You are on page 1of 29

1/29

EC114 Introduction to Quantitative Economics


5. Basic Statistical Inference
Department of Economics
University of Essex
8/10 November 2011
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
2/29
Outline
1
Introduction
2
Populations and Samples
3
Statistical Inference
4
Sampling Distribution of the Mean
Reference: R. L. Thomas, Using Statistics in Economics,
McGraw-Hill, 2005, sections 3.13.2.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Introduction 3/29
The concepts we have studied so far e.g. the mean and
variance, are related to a population.
Usually we do not have access to population information
and have to rely on samples from the population.
The aim is to infer things about the population using the
information contained in the sample.
This is known as statistical inference.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Populations and Samples 4/29
Suppose a random variable X is determined by the
following probability distribution:
X 4 5 6 8
p(X) 0.1 0.2 0.3 0.4
There are only four discrete outcomes (4, 5, 6 and 8) but
the population of X may consist of a large number (possibly
millions) of values determined by the distribution.
For example, 10% of all X values will equal 4, 30% will
equal 6 etc.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Populations and Samples 5/29
Populations can be characterised by certain constants
the population parameters.
For example, the population mean, , is a population
parameter.
From the previous probability distribution we can calculate
as follows:
=
4

i=1
x
i
p(x
i
)
= x
1
p(x
1
) + x
2
p(x
2
) + x
3
p(x
3
) + x
4
p(x
4
)
= 4(0.1) + 5(0.2) + 6(0.3) + 8(0.4) = 6.4.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Populations and Samples 6/29
Another population parameter is the population variance,
denoted
2
.
The variance can be found using
2
= E(X
2
)
2
.
From the table we have
E(X
2
) =
4

i=1
x
2
i
p(x
i
)
= x
2
1
p(x
1
) + x
2
2
p(x
2
) + x
2
3
p(x
3
) + x
2
4
p(x
4
)
= 16(0.1) + 25(0.2) + 36(0.3) + 64(0.4) = 43.
Hence
2
= 43 (6.4)
2
= 2.04.
The standard deviation, , is therefore equal to 1.43; this is
a measure of the average deviation of X around its mean.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Populations and Samples 7/29
Important: population parameters are constants i.e. xed
numbers.
Unfortunately, they are also usually unknown.
For example, suppose we are interested in the annual
income of residents of a city.
It is too expensive and time consuming to interview all
residents so we could take a sample of, say, 100 or 1000
residents.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Statistical Inference 8/29
Suppose it were possible to nd the annual income of the
citys residents by conducting a survey of all residents in
week 45 of a given year, 2006.
There could be many millions of values for income (X) in
this population!
But we could use them to nd the mean: = 17,670, for
example (we could also compute
2
if we wanted to).
Furthermore, suppose we wish to do the same thing for
week 45 of 2011 (this week), but either we dont have the
time or we cant afford to survey the entire population to
get this information.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Statistical Inference 9/29
Instead we could take a sample of, say, 400 residents.
We might nd the sample mean to be

X = 19,110.
Ignoring price changes, based on our sample mean

X, can
we say that the mean annual income of residents in the
population, , has risen in the last ve years?
The mean appears to have risen (based on our sample)
but how condent can we be that it has risen in the
population?
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Statistical Inference 10/29
This is a typical example of statistical inference, which we
can summarise as follows:
We need to nd a level of income, k, such that:
if

X k then we conclude that has risen;
if

X < k then we conclude that is unchanged (or
hasnt risen).
A word on notation:
Greek letters e.g. and
2
, denote population
parameters;
Roman letters e.g.

X and v
2
, denote sample values.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Statistical Inference 11/29
One of the difculties with statistical inference is that
different samples yield different outcomes.
For example, our sample of 400 residents resulted in

X =19,110, but a second sample might give



X =18,210
while a third might yield

X =20,880 etc.
This is known as sampling variability.
It is important for the sample to be representative of the
population in some sense.
Fortunately, if the sample is random then sampling
variability follows known and systematic laws.
Denition
A sample is said to be random if every combination of n
members of the population has the same probability of
becoming the sample that is actually drawn.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Sampling Distribution of the Mean 12/29
Suppose we could take a large number of samples from
the population.
Each sample would give a different value for the sample
mean,

X.
If

X was between 18,000 and 19,000 in 10% of the
samples, we could interpret this as the probability of
obtaining a sample mean in this range:
Pr(18, 000 <

X < 19, 000) = 0.10.
We could do this for all ranges
e.g. Pr(21, 000 <

X < 22, 000) = 0.20 etc.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Sampling Distribution of the Mean 13/29
By doing so we could build up the probability histogram for

X:

It has a jagged appearance because the class widths are
equal to 1000.
If we reduced the size of the class widths to 100 or 1 it
would become smoother.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Sampling Distribution of the Mean 14/29
The previous histogram is the sampling distribution of the
mean.
It is the distribution of

X if it were possible to take a large
number of samples of size n (here n = 400) from the given
population.
In practice we typically only have access to a small number
of samples.
In Economics we usually have just one sample to work
with.
Fortunately we can use theoretical results to deduce the
sampling distribution for

X that would result from a large
number of random samples.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Sampling Distribution of the Mean 15/29
Theorem
If random samples of size n are taken from an innitely sized
population with mean and variance
2
then the sample
distribution of the sample mean

X will have mean and variance
E(

X) = , V(

X) =
2

X
=

2
n
,
respectively.
The rst property suggests that

X should be a good
estimator of because E(

X) = .
For a xed sample size n, the larger the population
variance
2
the larger the variance of

X,
2

X
.
For a xed population variance
2
, the larger the sample
size n the smaller the variance of

X,
2

X
.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Sampling Distribution of the Mean 16/29
Example (Thomas, Example 3.2) A long series of random
samples, all of size 12, is taken from a population of X
values with mean and variance
2
. For each sample, the
sample mean

X is calculated. The average value of all the

X values obtained turns out to be 20 with a standard


deviation of 2. Guess the values of and
2
.
Solution Our best guess is that = 20 because this is the
average of all the sample means,

X. We are also told that

X
=

2
n
= 4;
given that n = 12 we can solve this to obtain

2
= 4 12 = 48.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Sampling Distribution of the Mean 17/29
The Theorem tells us a lot about the sampling distribution
of

X, but nothing about its shape.
For example, is the distribution normal, uniform or
something else?
The following theorem is important for statistical inference.
Theorem (Central Limit Theorem)
If sufciently large samples are randomly drawn from a
population with mean and variance
2
, then the sampling
distribution of

X will be approximately normally distributed with
mean E(

X) = and variance
2

X
=
2
/n, regardless of the
shape of the distribution of the population.
This tells us that X can have any distribution but, provided
n is sufciently large,

X will be normally distributed.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Sampling Distribution of the Mean 18/29
CLT Example - Underlying PDF
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Sampling Distribution of the Mean 19/29
CLT Example - Sample Size 2
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Sampling Distribution of the Mean 20/29
CLT Example - Sample Size 3
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Sampling Distribution of the Mean 21/29
CLT Example - Sample Size 4
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Sampling Distribution of the Mean 22/29
CLT Example - Sample Size 8
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Sampling Distribution of the Mean 23/29
CLT Example - Sample Size 16
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Sampling Distribution of the Mean 24/29
CLT Example - Sample Size 32
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Sampling Distribution of the Mean 25/29
The Central Limit Theorem (CLT) implies that we can write

X

approx. N

,

2
n

for sufciently large n.


We can standardise

X in the usual way:
Z =

X

approx. N(0, 1).


Another way to write this is, noting that
2

X
=
2
/n,
Z =

X
/

approx. N(0, 1).


EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Sampling Distribution of the Mean 26/29
How large does n need to be?
Often, even for samples of size 25, the sampling
distribution for

X has a shape close to that of the normal
distribution.
For larger samples the approximation gets better and
better.
Usually the population parameters and
2
are unknown.
But, to show how the CLT can be used, we shall look at an
example where they are taken as given.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Sampling Distribution of the Mean 27/29
Example (Thomas, Example 3.3) A population has a mean
of 75 and a variance of 25. A random sample of size 80 is
drawn from the population. Find the probability that the
sample mean

X:
(a) lies between 74 and 76; (b) exceeds 100; (c) equals 75.
Solution We have = 75,
2
= 25 and n = 80.
(a) We require Pr(74 <

X < 76). Standardising we obtain:
Pr

74 75

25/80
<

2
/n
<
76 75

25/80

= Pr(1.789 < Z < 1.789) = 2 Pr(0 < Z < 1.789).


From the N(0, 1) table we nd that
Pr(0 < Z < 1.79) = 0.4633.
Hence the required probability is 2 0.4633 = 0.9266.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Sampling Distribution of the Mean 28/29
(b) This time we require Pr(

X > 100). Standardising, we


obtain
Pr

2
/n
>
100 75

25/80

= Pr(Z > 44.72) = 0.


(c) Assuming, not unreasonably, that

X is a continuous
variable, then the probability that

X takes on any given
value is zero i.e. Pr(

X = 75) = 0.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference
Summary 29/29
Summary
Populations and samples.
Statistical inference.
The sampling distribution of the mean.
Next week:
Estimation and testing of population parameters.
EC114 Introduction to Quantitative Economics 5. Basic Statistical Inference

You might also like