MEI Mathematics in Education and Industry

MEI Structured Mathematics

Module Summary Sheets

Statistics 2
(Version B: reference to new book)

Topic 1: The Poisson Distribution

Topic 2: The Normal Distribution

Topic 3: Samples and Hypothesis Testing

1. Test for population mean of a Normal distribution
2. Contingency Tables and the Chi-squared Test

Topic 4: Bivariate Data

Purchasers have the licence to make multiple copies for use

within a single establishment

© MEI November, 2005

MEI, Oak House, 9 Epsom Centre, White Horse Business Park, Trowbridge, Wiltshire. BA14 0XG.
Company No. 3265490 England and Wales Registered with the Charity Commission, number 1058911
Tel: 01225 776776. Fax: 01225 775755.
Summary S2 Topic 1 − The Poisson Distribution

References: The Poisson distribution is a discrete random variable X E.g. The mean number of telephone calls to
Chapter 1 where an office is 2 every 10 minutes.
λr
Pages 1-4 P(X = r ) = e − λ The probability distribution (Poisson(2)) is as
r! follows:
The parameter, λ, is the mean of the distribution. P( X = 0) = e − 2 = 0.1353
We write X ~ Poisson(λ ). P( X = 1) = 2e − 2 = 0.2707
Example 1.1
Page 4 The distribution may be used to model the number of P( X = 2) =
2 2 −2
e = 2e − 2 = 0.2707
occurences of an event in a given interval provided the 2!
occurrences are: 2 3 −2
P( X = 3) = e = 0.1804
(i) random, (ii) independent, 3!
(iii) occurring at a fixed average rate. 4
2 −2
P( X = 4) = e = 0.0902
4!
Mean, E(X) = λ, Variance, Var(X) = λ
P( X > 4) = 1 − sum of above = 0.0527
Mean ≈ Variance is a quick way of seeing if a Poisson
Exercise 1A model might be appropriate for some data.
E.g. For λ = 2, find P(3) from tables.
Q. 7
It is possible to ca.culate terms of the Poisson distribution 23 8
by a recurrence relationship.
Note that P(3) = e -2 = 0.1353 ×
3! 6
λr = 0.1804
E.g. P( X = r ) = × e−λ ;
r! From tables, P(3) = P( X ≤ 3) − P( X ≤ 2)
λ r +1 λ = 0.8571 − 0.6767 = 0.1804
P( X = r + 1) = × e−λ = × P( X = r )
( r + 1)! ( r + 1)
Care needs to be taken over the cumulation of error s. E.g. Vehicles passing along a road were
counted and categorised as either private or
References: Use of cumulative probability tables commercial; on average there were 3 private
Chapter 1 and 2 commercial vehicles passing a point
Cumulative Poisson probability tables are on pages 40-42
Pages 5-6 every minute. It is assumed that the distribu-
of the Students' Handbook and are available in the examinations. tions are independent of each other. Find the
Example 1.2 They give cumulative probabilities, i.e. P(X ≤ r). probability that in a given minute there will
Page 6 So P(X = r ) = P(X ≤ r ) − P(X ≤ r − 1) be (i) no private vehicles, (ii) no vehicles,
For λ = 1.8 (Page 40), the second and third entry of the tables give (iii) exactly one vehicle.
Exercise 1A For the private vehicles, λ p = 3 and for
P(X ≤ 1) = 0.4628, P(X ≤ 2) = 0.7306
Q. 4, 5, 11 commercial vehicles λ c = 2.
i.e. P(X = 2) = P(X ≤ 2) − P(X ≤ 1) = 0.7306 − 0.4628 = 0.2678
The distribution may be modelled by the
Poisson distribution so that for private
References: Sum of Poisson distributions vehicles, X ~ Poisson(3) and for commercial
Chapter 1 Two or more Poisson distributions can be combined by vehicles, Y ~ Poisson(2).
Pages 12-15 addition providing they are independent of each other. For all vehicles, Z = X + Y ~ Poisson(3+2) =
Poisson(5)
If X ~ Poisson(λ) and Y ~ Poisson(μ),
Exercise 1B then X + Y ~ Poisson(λ +μ). (i) P(X = 0) = e−3 = 0.0498
Q. 4, 8
N.B. You may only add two Poisson distributions in this (ii) P(Z = 0) = e−5 = 0.0067
way if they are independent of each other. There is no cor- (iii) P(Z = 1) = 5e−5 = 0.0337
responding result for subtraction. Note that P(1 veh) = P(1 priv). P(0 comm)
+ P(0 priv) . P (1 comm)
= 3e−3.e−2 + e−3 .2e−2
References: Approximation to the binomial distribution
The Poisson distribution may be used as an approximation = (2 + 3)e−5 = 5e−5 = 0.0337
Chapter 1
Pages 18-22 to the binomial distribution B(n, p) when
E.g. Some equivalent values:
(i) n is large, (ii) p is small (so the event is rare).
Exercise 1C For X~B(40, 0.03), np = 1.2;
Q. 1(i), 3, 7 Then λ = np. P(0) = (0.97)40= 0.296 ≈ 0.3;
Note that the Normal distribution is likely to be a good For X~Poisson(1.2), P(0) = e-1.2
approximation if np is large. = 0.301 ≈ 0.3
Exercise 1D For X~B(80, 0.02), np = 1.6;
Statistics 2 P(0) = (0.98)80= 0.199 ≈ 0.2;
Q. 4, 5 Version B: page 2 For X~Poisson(1.6), P(0) = e-1.6
Competence statements P1, P2, P3, P4, P5 = 0.202 ≈ 0.2
© MEI
Summary S2 Topic 2 − The Normal Distribution

