212 times and shows a 33 four times. The exper­i­men­tal prob­a­bil­ity is 412=13\displaystyle \dfrac{4}{12} = \dfrac{1}{3}, but the the­o­ret­i­cal one is 16\displaystyle \dfrac{1}{6}. Why the dif­fer­ence? Twelve tri­als are just too few. With 600600 or 60006000 rolls, we would expect the exper­i­men­tal value to set­tle close to 16\displaystyle \dfrac{1}{6}. This ten­dency is called the law of large num­bers.

Prob­a­bil­ity from data and sam­ples

Data that has already been col­lected works like an exper­i­ment that has already been done.

Exam­ple 7. In a class of 4040, 1616 pre­fer mango, 1212 apple, 88 banana, 44 guava. For a stu­dent picked at ran­dom from the class, P(mango)=1640=0.4\displaystyle P(\text{mango}) = \dfrac{16}{40} = 0.4.

Bar chart of fruit preferences in a class of 40 students: mango 16, apple 12, banana 8 and guava 4; the mango bar is 16 out of 40.
Exam­ple 7: mango is pre­ferred by 16 of the 40 stu­dents.

Sup­pose the school has 12001200 stu­dents and this class is a fair sam­ple of the school. Then about 0.4×1200=4800.4 \times 1200 = 480 stu­dents pre­fer mango. A larger and more var­ied sam­ple, with stu­dents from sev­eral classes instead of one, gives an esti­mate you can trust more. Choos­ing the sam­ple well is called sam­pling.

Exam­ple 8. A fac­tory tests 500500 bulbs and 1515 fail. So the esti­mated prob­a­bil­ity that a bulb fails is 0.030.03. In a batch of 2000020000, we would expect roughly 600600 fail­ures.

The gam­bler's fal­lacy

A fair coin shows heads six times run­ning. Is tails now "due"? Many peo­ple feel it must be, but no. The coin has no mem­ory. The next toss is 12\displaystyle \dfrac{1}{2} heads and 12\displaystyle \dfrac{1}{2} tails, exactly as the first one was. Believ­ing that a run of one out­come makes the other more likely is called the gam­bler's fal­lacy. Prob­a­bil­ity describes the long run, not the very next trial.

Two words before we prac­tise. A fair or unbi­ased coin or die is sym­met­ric enough that no face is favoured. And "at ran­dom" means the draw is made with­out look­ing or inter­fer­ing.

Prac­tice

  1. Place each event on the scale, impos­si­ble, unlikely, even, likely, cer­tain, with a rea­son: (a) the sun rises tomor­row; (b) a fair die shows a num­ber less than 66; (c) a card drawn from a pack is a spade; (d) it snows in Chen­nai in May; (e) a fair coin shows tails.
  2. A spin­ner is spun 8080 times and stops on green 1818 times. Find the exper­i­men­tal prob­a­bil­ity of green.
  3. A sam­ple of 5050 sweets from a large jar has 1414 orange, 2020 lemon, 1616 mint. (a) Find the prob­a­bil­ity that a sweet from the sam­ple is lemon. (b) Esti­mate how many of the jar's 900900 sweets are mint.
  4. A let­ter is cho­sen at ran­dom from PROB­A­BIL­ITY. Find the prob­a­bil­ity of (a) B; (b) a vowel; (c) a let­ter that appears exactly once.
  5. A fair die is rolled. Find the prob­a­bil­ity of (a) a prime; (b) a num­ber greater than 22; (c) a 77.
  6. A die is rolled 3030 times and shows a 66 eight times. Give the exper­i­men­tal and the­o­ret­i­cal prob­a­bil­i­ties of a 66, and say what you would expect after 30003000 rolls.
  7. A coin has come up tails five times in a row. What is the prob­a­bil­ity of tails on the next toss, and what is the name of the mis­take in expect­ing heads?
  8. Which of these have equally likely out­comes? (a) A car starts or does not start. (b) A fair die is rolled. (c) A mar­ble is drawn from a bag of 22 red and 88 blue: red or blue. (d) A fair coin is tossed.

Answers

Show answers
  1. (a) Cer­tain, 11. (b) Likely, 56\displaystyle \dfrac{5}{6}. (c) Unlikely, 14\displaystyle \dfrac{1}{4}. (d) Impos­si­ble for prac­ti­cal pur­poses, 00. (e) Even, 12\displaystyle \dfrac{1}{2}.
  2. 1880=0.225\displaystyle \dfrac{18}{80} = 0.225.
  3. (a) 2050=0.4\displaystyle \dfrac{20}{50} = 0.4. (b) 1650×900=288\displaystyle \dfrac{16}{50} \times 900 = 288.
  4. Eleven let­ters. (a) 211\displaystyle \dfrac{2}{11}. (b) O, A, I, I (the I appears twice): 411\displaystyle \dfrac{4}{11}. (c) P, R, O, A, L, T, Y: 711\displaystyle \dfrac{7}{11}.
  5. (a) 2,3,52, 3, 5: 12\displaystyle \dfrac{1}{2}. (b) 46=23\displaystyle \dfrac{4}{6} = \dfrac{2}{3}. (c) 00.
  6. Exper­i­men­tal 830≈0.267\displaystyle \dfrac{8}{30} \approx 0.267; the­o­ret­i­cal 16≈0.167\displaystyle \dfrac{1}{6} \approx 0.167. After 30003000 rolls the exper­i­men­tal value should be close to 0.1670.167.
  7. 12\displaystyle \dfrac{1}{2}; the gam­bler's fal­lacy.
  8. (b) and (d) only. In (a) the two results are not sym­met­ric; in (c) the ten mar­bles are equally likely but the two colours are not.