Transcript STATISTICS FOR MANAGERS - Pompeu Fabra University

INTRODUCTION
APPLIED ECONOMETRICS
BASICS I
1900
200
180
1800
1700
140
120
Sales
1600
100
1500
80
60
1400
40
1300
20
1200
0
1999
2000
2001
2002
2003
2004
Advertising expenditure
160
Sales
Advertising exp.
BASIC PROBLEM I
Endogeneity
 Circularity
 Egg and chicken
 Causality and correlation: two very
different concepts!!

BASIC PROBLEM I
Interpretation of any graph/table
implies many assumptions.
 However, these assumptions are
almost never explicit.
 Assumptions for the interpretation of
the previous graph?

Direction
 Omission

BASICS II
1936 US presidential election
 Sampling list: mail out ballot cards
to residential telephone subscribers
and owners of cars.
 Result of the poll: Landon
(republican) will win with 57% of the
vote over Roosevelt (democract).

BASICS II
Outcome of the election: Roosevelt
won with 62.5% of the votes (523 of
the 531 electoral votes!!)
 What happened?

BASICS PROBLEM II
Therapy of hormonal replacement
for women with menopause. Does it
work?
 Possible problem.
 Solution.
 Why did the result with
observational data was wrong?

BASICS PROBLEM II
Sample selection problem.
 Training courses for employees.
 Annual physical: medical review.

SOLUTION if possible






Randomized experiment. Example.
In medical science only randomized
experiments are accepted.
Tobacco litigation.
Food and Drug administration.
Many times experiments are NOT
AVAILABLE AND ARE VERY EXPENSIVE.
Look for other designs: “clever”
regression and proper estimation
procedures.
EXOGENEITY

Basic problem: to find an exogenous
source of variation.
Impossible (Lucas’ Critique):
everything in economics is set
simultaneously->DGEM and
computable models. “Deep” parameters
 Construct experiments or look for
natural and pseudo-experiments. Find
credible sources of exog. variation.

IMPORTANT!!
Statistical methods are never wrong!
 It is their application by clumsy, unexperienced or careless researchers
that makes the results wrong.
 Remember: if you get the design
/assumptions /data right the results
will always be right.

The NYT litigation




Last year Morgan Stanley agreed to pay
$54 millions for a sexual discrimination
demand. Now Walmart.
In the 70’s some women journalist at the
NYT claimed that they were
discriminated.
Is a simple difference between the salary
of men and women a good indicator of
discrimination?
Judge decision.
Regressions
(1)
wagei   0  96,78womani  ui
(9,35)
(2)
wagei   0  77,14womani   2 educi  ui
(8,51)
(3)
wagei  0  13,86womani   2educi  3 prizesi   2 promoi  3inceni  ui
(11,05)