References: The Normal distribution, N(μ,σ2), is a continuous, For N(0,1), P(Z < z1) can be found from tables
Chapter 2 symmetric distribution with mean μ and standard (Students Handbook, page 44)
Pages 32-44 deviation σ. The standard Normal distribution N(0,1) has
mean 0 and standard deviation 1. E.g. P(Z < 0.7) = 0.7580
P(Z > 0.7) = 1 − 0.7580 = 0.2420
Example 2.1 P(X<x1) is represented by the area under the curve
Page 35 below x1.
(It is a special case of a continuous probability density
function which is a topic in Statistics 3.)
The area under the standard Normal distribution curve For N(2,9) [μ = 2, σ = 3]
can be found from tables. P(X < 5) = P(Z < z1) where z1 = (5 − 2)/3 = 1
To find the area under any other Normal distribution = 0.8413
Exercise 2A
curve, the values need to be standardised by the formula
Q. 4
x − μ E.g. The distribution of masses of adult males
z =
σ may be modelled by a Normal distribution with
mean 75 kg and standard deviation 8 kg. Find
References: Modelling the probability that a man chosen at random
Chapter 2 Many distributions in the real world, such as adult will have mass between 70 kg and 90 kg.
Pages 49, 50 heights or intelligence quotients, can be modelled well by
a Normal distribution with appropriate mean and We require P(70< X < 90) = P( z1 < Z < z 2 )
variance. 70 − 75
where z1 = = − 0.625
Given the mean μ and standard deviation σ, the Normal 8
distribution N(μ,σ2) may often be used. 90 − 75
Example 2.3 When the underlying distribution is discrete then the and z 2 = = 1.875
8
Page 49 Normal distribution may often be used, but in this case a
continuity correction must be applied. This requires us to
take the mid-point between successive possible values
Exercise 2B when working with continuous distribution tables. ⇒ P(70 < X < 90)
Q. 1
= P( Z < 1.875) − P ( Z < − 0.625 )
E.g. P(X) > 30 means P(X > 30.5) if X can take only
integer values. = P(Z < 1.875) − (1 − P( Z < 0.625) )
= 0.9696 − (1 − 0.7340) = 0.7036
References: The Normal approximation to the binomial
Chapter 2 distribution.
Pages 50-52 This is a valid process provided E.g. Find the probability that when a die is
(i) n is large, thrown 30 times there are at least 10 sixes.
(ii) p is not too close to 0 or 1. Using the binomial distribution requires
P(30 sixes) + P(29 sixes) + ..........+ P(10 sixes).
Mean = np, Variance = npq. However, using N(np, npq) where n = 30 and
Exercise 2B The approximation will be N(np, npq). p = 1/6, gives N(5, 4.167).
Q. 4 P(X > 9.5) = P(Z > z1)
A continuity correction must be applied because we are
approximating a discrete distribution by a continuous where z1 = (9.5 − 5)/2.04 = 2.205
distribution.
= 1 − 0.9863 = 0.0137
References: The Normal approximation to the Poisson N.B. a continuity correction is applied because
Chapter 2 distribution. the original distribution (binomial) is being
Pages 52-54 This is a valid process provided λ is sufficiently large for approximated by a continuous distribution
the distribution to be reasonably symmetric. A good (Normal).
guideline is if λ is at least 10.
Exercise 2B For a Poisson distribution, mean = variance = λ E.g. A large firm has 50 telephone lines. On
Q. 11 The approximation will be N(λ,λ). average, 40 lines are in use at once and the dis-
tribution may be modelled by Poisson(40). Find
As with the Binomial Distribution, a continuity the probability of there not being enough lines.
Exercise 2C
correction must be applied because we are approximat-
Q. 1, 6, 7 The distribution is Poisson(40).
ing a discrete distribution by a continuous distribution.
Approximate by N(40,40).
Statistics 2 Then we require P(X > 50) = P( Z > z1 )
Version B: page 3
50.5 − 40
Competence statements N1, N2, N3, N5 where z1 = = 1.66
© MEI 40
⇒ P(X > 50) = 1 − 0.9515 = 0.0485
Summary S2 Topic 3 − Samples and Hypothesis Testing
1: Estimating the population mean of a Normal distribution

