Nicolas Hérault & Nicolás Salamanca Assignment 2, ECOM20001

ECOM20001: Econometrics 1

Assignment 2: Who Wants to be a Billionaire (in a non-linear world)?

Student
Tutorial Day & Tutorial
Name ID Tutor
Time Location
Number

Nicolas Hérault 321415 Friday

1
Nicolas Hérault & Nicolás Salamanca Assignment 2, ECOM20001

Assignment 2: Suggested Solutions

1. Summary stats for the entire dataset:

Typical country has about 8.9 billionaires, generates 25 billion dollars of natural resources
rents, has a population of 41.1 million people, a GDP per capita of $29,426, and a rule of law
index of 0.67.
In terms of scaling for regressions, natural resources rents will be divided by 1,000,000,000
so it will be in terms of billions of dollars, and population will be scaled by 1,000,000 so it
will be in millions of people.
2. Scatter plots produced on the next page and also include lines of best fit to highlight the
relationship.
• They reveal a positive relationship between the number of billionaires and population
size, a weakly positive relationship between the number of billionaires and natural
resources rents, and a positive relationship between natural resources rents and
population size.
• A single linear regression of the number of billionaires on natural resources rents would
therefore have an upward biased coefficient if population size was left in the error term.
• A positive regression coefficient on natural resources rents in such a regression could be
driven by a true positive relationship between the number of billionaires and natural
resources rents, OR by the fact that the omitted variable in the regression, population
size, simultaneously causes the number of billionaires and natural resources rents to
increase.

2
Nicolas Hérault & Nicolás Salamanca Assignment 2, ECOM20001

3
Nicolas Hérault & Nicolás Salamanca Assignment 2, ECOM20001
3. Scatter plots produced on the next page and also include lines of best fit to highlight the
relationship.
• They reveal a weakly positive relationship between the number of billionaires and GDP
per capita, a positive relationship between the number of billionaires and the rule of law
index, and a positive relationship between GDP per capita and the rule of law index.
• A single linear regression of the number of billionaires on GDP per capita would
therefore have an upward biased coefficient if the rule of law index was left in the error
term.
• A positive regression coefficient on GDP per capita in such a regression could be driven
by a true positive relationship between the number of billionaires and GDP per capita,
OR by the fact that the omitted variable in the regression, the rule of law index,
simultaneously causes the number of billionaires and GDP per capita to increase.

4
Nicolas Hérault & Nicolás Salamanca Assignment 2, ECOM20001

5
Nicolas Hérault & Nicolás Salamanca Assignment 2, ECOM20001
4. Regression results are presented in the table below.

6
Nicolas Hérault & Nicolás Salamanca Assignment 2, ECOM20001
Answer to question 4:
• The coefficients across the column for natural resources rents range from -0.001 to 0.033
depending on the set of controls but the two negative estimates (-0.001 and -0.005) are not
statistically significant. Including population in the regression (columns 2 & 3) causes the
coefficient to become slightly negative and not statistically significant. But when we then add
the rule of law index (columns 4 & 5), the coefficient’s statistical significance is restored.
In columns 2 and 3 the coefficient is not statistically different from zero but the other columns
show a coefficient for natural resources rents that ranges from 0.022 to 0.033 implying 2.2 to
3.3 more billionaires from a $100bn increase in natural resources rents (or 0.022 to 0.033 more
billionaires from a $1bn increase).

• Interpreting a 1-unit change in the rule of law index makes less sense: the index runs from 0
to 1, which means a 1-unit change means changing from minimum to the maximum value, an
extreme change. It makes more sense to use a 0.1-unit change, which means a 10 percentage
points change on the rule of law 0-1 scale.
• Comparing columns (3) and (4) clearly highlights the substantial impact of controlling for the
rule of law in isolating the relationship between the number of billionaires and GDP per capita;
if it is not controlled for, it causes substantial upward bias in the estimated effect.

• The coefficient estimate on the 2013 year dummy is 6.1 and statistically significant at the 1%
level, indicating that, holding other factors constant, there are 6.1 more billionaires per country
in 2013 than in 2005, the base year.

7
Nicolas Hérault & Nicolás Salamanca Assignment 2, ECOM20001

5. Regression results are presented in the table below.

Answers to question 5:
• In the linear-log model in column (1), the interpretation is that an increase in population size
by 1% leads to 0.01 x 3.866 = 0.03866 more billionaires, a statistically significant effect at the
1% level.
• In the log-linear model in column (2), the interpretation is that an increase in population size
by 1 million leads to a 1 x 0.010 x 100 = 1% increase in the number of billionaires, a statistically
significant effect at the 1% level.

• In the log-log model in column (3), the interpretation is that a 1% increase in population size
to a 0.442% increase in the number of billionaires, again a statistically significant effect at the
1% level.

