Interactive Lessons

Normal distribution

Lecture notes

Definition

The normal distribution

A normal variable has a bell-shaped, symmetric curve. Two numbers fix it: the mean (the centre) and the standard deviation (the spread). We write ; the second number is the variance .
μ146154162170178186194
Example
Adult heights in a town have mean cm and standard deviation cm.: the curve is highest at and falls away on both sides, most of the people are within cm of the mean.
Definition

Symmetry and area 1

The curve is symmetric about : the mean, the median and the mode are all equal to , and half of the values lie on each side. The whole area under the curve is (the total probability), whatever and are.
0.50.5146154162170178186194
Example
Heights, : . and are both cm from the mean, so .
Rule

The mean moves, the spread stretches

A different moves the curve along the axis; its shape stays. A larger stretches it: wider and lower. The area stays , so the peak height is .
0.026670100130
Example
has peak height .With (orange) the curve is twice as wide and the peak is half as high: .
Rule

Points of inflection at μ ± σ

In the middle the curve bends downward, in the tails upward. It changes its bend at the two points of inflection and . So you can read off a picture: it is the distance from the centre to where the curve starts to bend the other way.
0.079845505560657075
Example
bends at and .There the height is the peak times : .
Rule

The rule of thumb 68 – 95 – 99.7

For a normal distribution about 68% of the values lie within one standard deviation of the mean, about 95% within two and about 99.7% within three. Almost nothing lies further out than .
99.7%95%68%146154162170178186194
Example
Heights, :between and about 68%; between and about 95%; between and about 99.7%.Of people about are between and and about are outside.
Tip

The exact shares

The rule of thumb is rounded. The exact shares are 68.27%, 95.45% and 99.73%. For exactly 95% you need standard deviations, not ; for 99% you need .
95%−1.961.96−3−2−10123
Example
Within two standard deviations lies , so of people on average and not exactly .And : that is where the comes from.
Definition

Fitting a normal curve to data

To model data with a normal curve, take the mean of the data as and the standard deviation of the data as . The curve should then follow the tops of the histogram bars.
Example
Twelve heights (cm): .Sum , so . Squared distances: , so , .
160165170175180185
Rule

Sample or whole group: n − 1 or n

There are two standard deviations. For a sample from a larger group divide by : . For the whole group divide by : . For a large they are almost equal.
Example
The twelve heights as a sample: .As a whole group: .With the two differ by only about %.
Tip

Check the fit with the bands

Count how many values lie within one and within two standard deviations of the mean. For data from a normal distribution that is about 68% and about 95%. If the shares are far off, or the histogram is clearly lopsided, doubt the model.
Example
The twelve heights, , :within one standard deviation ( to ) lie of values, that is % (rule: 68%);within two all , that is % (rule: 95%). The fit is good.
160165170175180185
Watch out!

When a normal model is not appropriate

Do not use a normal curve when the data are clearly skewed, have two peaks, run into a hard limit that the curve would cross, or are too few to judge. A quick test: must not fall below a hard limit such as .
Example
Twelve reaction times (s): .Mean , , so : below , but a time cannot be negative. The median lies below the mean: skewed to the right. A normal model is not appropriate.
−0.150.050.250.450.65
Definition

A continuous variable: probability is area

A continuous variable, such as a height, can take every value in an interval. The probability that lies between and is the area under the curve from to : .
0.4680165175146154162170178186194
Example
Heights, : the area between and is .So about % of the people are between and cm.
Rule

One value has probability 0

For a continuous variable : one value has no width, so no area. Hence and : the end points do not matter. For a discrete variable such as a binomial one they do.
0.8413178146154162170178186194
Example
: .But : and ; the value itself has probability .
Rule

Four kinds of area

A normal probability is one of four areas: the left tail , the right tail , the area between and the two tails . Always draw the curve and shade the area first.
1151158513085130
Example
A test score has ., , and the two tails .
Rule

The calculator line

A graphing calculator gives the area directly. The arguments are lower limit, upper limit, mean and standard deviation: TI , Casio (there the standard deviation comes before the mean). For an open end type or .
Example
and :TI , Casio ; the answer is .: gives .
Rule

Standardising: z and Φ

Every normal curve becomes the standard normal curve when you standardise: . Then with , where is the area of the standard normal curve to the left of .
0.841311555−370−285−11000115113021453z
Example
, : , so .The same area, only the axis changed: below the axis stands the axis.
Definition

The z-score

The z-score tells how many standard deviations lies from the mean: positive above, negative below. Back: .
13055−370−285−11000115113021453z
Example
: has : two standard deviations above the mean. gives .
Definition

The printed table of Φ(z)

A printed table gives to 4 decimals for . Find the row of to one decimal and the column of the second decimal; the number where they meet is .
z0.040.050.060.070.081.10.87290.87490.87700.87900.88101.20.89250.89440.89620.89800.89971.30.90990.91150.91310.91470.91621.40.92510.92650.92790.92920.9306
Example
: row , column : .
Rule

Negative z and the symmetry rule

The table has only . For a negative use the symmetry: . Also and .
0.10380.1038−1.261.26−3−3−2−2−1−100112233z
Example
, and is the same by the mirror.For that is , because .Between: .
Watch out!