References: The distribution of sample means E.g. If the parent population is N(10, 16) and a
Chapter 3 If a population may be modelled by a Normal distribu- sample of size 25 has mean 8.6, then this value
Pages 68-71 tion and samples of size n are taken from the comes from the sampling distribution of means
population, then the distribution of means of these which is N(10, 0.64).
samples is also Normal.
E.g. It is thought that the parent population is Nor-
Example 3.1 ( )
If the parent population is N μ , σ 2 then the mally distributed with mean 20.
Page 70 ⎛ σ2 ⎞ A random sample of 50 data items has a sample
sampling distribution of means is N ⎜ μ , ⎟. mean of 24.2 and s.d. 8.3.
⎝ n ⎠ Is there any evidence at the 0.1% significance
level that the mean of the population is not 20?
H 0 : μ = 20
References: Hypothesis test for the mean using the Normal H1 : μ ≠ 20
Chapter 3 distribution (Note that although the mean of the sample
Pages 71-73 Tests on the mean using a single sample.
H0 is μ = μ0 where μ0 is some specified value. is greater than the proposed mean, we do not
H1 may be one tailed: μ <μ0 or μ >μ0 have μ >20 because of the wording of the
or two tailed: μ ≠μ0. question.)
Exercise 3A
In other words, given the mean of the sample taken we _
Q. 1(i),(iii) x−μ 24.2 − 20
ask the question, “Could the mean of the parent ⇒z= = = 3.578
population be what we think it is?” σ 8.3
50
( )
n
Suppose the parent population is N μ , σ 2 , then the
Critical value from tables for two-tailed,
⎛ σ2 ⎞ 0.1% significance level is 3.27
sampling distribution of means is N ⎜ μ , ⎟ .The
⎝ n ⎠ Since 3.578>3.27 we reject H 0in favour of H1.
σ There is evidence that the mean of the
critical values are therefore μ ± k where the
n population is not 20.
value of k depends on the level of significance
and whether it is a one or two-tailed test. E.g. A population has variance 16. It is required to
_ test at the 0.5% level of significance whether the
x− μ mean of the population could be 10 or whether it
We therefore calculate the value z = and is less than this. A random sample size 25 has a
σ
n mean of 8.6.
References: compare it to the value found in tables. H0 : μ = 10
Chapter 3
Pages 73-74 Alternatively, if the value of the mean of the sample H1 : μ < 10
lies inside the acceptance region then we would accept k = 2.58 (for 0.5% level, 1-tailed test)
H0, but if it lay in the critical region then we would
Example 3.3 reject H0 in favour of H1. σ 4
Critical value is 10 − 2.58× = 10 − 2.58× = 7.936
Page 74 n 5
Alternatively, calculate the probability that the value is Since 8.6 > 7.936 we accept H0 ;there is no evidence
greater than the value found and see if it less than the
at the 0.5% level of significance that the mean is
significance level be used.
less than 10.
Alternatively, if the mean is 10 then the
sampling distribution of means is N(10, 0.64)
Exercise 3A Known and estimated standard deviation ⎛ 10 − 8.6 ⎞
Q. 6 The hypothesis test described above requires the value Then P( X ≤ 8.6) = 1 − Φ ⎜ ⎟ = 1 − Φ(1.75)
of the standard deviation of the parent population. ⎝ 0.8 ⎠
In reality the standard deviation of the parent = 1− 0.9599 = 0.0401.
population will usually not be known and will have to Since 0.0401 > 0.005 we accept H0
be estimated from the sample data.
If the sample size is sufficiently large, the s.d. of the Statistics 2
sample may be used as the s.d. of the parent popula- Version B: page 4
tion.
Competence statements N6
A good guideline is to require n ≥ 50.
© MEI
Summary S2 Topic 3 − Samples and Hypothesis Testing
2: Contingency Tables and the Chi-squared Test
References: Contingency Tables E.g. a group of 50 students was selected at random
Chapter 3 Suppose the elements of a population have 2 sets of dis- from the whole population of students at a
Pages 81-85 tinct characteristics {X, Y}, each set containing a finite College. Each was asked whether they drove to
number of discrete characteristics X = {x1, x2,….,xm} and College or not and whether they lived more than
Y = {y1,y2,…….,yn} then each element of the population or less than 10 km from the College. The results
will have a pair of characteristics (xi,yj). are shown in this table.
The frequency of these m × n pairs (xi,yj) can be tabulated
Nearer Further than 10
into an m × n contingency table. than 10 km km

