Causal Inference on

Human Society

with Natural Science Data

Masayuki Kudamatsu

Osaka School of International Public Policy (OSIPP)

Osaka University

16 May, 2018

Data Science

Prediction

Causal inference

Data Science

Prediction

Causal inference

Data Science

Prediction

Causal inference

Today's lecture:

3

examples of

how natural science data

helps economists

conduct causal inference

Suitability for Agriculture

Natural science data #1

Weather "forecast" in 1871

Natural science data #2

(The above image refers to the mean temperature for 1951-1980)

Potential crop yields (FAO-GAEZ)

Natural science data #3

These natural science data help us learn

1

2

A cause of ethnic diversity

Impact of Tsetse flies

on African economic development

3

Origin of states

Before diving in to these 3 examples...

Ordinary Least Squares (OLS) explained in 5 min.

Image adapted from Figure 3.1 of James et al. (2013)

X

Y

Y_i=\alpha + \beta X_i +\varepsilon_i
Yi=α+βXi+εiY_i=\alpha + \beta X_i +\varepsilon_i
\min_{\alpha,\beta} \sum_i \varepsilon_i^2
minα,βiεi2\min_{\alpha,\beta} \sum_i \varepsilon_i^2

Suitability for Agriculture

Natural science data #1

This data is created as follows:

[1] Obtain spatial distribution of croplands from satellite images

[2] Divide the earth into 0.5-degree cells

[3] Calculate % of cells cultivated

[4] Regress [3] on degree-days, moisture, soil carbon density, soil pH

[5] Predict % of cells cultivated from [4]

This data is created as follows:

[1] Obtain spatial distribution of croplands from satellite images

[2] Divide the earth into 0.5-degree cells

[3] Calculate % of cells cultivated

[4] Regress [3] on degree-days, moisture, soil carbon density, soil pH

[5] Predict % of cells cultivated from [4]

Process [4] removes

human factors for cultivation

e.g. wars, population growth

Suitability for Agriculture in Asia

Suitability for Agriculture in Oceania

Suitability for Agriculture in Africa

Suitability for Agriculture in Europe

Suitability for Agriculture in North America

Suitability for Agriculture in South America

In economics, it's been known that

ethnic diversity is associated with lower economic growth

In antholopology, some studies suggest

linguistic diversity is associated with ecological diversity

Stelios Michalopoulos,

an economist at Brown University,

provides systematic evidence

for language-ecology relationship

in his 2012 paper.

A seemingly irrelevant topic in social science:

Origin of Ethnic Diversity

Calculate the standard deviation of suitability for agriculture

within each country

Measuring ecological diversity for humans:

Count the number of languages spoken within each country

from the World Language Mapping System

Measuring linguistic/ethnic diversity

Example: Nepal vs Greece

Source: Figure 2 and page 1514 of Michalopoulos (2012)

# of languages

107

14

Nepal

Greece

Whole world plot

Source: Figure 4 of Michalopoulos (2012)

A case of everywhere cultivable

Japan

Source: Figure 4 of Michalopoulos (2012)

Saudi Arabia

Source: Figure 4 of Michalopoulos (2012)

A case of everywhere deserted

Ethiopia

Source: Figure 4 of Michalopoulos (2012)

A case of numerous ethnic groups in Africa

Senegal

Source: Figure 4 of Michalopoulos (2012)

A case of numerous ethnic groups in Africa

Source: Figure 4 of Michalopoulos (2012)

Swaziland

A case of ethnically homogenous Africa

Source: Figure 4 of Michalopoulos (2012)

Can we interpret this as causal?

Natural science data gives random variation in X

Image adapted from Figure 3.1 of James et al. (2013)

X

Y

Then we can claim that X causes Y

Image adapted from Figure 3.1 of James et al. (2013)

X

Y

Natural science data gives random variation in X

Image adapted from Figure 3.1 of James et al. (2013)

X

Y

Known as

"Natural Experiments"

Source: Figure 4 of Michalopoulos (2012)

Can we interpret this as causal?

