Applied Business Statistics

1
 Textbook
Wooldridge, Jeffrey (2008). Introductory Econometrics:
A Modern Approach. 4th edition, paperback. South-
Western, Division of Thomson Learning.

2
 The Nature of Business Statistics Data
• Reference: Wooldridge, Chapter 1.
• Business Statistic is used for:
– Estimating Business Models
– Evaluating & implementing policy
– Forecasting
• What is the effect of education on wages?
• How do training programs impact productivity?
• How will share prices develop in the future?

3
• Key ingredient: Data – typically in the form of
large samples.
• Data = information.
• Business Statistic= a method for processing
data and learning about general patterns in
the population of interest.
• For example, what is the effect of education
on labor market outcomes in the US?

4
 Common structures of Business data
1. Cross-sectional data: Sample of individuals, households, firms,
taken at a given point in time; often obtained from random sampling
from the underlying population.
2. Time series data: Observations on one or several variables over
time (e.g. GDP for Sweden 1971-2011). Time series observations are
unlikely to be independent over time which implies certain
methodological problems that we will study later.
3. Pooled cross sections: Combines cross-section datasets for
different time periods.
4. Panel (or longitudinal) data: Combines cross-section datasets
for different time periods for the same individuals.

5
 Example 1.1:
Becker’s model of crime
• Certain crimes have clear economic rewards,
but they also have costs.
• From Becker’s (1968) perspective, the decision
to participate in illegal activity is influenced by
the rewards and costs.
• Now write down an equation describing the
time spent in criminal activity as a function of
various factors:

6
A model of crime:

y = f ( x1, x2, x3, x4, x5, x6, x7 )

where f ( ) is a function (which remains unspecified for the moment)

y = hours spent in criminal activities

x1 = ”wage” for an hour spent in criminal activity

x2 = hourly wage in legal employment

x3 = other income

x4 = probability of getting caught

x5 = probability of getting convicted if caught

x6 = expected sentence if convicted

x7 = age

Think about whether the various x-variables likely impact on y positively or negatively. 7
 Model Specification
• Before we can undertake statisical analysis
linked to crime or worker productivity, the
models above must be made specific.
• This means we must decide exactly what the
function f( ) looks like.
• A second issue is how to deal with variables
tha cannot be observed (e.g. the wage that
someone can earn in criminal activity).

8
 A model of crime

where:
crime = measure of the frequency of criminal activity
wage = wage that can be earned legally;
othinc = income from other sources;
freqarr = frequency of arrests for prior crimes;
freqconv = freqency of conviction;
avgsen = average sentence length after conviction;

9
 Causality
• A common goal for applied statisian is to estimate
the causal effect of one variable on some outcome of
interest.
• Important: Distinguish correlation (association) from
causation.
• Ceteris paribus: other relevant factors being equal,
what is the effect of…
– a price increase on consumer demand
– training on worker productivity

10
 Causality (cont’d)
If…
a) …we succeed in holding all other relevant
determinants of (say) productivity constant;
and
b) …find a link between training and
productivity,
 …then we can conclude that training has a
causal effect on productivity.

11
 Causality (cont’d)
• Ideal setting is experimental: laboratory –
administer treatment to half the sample and
use the other half as control.
• Much of the research in Busines and
economics use non-experimental data
• A key challenge in Business Statistics is to
condition on enough other factors, so that a
case for causality can be made.

12
 Causality: Example
• Goal: Estimate the causal effect of education on wages
• Data: WAGE1.DTA. (Source: 1976 Current Population Survey in the US).
• Scatter plot:
25
20
average hourly earnings
10 5
0 15

0 5 10 15 20
years of education

13
 Causality: Example
• This of course doesn’t imply that education causes wages
• Wages are determined by many other factors except education –
for example, innate ability
– High ability => high wages
– High ability => high education (e.g. intelligent individuals choose high education )
• Perhaps the correlation between education and wages visible
in the graph is driven by ability rather than education?
• To credibly estimate the causal effect of education, we must find
a way of determining the link between education and wages
holding innate ability constant!
14
 Chapter 2:
The Simple Regression Model

15
 The simple regression model
Suppose we want to ”explain y in terms of x”.
Three issues:
1. Since there’s never an exact relationship between two
variables: how allow for other factors affecting y?
2. What is the functional form?
3. Are we capturing a ceteris paribus (causal)
relationship between y and x?

16
The simple linear regression model
Assume that, in the population, outcome variable y can be
modeled as a function of x as follows:

u: error term; disturbance term;

residual; noise

β0, β1: parameters, coefficients,

constants

17
 Simple regression: The functional
relationship between y and x is linear:
• If other factors in u are held fixed, so that the
change in u is zero (Δu=0), then in a linear model x
has a constant effect on y:

• Hence, β1 is the slope parameter, holding other

factors in u fixed – a parameter of primary interest
in applied Business
• The intercept parameter β0 (sometimes called the
constant term) is rarely central to an analysis.
18
 Examples:

• How interpret β1 in these equations?

• What are the ”other factors” that make up u in
these settings?

19
• To get reliable estimators of β0 and β1 from a random
sample of data we have to make an assumption
restricting how unobservable u is related to the
explanatory variable x.

• The crucial assumption:

– The left-hand side is a conditional expectation
– The right-hand side is an unconditional expectation (just
the expected value of u, regardless of x).
– So, this expression says that the expected value of u is
independent of x. Formally, we say that u is mean
independent of x.

20
 Detour: Expected Values

• The expected value is one of the most important

concepts related to probability that we will come
across
• If X is a random variable, the expected value of X is
denoted E(X), or sometimes μ.
• Sometimes the expected value is called the
population mean, emphasizing that X represents
some variable in a population.