y1 y2 yn Drives 11 17 28
x1 f1,1 f1,2 f1,n Does not 15 7 22
drive
x2 f2,1 f2,2 f2,n
26 24 50

xm fm,1 fm,2 fm,n E.g. If driving to College and the distance lived
are not associated events then if one student is
The marginal totals are the sum of the rows and the sum chosen at random the estimated probabilities are
of the columns and it is usual to add a row and a column 28 24
for these. P(drives) = , P(lives further than 10 km) =
50 50
The requirement is to determine the extent to which the 28 24
variables are related. and P(drives and lives further than 10 km) = ×
50 50
If they are not related but independent, then theoretical
= 0.2688
probabilities can be estimated from the sample data.
You now have two tables, one containing the actual So out of 50 people we would expect 50 × 0.2688
(observed) frequencies and the other containing the esti- = 13.44
mated expected frequencies based on the assumption that
the variables are independent. In a similar way the entries in the other three
boxes are calculated to give the following:
The hypothesis test
Nearer Further than 10
H0: The variables are not associated. than 10 km km
H1: The variables are associated.
Drives 14.56 13.44 28
2
References: The χ Statistic (Chi-squared statistic) Does not 11.44 10.56 22
Chapter This statistic measures how far apart are the set of drive
Pages 87-92 observed and expected frequencies.
26 24 50
( observed frequency - expected frequency )
2

X =∑
2

expected frequency
( fo − fe )
2
We test the hypotheses:
=∑ H0: the two events are not associated
fe
H1: The two events are associated.
References: Degrees of freedom ( fo − fe )
2

Chapter 3 The distribution depends on the number of free variables X2 = ∑

fe
Page 85 there are, called the degrees of freedom, ν.
(11−14.56) (17 −13.44) (15 −11.44) ( 7 −10.56)
2 2 2 2

This is the number of cells less the number of restrictions = + + +

14.56 13.44 11.44 10.56
placed on the data.
= 0.8704 + 0.9430 + 1.1078 + 1.2002
For a 2×2 table such as the example given the number of = 4.1214
cells to be filled is 4, but the overall total is 50 which is a
restriction and the proportions for each variable were also If the test is at the 5% level then the tables on
Exercise 3B estimated from the data, giving two further restrictions. page 45 of the MEI Students’ Handbook gives the
Q. 4, 5 So the number of degrees of freedom in the example is 1. critical value of 3.841 (v = 1).
Since 4.1214 > 3.841 we reject the null hypothe-
In general the number of degrees of freedom for an m × n sis, H , and conclude that there is evidence that the
0
table is (m − 1)(n − 1). two events are associated.
Exercise 3C If the test were at the 1% significance level then
Q. 1, 8 Statistics 2 we would conclude that there was not enough evi-
Version B: page 5 dence to reject the null hypothesis.
Competence statements H1, H2
© MEI
Summary S2 Topic 4 − Bivariate Data − 1