The last digit may differ

Every table entry is rounded to 4 decimals. A result built from two entries can therefore differ in the last digit from the calculator value. Both are right to that accuracy: give 4 decimals and do not round further.
Example
, : and .Table: .Calculator : , because .
Rule

Add columns for the third decimal

Some formula booklets (such as the MF19 of Cambridge 9709) give for two decimals of and an ADD block for the third decimal. Take the entry for the first two decimals and add the number of the third decimal from the ADD block (in units of ).
z0.050.060.07+41.20.89440.89620.898071.30.91150.91310.91476
Example
: the entry for is ; in row the third decimal adds (that is )..Where the comes from: from to the entry rises by ; the third decimal is of that step: .
Rule

Two tails and the critical values

is the area of both tails together. The values you meet most: leaves % in the two tails, leaves % and leaves %.
95%2.5%2.5%−1.961.96−3−2−10123
Example
: % in each tail.For : , the same as .
Rule

Comparing with z-scores

Values on different scales cannot be compared directly. Standardise both: the larger is the better position within its own group. A z-score compares relative position, not size.
93.3%650−3−2−15000123z91.9%28−3−2−1210123z
Example
A scores on a test with mean and standard deviation : .B scores on a test with mean and standard deviation : .A stands higher: % of the group of A scored lower, against % for B.
Definition

Percentile rank and the value at a percentile

The percentile rank of is the share of the values below it: %. The other way round, the value at the -th percentile is with . A percentile is not a percentage of right answers.
90%119.2557085100115130145
Example
gives : the rd percentile (% lie below).The th percentile of : , so .A mark on a test with mean and standard deviation has : only % of the group is lower, although % is right.
Definition

The inverse normal

The inverse normal goes from an area to a boundary: given , find . A calculator does it in one line ( on a TI, on a Casio). With a table read backwards and use .
00.51119.220.9557085100115130145
Example
and .Table: gives , so .Calculator: gives . Check: .
Rule

Inverse: left, right and middle

The inverse normal always needs the left area. Right tail : use the left area . Middle , symmetric about : each tail holds , so has left area and .
10%180.2515417018690%156.84183.16154170186
Example
Heights, .The tallest %: left area , , cm.The middle %: left area , , so : from to cm.In the table lies between () and ().
Rule

One unknown: μ or σ

When or is unknown, one condition fixes one unknown. Turn into a z-value with the inverse normal, then solve for the unknown: or .
0.24048.42
Example
and : the left area gives , so .If instead is known: .
Rule

Two unknowns: two equations

Two conditions give two equations and . Subtract them: , then . Always check by putting the answers back into the two conditions.
0.20.1405545.946
Example
and . and , so .. Check: and .
Rule

The middle interval: 1.96σ or 2σ

The middle interval that holds a share is , where is the with left area : for 90%, for 95%, for 99%. The “two sigma” of the rule of thumb is a round : it holds %.
95%70.6129.4557085100115130145
Example
, the middle %: , from to .With “two sigma” the interval is to , which holds .
Definition

The normal approximation of the binomial

For a large the bars of form a bell. The normal curve with the same mean and variance fits them: with . The standard deviation is .
3335373941434547495145.5
Example
: the mean is , the variance , the standard deviation .So use .
Rule

When is the approximation good?

The bars must not be lopsided: and both at least 5 (take 10 to be safe). Otherwise the error is large and you use the exact binomial. A measure is the largest difference between the exact and the approximate over all .
−4−3−2−101234z
Example
: and , both large: the largest error is only .: and : the first is below , the largest error is , twenty times as big. Use the exact binomial here.
Rule

The continuity correction

The bar of covers to . So is approximated by the normal area left of , not left of . The extra half-unit strip is what the continuity correction adds.
Example
, : exact .With the correction: , (error ).Without: , (error ).The strip from to has area .
33353739414345474951453335373941434547495145.5
Rule

The correction for every event

Write the event with or first ( is , is ). The corrected boundary lies half a unit outside the last value included: : to ; : from ; : to ; : to .
32343638404244464837.542.5
Example
, :: from : exact , normal .: to : exact , normal .A common slip: needs (normal , exact ); would give , that is .
Rule

A linear function of a normal variable

If , then is normal again, with mean and standard deviation (variance ). The bell is moved and stretched, never bent out of shape.
23.8914202675526884
Example
A temperature in °C is . In °F it is : mean , standard deviation , so .: , so ; the same as .
Rule

Sums of independent normals

If and are independent and normal, then is normal with mean and variance . In general has mean and variance .
Example
and are independent.Mean , variance , so ..
Watch out!