21
 The expected value

• The expected value is a weighted average of all

possible values of X, where the weights are
determined by the pdf.
• Example: Suppose X takes on the values -1, 0, and 2
with probabilities 1/8, 1/2, and 3/8, respectively.
Then,

which is equal to 5/8 (or 0.625).

22
 Conditional Expectation
• We can summarize the relationship between one
variable Y and another variable X by looking at
the conditional expectation of Y given X.
• Basic idea: Suppose X has taken on a particular
value, say x.
• Then we can compute the expected value of Y,
given the outcome of X (i.e. x).
• We denote this expected value by E(Y|X=x), or
sometimes just E(Y|x).
23
 Example: Conditional Expectation
• Let (X,Y) represent the population of all working
indivduals, where X is education and Y is hourly
wage.
• E(Y|X=12) is simply the average hourly wage for all
people in the population with 12 years of education.
• Similarly, E(Y|X=16) the average hourly wage for all
people in the population with 16 years of education

24
Illustration:

25
 Now back to Chapter 2
• We encountered the following ’crucial assumption’:

• Look at the figure on the previous slide.

• How would you draw a figure for which
?

26
 Example
Model:

Assumption: E[u | educ] = E[u]

• What does the assumption mean in this

context?
• Does this assumption make sense, given the
context?

27
 A more innocent assumption
• As long as the intercept β0 is included in the
equation, we can always assume that the average
value of u in the population is zero:

28
Model:

Assumption:

Assumption:

Now show that the following is true:

• E(y|x) is the population regression function (PRF).

• It is a linear function of x.
• A one-unit increase in x changes the expected value of y by β1
29
Dispersion around the
average, given X = x3

30
 Interpretation:
Breaking y into two parts

Given the PRF, it follows that

= systematic part of y + unsystematic part of y

31
 Deriving the
Ordinary Least Squares Estimates

• We will now discuss how to estimate the

(unknown) parameters of this model.
• Let’s suppose we have a random sample of
size n drawn from the population:
• Note the i-subscripts on the two variables
(observation indices). The regression model:

32
 Estimation procedure

Assumption:

Assumption:
The first of these (mean independence) implies zero covariance
between x and u. We can now re-write the above assumptions as

(2.11)

(2.10)

(Note: Cov(x,u) means the covariance between x and u.

33
See Section B.4 in Appendix B.)
Note:

Re-write the above assumptions again:

(2.12)

(2.13)

• We have two unknown parameters. Can’t we just solve for

these?
• Not quite, because we don’t know the expected values of
y and x in the population.
• It’s precisely for this reason that the best we can hope to
do is estimate the parameters 34
 This applies for the population:

(2.12)

(2.13)

The ingredients are unobserved.

Suppose we choose estimates
to solve the sample counterparts of (2.12) & (2.13):

35
• Show how we can solve for and from these
equations ( you need to know how to do this).

where and
36
 (2.19)

(2.17)

• These are the OLS estimates of the

parameters of the simple regression model.
• Eq. (2.19): Covariance between x and y
DIVIDED by the variance of x.
• Hence, the sign of is always the same as
the sign of Cov(x,y) for this model.
37
Note: We require
• If this does not hold, is not defined.
• Why?
• What does this imply in practice?

38
 Why is this estimator called the ’ordinary
least squares’ (OLS) estimator?
• To see why, first define a fitted value for y when
x=xi as

• Next, define the residual for observation i as

Note that there are n such residuals.

• The OLS estimates minimize the sum of squared residuals:

Least squares…

39
40
 Some related concepts…
• The OLS regression line (or, the sample
regression function; SRF):

• Interpretation:

What is the difference between the sample regression

function and the population regression function?
41
 Taking stock
• We’ve derived the OLS estimator from two explicit
assumptions. Understanding these and why they
matter is important.
• We have said nothing so far about the statistical
properties of OLS
• So still some way to go… But we’ve made a start!
Now let’s look at an example.

42
 Example:
CEO Salary and Return on Equity

• These data were obtained by Wooldridge from the

May 6, 1991 issue of Businessweek.
• It would be interesting to do a similar data collection
exercise now and then investigate if the relationship
between RoE and pay has changed.

43
 15000
Scatter plot:

Cov(salary,roe) = 1342.5
1990 salary, thousands $
10000

Corr(salary,roe)= 0.11

Var(roe) = 72.6
5000 0

0 20 40 60
return on equity, 88-90 avg

We have enough information to figure out that the

regression coefficient on roe will be equal to…….
44
Results from simple OLS regression
. reg salary roe

Source SS df MS Number of obs = 209

F( 1, 207) = 2.77
Model 5166419.04 1 5166419.04 Prob > F = 0.0978
Residual 386566563 207 1867471.32 R-squared = 0.0132
Adj R-squared = 0.0084
Total 391732982 208 1883331.64 Root MSE = 1366.6

salary Coef. Std. Err. t P>|t| [95% Conf. Interval]

roe 18.50119 11.12325 1.66 0.098 -3.428196 40.43057

_cons 963.1913 213.2403 4.52 0.000 542.7902 1383.592

• How do we interpret this equation?

45
15000
10000
5000
0
The regression line

0 20 40 60
return on equity, 88-90 avg

1990 salary, thousands $ Fitted values

46
Assignment: Use the following model:

Compute OLS estimates based on the following data:

+---------+
| Y X |
|---------|
| 2 1 |
| 6 9 | Note: Complete the assignment
| 4 3 | using pencil, paper and a pocket
| 12 18 | calculator. Once you have an
| 4 5 | answer, you may want to check it
| 10 15 | using some ststistical software
| 6 7 | (e.g. Stata).
| 6 7 |
| 16 20 |
| 12 13 |
| 8 14 |
| 6 5 |
+---------+
47