References: Bivariate Data are pairs of values (x, y) associated with a Example 1
Chapter 4 single item. The length (x cm) and width (y cm) of leaves
Pages 104-109 e.g. lengths and widths of leaves. of a tree were measured and recorded as
The individual variables x and y may be discrete or follows:
continuous. x 2.1 2.3 2.7 3.0 3.4 3.9
y 1.1 1.3 1.4 1.6 1.9 1.7
A scatter diagram is obtained by plotting the points (x1, y1),
(x2, y2) etc. The scatter graph is drawn as shown.
Correlation is a measure of the linear association between y
2.0
the variables.
A line of best fit is a line drawn to fit the set of data points
as closely as possible. 1.5
_ _
This line will pass through the mean point ( x, y ) where
_ _
x is the mean of the x values and y is the mean of the 1.0 x
2.0 3.0 4.0
y values.
_ _
The mean point is ( x, y ) which is (2.9, 1.5)
There is said to be perfect correlation if all the points lie on The line of best fit is drawn through the point (2.9,1.5)
a line.
E.g. 50 students are selected at random and
Correlation and Regression
their heights and weights are measured. This
If the x and y values are both regarded as values of random
will require correlation analysis.
variables, then the analysis is correlation. Choose a sample
A ball is bounced 5 times from each of a
from a population and measure two attributes.
number of different heights and the height is
If the x value is non-random (e.g. time at fixed intervals)
recorded. This will require regression
then the analysis is regression. Choose the value of one vari-
analysis.
able and measure the corresponding value of another.

References: Notation for n pairs of observations ( x , y ).

Chapter 4 For the data above:
⎛ _
⎞⎛ _
⎞ ⎛ _
⎞⎛ _
⎞
Pages 110-111 S xy = ∑ ⎜ xi − x ⎟⎜ yi − y ⎟ , S xx = ∑ ⎜ xi − x ⎟⎜ xi − x ⎟
⎝ ⎠⎝ ⎠ ⎝ ⎠⎝ ⎠
⎛ _
⎞⎛ _
⎞
∑ x = 17.4; ∑ y = 9.0; ∑ xy = 26.97
S yy = ∑ ⎜ yi − y ⎟⎜ yi − y ⎟ 17.4 × 9.0
⎝ ⎠⎝ ⎠ S xy = 26.97 − ⇒ S xy = 0.87
6
The alternative form for S xy is

S xy = ∑ xi yi −
∑ x ∑ y =∑ x y − n x y
i i
_ _

i i
n

For the data above:

∑x 2
= 52.76 ⇒ S xx = 2.3
Pearson’s Product Moment Correlation Coefficient
provides a standardised measure of covariance.
∑y 2
= 13.92 ⇒ S yy = 0.42
S xy 0.8 7
⎛ _
⎞⎛ _
⎞ r = = = 0.8 85
S xy ∑ ⎜⎝ xi − x ⎟⎠ ⎜⎝ yi − y ⎟⎠ S xx S yy 2 .3 × 0.4 2
References: r= =
2 2
Chapter 4 S xx .S yy ⎛ _
⎞ ⎛ _
⎞
Pages 111-114 ∑ ⎜⎝ x − x ⎟⎠ ∑ ⎜⎝ y − y ⎟⎠
i i
r can be found directly with an appropriate
calculator.
_ _
Example 4.1
=
∑x y −nx y
i i
Page 112 ⎛ 2 ⎞⎛ 2⎞
_ _