Variances add, standard deviations do not

Add the variances, never the standard deviations. For a difference the variances also add: , while the mean is . A difference is never less spread out than its parts.
105.5810.5
Example
Same and : the standard deviation of is , not . With you would find instead of ., so .
Tip

Which model fits?

Check the situation before you calculate. Binomial: a fixed number of independent trials, success or failure, the same chance each time. Normal: a measurement or a sum that is symmetric, has one peak and no hard limit. A binomial with and at least 5 is both.
Example
Heads in coin tosses: binomial, and , so also nearly normal.Waiting time for a bus: not normal (skewed, hard limit at ).A guessed test of questions with choices: binomial with and , just at the border; the largest error of the approximation is .Adult heights: normal.
Tip

Everyday: heights

Read the words: “the tallest %” is a right tail and needs the inverse normal; “taller than a door” is a right-tail area; “between two heights” is an area between. Give the answer in a sentence with the unit.
10%180.25146154162170178186194
Example
Heights, .The tallest % start at cm (left area ).Taller than a door of cm: , : about % of the people.
Tip

Everyday: the filling machine

A machine fills bottles with a mean setting and a spread . The share under the label volume is a left tail. To keep at most a share under the label, raise the mean setting to , or reduce .
0.3085500490494498502506510514
Example
Label ml, ml. At mean : and , about % under the label.For at most % under: ml. With it is ml.
Tip

Everyday: exam grades

A mark’s percentile rank is %; a mark at a given percentile needs the inverse normal; a pass mark gives the share that passes as a right tail.
0.72025526385062748698
Example
Marks .Mark : , percentile rank .The th percentile: .Pass mark : and : about % pass.
Tip

Everyday: shaft tolerance

A part is good when its size lies in the tolerance band . The share outside is two tails: . For at most a share outside choose with , where has left area .
6.7%6.7%19.9720.0320
Example
A shaft of nominal diameter mm, mm, tolerance mm: and the share outside is , about %.For at most % outside: , so mm. Check: gives .
Tip

Which course asks what

Cambridge 9709: the normal distribution with tables (Paper 5), the approximation with continuity correction, sums of normals (Paper 6). UK A level (DfE, Edexcel, AQA): the normal as a model and choosing the distribution. IB (AA, AI): curve, standardising, inverse normal, unknown mean or standard deviation; sums of normals in AI HL. AP Statistics: z-scores, the empirical rule, areas and intervals. Common Core: fit a normal curve to data. Netherlands: the three rules of thumb (wiskunde A havo), the standard normal (wiskunde D vwo).
Example
A typical exam question: ; find and the value with ., so ; and .

