Principles of Macroeconomics

Twelfth Edition

Part VI
METHODOLOGY

21-1
Principles of Macroeconomics
Twelfth Edition

Chapter 21
Critical Thinking about
Research

Copyright 2017 Pearson Education, Inc. 21-2

Copyright

Copyright © 2017 Pearson Education, Inc. 21-3

Chapter Outline and Learning Objectives
21.1 Selection Bias
• Give some examples of studies that might suffer from selection bias.

21.2 Causality
• Understand the difference between correlation and causation.
21.3 Statistical Significance
• Understand how researchers decide whether their results are
meaningful.

21.4 Regression Analysis

• Understand how regression analysis can be used for both estimation
and testing.

Copyright © 2017 Pearson Education, Inc. 21-4

Chapter 21 Critical Thinking about
Research
• You have seen in this text how we can begin to answer
many questions with economic theory.
• To get quantitative answers, we need to use statistical
methods to look at real-world data.
• This chapter introduces the tools economists and other
social scientists use to look at data.

Copyright © 2017 Pearson Education, Inc. 21-5

Statistical Bias
• selection bias Selection bias occurs when the sample
used is not random.
• One example is a comparison of the bone densities of
70-year-old men and 90-year-old men.
• survivor bias Survivor bias exists when a sample
includes only observations that have remained in the
sample over time, making that sample unrepresentative of
the broader population.
• This bias results in comparing apples and oranges.

Copyright © 2017 Pearson Education, Inc. 21-6

Causality
Correlation versus Causation
• correlated Two variables are correlated if their values
tend to move together.
• Two variables are positively (negatively) correlated if they
move in the same (opposite) direction(s).
• Most people who have minivans also have children: This is
obviously a correlation but does not illustrate not causality.

Copyright © 2017 Pearson Education, Inc. 21-7

Random Experiments
• Random experiments are the gold standard for empirical
work.
• They are common in medical research but not always
possible to carry out.
• A study of the causal relationship between smoking and
cancer found that people who smoke may also have made
other choices that are unhealthy.
• intention to treat A method in which we compare two
groups based on whether they were part of an initially
specified random sample subjected to an experimental
protocol.
Copyright © 2017 Pearson Education, Inc. 21-8
Regression Discontinuity (1 of 2)
• Economists often make inferences from market data
instead of using random experiments.
• Market data reflect real choices and are easily available to
researchers.
• However, identification of causality is difficult because
individual choices are made in relatively uncontrolled
settings.

Copyright © 2017 Pearson Education, Inc. 21-9

Regression Discontinuity (2 of 2)
• regression discontinuity Identifies the causal effects of
a policy or factor by looking at two samples that lie on
either side of a threshold or cutoff.
• Effectively compares outcomes from individuals who are
close to either side of a dividing line.

Copyright © 2017 Pearson Education, Inc. 21-10

ECONOMICS IN PRACTICE
Moving to Opportunity

• In the mid-1990s, the Moving to

Opportunity program offered randomly
selected families living in high-poverty
housing projects housing vouchers for
moving to lower-poverty neighborhoods.

• Economists found substantial effects on

those who moved as children with their
families, with average gains of 31%
higher incomes.

THINKING PRACTICALLY

1. Some of the same researchers whose work is described also did another study looking
at the outcomes of households that moved versus those that did not in the general
population. To control for selection bias, the researchers compared children of different
ages within families to see how much more time in the better neighorbood influenced
younger versus older children. How does this attenuate the selection bias issue?

Copyright © 2017 Pearson Education, Inc. 21-11

ECONOMICS IN PRACTICE
Birth Weight and Infant Mortality

• Very low birth weight (VLBW) babies—

those with a birth weight less than 1500
grams—receive extraordinary care at
birth.

• In a regression discontinuity study,

economists looked at outcomes for
babies on either side of the VLBW line.
They found that babies below the line
had a 1% lower one-year mortality rate
than the slightly heavier babies.

THINKING PRACTICALLY

1. Can you think of another medical designation for which a regression discontinuity
technique might be useful?

Copyright © 2017 Pearson Education, Inc. 21-12

Difference-in-Differences (1 of 2)
• difference-in-differences Difference-in-differences is a
method for identifying causality by looking at the way in
which the average change over time in the outcome
variable is compared to the average change in a control
group.

Copyright © 2017 Pearson Education, Inc. 21-13

Difference-in-Differences (2 of 2)
• Consider the effect of a gardening program on housing
prices in community a:
effect  penda  pbega  (pendb  pbegb)
where:
pbega = average housing value in community a.
pbegb = average housing value in community b.
penda = average housing value in community a after 4 years.
pendb = average housing value in community b after 4 years.

Copyright © 2017 Pearson Education, Inc. 21-14

ECONOMICS IN PRACTICE
Using Difference-in-Differences to Study the Minimum Wage

• In the 1990s, New Jersey raised its

minimum wage, while its neighboring
state, Pennsylvania, did not.

• Two economists looked at the difference

in employment changes in between
restaurants of those two states. They
found no effect from the minimum wage
law change.

THINKING PRACTICALLY

1. Design another experiment using difference-in-differences to understand the effect of a

policy change at your college.

Copyright © 2017 Pearson Education, Inc. 21-15

Statistical Significance
• p-value The probability of obtaining the result that you find
in the sample data if the null hypothesis of no relationship
is true.
• statistical significance A result is said to be statistically
significant if the computed p-value is less than some
presubscribed number, usually 0.05.
• One has more confidence in results obtained from
populations with low variation than from those with high
variation.

Copyright © 2017 Pearson Education, Inc. 21-16

Regression Analysis (1 of 4)
• Regression analysis is the most important statistical tool in
empirical economics.
• Example:
V a b g
where:
V = incumbent party’s presidential vote share.
g = growth rate of the economy in the year of the election.

• The job of regression analysis in this case is to find the

values of the coefficients a and b that provide a good fit of
the data.

Copyright © 2017 Pearson Education, Inc. 21-17

FIGURE 21.1 Hypothetical Plot of Points between the Vote Share
and the Growth Rate

Copyright © 2017 Pearson Education, Inc. 21-18

Regression Analysis (2 of 4)
• The best line is one that has the sum of all squared vertical
distances between the actual data points and the line
(sum).
• The values of a and b that correspond to the smaller value
of sum are the best-fitting intercept and slope, respectively.
• least squares estimates Least squares estimates are
those that correspond to the smallest sum of squared
distances, or errors.

Copyright © 2017 Pearson Education, Inc. 21-19

Regression Analysis (3 of 4)
• How do we test whether b is zero?
1. Postulate the null hypothesis that b is in fact zero.
2. Use regression analysis to estimate b and compute
the p-value that b is zero.
3. If the p-value is low—for example, less than 0.05—the
estimate of b is statistically significant.
4. Reject the null hypothesis.

Copyright © 2017 Pearson Education, Inc. 21-20

Regression Analysis (4 of 4)
• Most theories in economics are more complicated than
one variable affecting another.
• We can add one explanatory variable to the voting
example above:
V a  bg c p

where p = inflation

• The fitting idea works the same way: The least squares
estimates of the three coefficients (a, b, and c) correspond
to the smallest value of sum

Copyright © 2017 Pearson Education, Inc. 21-21

REVIEW TERMS AND CONCEPTS
• correlated

• difference-in-differences

• intention to treat

• least squares estimates

• p-value

• regression discontinuity

• selection bias

• statistical significance

• survivor bias

Copyright © 2017 Pearson Education, Inc. 21-22