⎜ ∑ xi − n x ⎟ ⎜ ∑ yi − n y ⎟
2 2

⎝ ⎠⎝ ⎠
Statistics 2
Exercise 4A
Q. 2 The pmcc lies between -1 and +1. Version B: page 6
Competence statements b1, b2, b3
© MEI
 Summary S2 Topic 4 − Bivariate Data − 2
References: Testing a parent population correlation by means of a For the data of Example 1:
Chapter 4 sample where r has been found
The value of r found for a sample can be used to test r = 0.885
Pages 118-124
hypotheses about the value of ρ, the correlation in the parent We wish to test the hypothesis that there is
population. positive correlation between lengths and
Conditions: widths of the leaves of the tree.
(i) the values of x and y must be taken from a bivariate
Exercise 4C Normal distribution, Ho: ρ = 0 There is no correlation between
Q. 3 (ii) the data must be a random sample. the two variables.
An indication that a bivariate Normal distribution is a valid H1: ρ > 0 There is positive correlation
model is shown by a scatter plot which is roughly elliptical between the two variables
with the points denser near the middle. (1-tailed test).
H 0 : ρ = 0 There is no correlation between
From the Students’ Handbook, the critical
the two variables. value for n = 8 at 5% level (one tailed test) is
H1 : ρ ≠ 0 There is correlation between 0.6215.
the two variables (2 - tailed test.)
Since 0.885>0.6215 there is evidence that H0
Or : can be rejected and that there is positive
H1 : ρ > 0 There is positive correlation between correlation between the two variables.
the two variables (1 - tailed test.)
Example 2
Or :
2 judges ranked 5 competitors as follows:
H1 : ρ < 0 There is negative correlation between
the two variables (1 - tailed test.) Competitor A B C D E
Judge 1 1 3 4 2 5
References: Spearman’s coefficient of rank correlation Judge 2 2 3 1 4 5
Chapter 4 If the data do not look linear when plotted on a scatter graph d 1 0 -3 2 0
Pages 132-134 (but appear to be reasonably monotomic), or if the rank order d2 1 0 9 4 0
instead of the values is given, then the Pearson correlation 5
6 x14
Example 4.3
coefficient is not appropriate. ∑d 1
2
= 14 ⇒ rs = 1 −
5 x 24
= 0.3
Instead, Spearman’s rank correlation coefficient should be
Page 134
used. It is usually calculated using the formula For the data of example 2:
6∑ d i 2
rs = 0.3
Exercise 4C rs = 1 −
Q. 5 n ( n − 1)
2
Ho: ρ = 0 There is no correlation between
where d is the difference in ranks for each data pair. the two variables.
This coefficient is used: H1: ρ > 0 There is positive correlation
Statistics 2 (i) when only ranked data are available, between the two variables
Version B: (ii) the data cannot be assumed to be linear. (1-tailed test).
page 7 In the latter case, the data should be ranked. For n = 5 at the 5% level (1 tailed test), the
Competence Where rs has been found the hypothesis test is set up in the critical value is 0.9.
statements same way. The condition here is that the sample is random. Since 0.3 < 0.9 we are unable to reject Ho
b4, b5, b6, b7, b8 Make sure that you use the right tables! and conclude that there is no evidence to
© MEI suggest correlation.
Tied Ranks
If two ranks are tied in, say, the 3rd place then each should be
given the rank 31/2. E.g. For the data x 1 2 3 4 5
y 1.1 2.4 3.6 4.7 6.1
References:
Chapter 4
The least squares regression line
For each value of x the value of y given and the value on the
∑ x = 15, ∑ y = 17.9, ∑ x 2
= 55, ∑ xy = 65
_ _
15 17.9
Pages 142-144 line may be different by an amount called the residual. ⇒x= = 3, y = = 3.58
If the data pair is (xi,yi) where the line of best fit is 5 5
_
y = a + bx then yi - (a + bxi ) = ei giving the residual ei .
⇒ S xx = ∑ x 2 − n x 2 = 55 − 6 × 32 = 10
Exercise 4D The least squares regression line is the line that minimises the _ _
Q. 4 sum of the squares of the residuals. ⇒ S xy = ∑ xy − n x y = 65 − 5 × 3 × 3.58 = 11.3
11.3
S xy
( x − 3) ⇒ y = 1.13 x + 0.19
_ _
The equation of the line is y − y = ( x − x) ⇒ y − 3.58 =
S xx 10