Exercises

Heights in a village: (cm). Give the mean, the standard deviation, the two points of inflection and . Easy
μ151158165172179186193
Show solution
Mean , standard deviation (the variance is ).
The points of inflection are at : and .
By symmetry half of the area lies to the left: .
Mean , standard deviation , bends at and ,
Check → and are both away from .
The marks of a test are . Use the rule of thumb: between which marks do about 95% of the pupils lie? About what share scores more than ? Easy
95%52586470768288
Show solution
About 95% lie within standard deviations: , from to .
About 99.7% lie within : , from to . So % lie outside and half of it, about %, above .
From to ; about % score above
Check → the exact value is , about %: close to the rule of thumb.
. Find the z-score of , and the value that has . Easy
Show solution
.
.
and
Check → is standard deviations () above ; a negative gives a value below the mean.
. Find with the table of . Easy
?6226344250586674
Show solution
Standardise: .
In the table: row , column : .
So . (The single value has probability , so or gives the same.)
Check → lies above the mean , so the answer is above .
Use the table of : find , and . Easy
z0.010.020.030.040.050.70.76110.76420.76730.77040.77340.80.79100.79390.79670.79950.80230.90.81860.82120.82380.82640.8289
Show solution
Row , column : .
.
By symmetry .
, and
Check → .
. Find and . Easy
Show solution
is continuous, so a single value has no area: .
, so .
and
Check → is the same number.
A bottle holds ml. Find with the table. Medium
?498505490494498502506510514
Show solution
and .
, and by symmetry.
.
Check → the calculator line gives as well.
. The highest % of the values lie above a boundary . Find to decimals. Medium
5%?557085100115130145
Show solution
A right tail of means a left area of : .
The inverse normal gives .
.
Check → gives the same; lies standard deviations above .
The marks of a test are . Between which two marks do the middle % of the pupils lie? Medium
Show solution
The two tails hold , so in each tail.
The lower boundary has left area : . The upper one has left area : .
, so the boundaries are and .
From to
Check → the interval is symmetric about ().
Pupil A scores on a test with . Pupil B scores on another test with . Who did better compared with their own group? Give both percentile ranks. Medium
Show solution
and .
The larger is the better relative position: A.
Percentile ranks: % for A and % for B.
A did better: against ; ranks % and %
Check → both are positive, so both percentile ranks are above %.
and . Find the mean to decimals. Medium
0.150?
Show solution
A right tail of is a left area of , so for the boundary .
, so .
.
Check → for the value lies standard deviations above the mean: .
. Use the normal approximation with the continuity correction to estimate . Medium
91317212529333720.5
Show solution
and are both at least ; , so the standard deviation is . Use .
becomes the area left of : .
with the calculator. (With the table and rounded to you get .)
Check → the exact binomial value is ; without the correction you would get .
and are independent. Find . Medium
?25381318232833
Show solution
The mean of is and the variance is , so the standard deviation is (not ).
, so .
.
Check → the standard deviation is larger than and than , but smaller than .
with and . Find and to significant figures. Hard
0.10.63050?
Show solution
The inverse normal gives for the left area and for .
Two equations: and .
Subtract: , so .
Then .
and
Check → and , the two z-values.
A machine fills bottles with ml; the label says ml and at most % of the bottles may hold less. How high must the mean be set (2 decimals)? How much more than the present ml? Hard
Show solution
: the left area gives .
, so .
At : , about % under the label. The mean must go up by ml.
ml; now % is under the label; ml more
Check → .
. Use the normal approximation with the continuity correction for . Is it suitable? Hard
5256606468727680848859.575.5
Show solution
and , both far above : suitable. : use .
The boundaries are and : , , so .
; the approximation is suitable
Check → exact ; without the correction .
A shaft has nominal diameter mm and tolerance mm; the diameters are (mm). What share is outside? Which (3 significant figures) keeps at most % outside? Hard
Show solution
The limits are standard deviations away: . Two tails: , about %.
For at most % each tail is at most : the limit must lie standard deviations from the mean.
; take mm.
About % outside; mm
Check → gives and outside, below .
Eight people get into a lift; their weights are independent, each (kg). The limit is kg. Find the probability that the total is over the limit. Hard
?650600
Show solution
The total is normal with mean and variance , so the standard deviation is (not ).
, so .
, about %
Check → the wrong standard deviation would give , far too big.
Ten times (s): . Fit a normal model (use ). How many values lie within ? Estimate and say how far you trust it. Hard
475053565962
Show solution
Sum : . Squared distances : , .
Within ( to ): of , % (rule: 68%): a good fit.
, : with only values trust the order of magnitude.
, , of within , (rough)
Check → the largest value is standard deviations above the mean: nothing is far out.
Are you a teacher?

Start live lessons with these applications for your whole class. Free to try for one month.

Try it free

Fits these school years

IB (MYP, Diploma): Diploma, year 1, Diploma, year 2

Cambridge International: AS Level, A Level

Netherlands: Havo · Mathematics A, Vwo · Mathematics A, Vwo · Mathematics C, Vwo · Mathematics D

England (GCSE, A Level): Year 12 (A Level), Year 13 (A Level)

USA (Common Core): Algebra 2, Precalculus

What the applications of this topic teach 6 applications

Normal curve: shape, rule of thumb and fit to data

  • Describing a normal curve: bell-shaped, symmetric, centre μ, spread σ
  • Using the 68 – 95 – 99.7 rule of thumb
  • Fitting a normal curve to data

Normal probabilities: areas, tables and calculator

  • Finding P(X < a), P(X > a) and P(a < X < b) for a normal variable
  • Standardising and reading Φ(z) from the table
  • Using two tails and the critical values 1.645, 1.96 and 2.576

z-scores and percentiles

  • Computing a z-score and going back with x = μ + zσ
  • Comparing values from different distributions with z-scores
  • Finding a percentile rank and the value at a percentile

Inverse normal: find x, find the mean or the standard deviation

  • Going from an area to a boundary with the inverse normal
  • Finding an unknown μ or σ from one or two conditions
  • Finding the symmetric interval that holds 90%, 95% or 99%

Normal approximation and sums of normals

  • Approximating a binomial probability with the normal distribution
  • Finding the distribution of aX + bY for independent normal variables
  • Deciding which model fits: binomial, normal, both or neither

Normal distribution in everyday life

  • Turning a story into a normal model in four stages
  • Answering questions about heights, a filling machine and exam grades
  • Writing the result as a sentence with its unit