8
Nicolas Hérault & Nicolás Salamanca Assignment 2, ECOM20001

6. Regression results in terms of the R Studio output:

• The coefficient on log(pop) alone is 0.41 and statistically significant at the 1% level,
which given 2005 as the base year, implies a positive and statistically significant
relationship between the log of population size and the number of billionaires in 2005, with
the number of billionaires increasing by 0.41% from a 1% increase in population.
• None of the log(pop) and year dummy variables interactions is statistically significant,
implying that there is no evidence of a stronger or weaker relationship between the log of
population size and the number of billionaires in any of these years when compared to
2005.

• To obtain the partial effect of the log(pop) on the number of billionaires in 2013, we add
the base group and the 2013 log(pop)-year dummy interaction, implying an effect of
0.4126+0.0444=0.457. This implies that in 2013, a 1% increase in population size is
associated with a 0.457% increase in the number of billionaires. To test the statistical
significance of this effect we need to test the joint null that the sum of the coefficient
estimates on log(pop) + lnpop_d2013=0. Doing so with an F-test, we obtain an F-statistic
of 41.574, df1=1 and df2=423, and a p-value<0.00001, implying a statistically significant
relationship at the 1% level.
• The first way the dummy variable trap is avoided is to omit the dummy variable for 2005
from the regression, d2005; otherwise the sum of the year dummy variables in the

9
Nicolas Hérault & Nicolás Salamanca Assignment 2, ECOM20001
regression would equal the constant regressor for all observations in the sample, leading to
perfect multicollinearity.
• The second way the dummy variable trap is avoided is to omit the interaction of
lnpop_d2005 (the log population and year 2005 interaction); otherwise the sum of all the
log(pop) - year dummies interactions variables would equal log(pop) for all observations
in the sample, learning to perfect multicollinearity.
7. The joint test yields a F-statistic=0.1492 with df1=8 and df2=423, and a p-value=0.9966,
meaning we fail to reject the null at the 5% (or even 10%) level. In words, we fail to
reject the null that the change in the relationship between the log of population and the
number of billionaires in any year after 2005 is different from the relationship in 2005.
That is, we find no evidence that the relationship varies over time.
8. Regression results in terms of the R Studio output:

• The coefficients on log_gdppc and roflaw are statistically significant at the 1% level, while
the coefficient on log_gdppc_roflaw is significant at the 5% level. Their signs imply that the
log of the number of billionaires is increasing with log_gdppc and roflaw (positive signs), but
does so at a decreasing rate due to the negative sign on log_gdppc_roflaw. So as the rule of law
improves, the rate at which the number of billionaires increases with GDP per capita decreases.
9. Following similar calculations from textbook on page 320, “Case III: Both X and Y are
in logarithms”, you can show that the regression equation estimated in question 8 yields
an elasticity of the number of billionaires with respect to GDP per capita of

𝛽𝛽1 + 𝛽𝛽3 × 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟

where the regression coefficients estimates with the “1” and “3” subscripts are the for the
log_gdppc (0.682) and log_gdppc_roflaw (-0.422) terms in the regression. Using this formula,
we can compute the elasticity of the number of billionaires with respect to GDP per capita for
any given value of the rule of law index. To compute the p-value associated with the test of the
null that the elasticity equals 0, we specify a value for roflaw and conduct the following joint
hypothesis test using the F-test:

𝐻𝐻0 : 𝛽𝛽1 + 𝛽𝛽3 × 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 = 0; 𝐻𝐻1 : 𝛽𝛽1 + 𝛽𝛽3 × 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 ≠ 0

10
Nicolas Hérault & Nicolás Salamanca Assignment 2, ECOM20001
Listing the test results for the various roflaw values in the question we obtain the following
table:
Roflaw Elasticity F-statistic P-value
0.1 0.640 49.6 0.000
0.2 0.598 59.1 0.000
0.3 0.556 69.9 0.000
0.4 0.514 78.3 0.000
0.5 0.472 76.6 0.000
0.6 0.429 61.1 0.000
0.7 0.387 40.4 0.000
0.8 0.345 23.9 0.000
0.9 0.303 13.4 0.000
1.0 0.261 7.3 0.007

The elasticity is statistically significant at the 1% level at all levels of roflaw, which is hardly
surprising given the statistically strong effects found for log_gdppc, roflaw and
log_gdppc_roflaw in Question 8.
As expected from the coefficient estimates, the elasticity decreases as the rule of law improves,
though it always remains positive. An economic interpretation for this finding is that as the rule
of law increases in a country, there are more checks and balances and the capture of economic
resources by a small elite becomes less likely.
Alternatively, the interpretation could be that increases in GDP generate more billionaires in
countries with weaker democratic institutions as measured by the rule of law, as it facilitates
the capture of economic resources by a few.

11