Look at 2.5-degree cells, not at countries

Source: Map 1 of Online Appendix to Michalopoulos (2012)

Weather "forecast" in 1871

Natural science data #2

(The above image refers to the mean temperature for 1951-1980)

The 20th Century Reanalysis project

Create the start-of-the-art model of climate

at the spatial resolution of 2-degree cells across the earth

"Forecast" the weather 6 hours ago from the current weather

Correct the "forecast" whenever actual observations are available

Repeat this up to the year of 1871

A seemingly irrelevant topic in social science:

Tsetse flies 

& Africa's

economic development

Tsetse flies kill domesticated animals

Image sources: Encyclopedia Britannica and Wikipedia

Livestock is

historically important

for agriculture

Manure

Draft power

Tsetse flies

only found in Africa

Anthropologists and historians speculate

Tsetse flies were the main cause for

Africa's economic backwardness

Africa is historically...

agricultural productivity: low

population density: low

Marcella Alsan

health economist in Stanford University

provides systematic evidence

on the impact of tsetse flies in her 2015 paper

Tsetse flies survive the longest

under warm but not too hot temperature

Source: Figure 2 Panel A of Alsan (2015)

Tsetse flies die more likely

when the air gets drier

Source: Figure 2 Panel B of Alsan (2015)

Temperature & humidity

in 1871

Tsetse suitability index

Source: Figure 3 Panel A of Alsan (2015) and Wikipedia

Historical data on Africa: Ethnographic Atlas

Source: Figure 5 Panel A of Alsan (2015)

Spatially match these two datasets

Tsetse Suitability Index

Ethnographic Atlas

The more suitable for tsetse, the less developed

Source: Figure 4 Panel A of Alsan (2015)

The more suitable for tsetse, the less developed

Source: Figure 4 Panel A of Alsan (2015)

The more suitable for tsetse, the less developed

Source: Figure 4 Panel A of Alsan (2015)

The more suitable for tsetse, the less developed

Source: Figure 4 Panel A of Alsan (2015)

(Over 20,000)

The more suitable for tsetse, the less developed

Source: Figure 4 Panel A of Alsan (2015)

The more suitable for tsetse, the less developed

Source: Figure 4 Panel A of Alsan (2015)

Potential crop yields (FAO-GAEZ)

Natural science data #3

How potential crop yields are calculated

For 5 arc-minute cells across the earth

Daily average weather

temperature, precipitation, humidity, wind speed, sun exposure

Soil type, Elevation, Land gradient

For each crop (wheat, rice, potatoes, cassava, etc.)

Crop growth parameters

How sensitive to degree-days, water scarcity, etc.

at each of the four stages of crop growth

A seemingly irrelevant topic in social science:

Origin of states

Historians and anthlopologists debate

whether agriculture led to the emergence of states

A seemingly irrelevant topic in social science:

Origin of states

These four economists argue

it's appropriability of crop harvests that matters for states to emerge

Luigi

Pascali

(Pompeu Fabra)

Omer

Moav

(Warwick)

Zvika

Neeman

(Tel Aviv)

Joram

Mayshar

(Hebrew)

Lucrative

for the powerful

to collect as tax

CEREALS

Wheat, rice, maize...

Storable

Harvest within a short season

ROOTS AND TUBERS

Cassava, yam, taro, bananas...

Perishable upon harvest

Harvesting is non-seasonal

Difficult and costly

to collect as tax

States should have emerged where

cereals are relatively more productive

than roots and tubers

Measuring relative productivity of cereal crops

Convert each crop's

potential yield

into calory units

Step 1

Obtain the maximum caloric yield among cereals

Step 2

Obtain the maximum caloric yield among roots and tubers

Step 3

Step 4

Take the difference

Measuring relative productivity of cereal crops

Relative productivity of cereal crops

To measure the emergence of states

Use Ethnographic Atlas across the world

Statistical analysis of 952 ethnic groups reveals:

on average

# of administrative levels

1.89

2.13

Relative

productivity

of cereals

higher

by 1 s.d.

1.65

Relative

productivity

of cereals

lower

by 1 s.d.

Takeaways

Data science's most basic tool:

Ordinary Least Squares (OLS)

Image adapted from Figure 3.1 of James et al. (2013)

Y

X

Can be used for prediction and causal inference

Image adapted from Figure 3.1 of James et al. (2013)

Y

X

Need random variation in X for causal inference

Image adapted from Figure 3.1 of James et al. (2013)

X

Y

Nature offers random variation in X

Image adapted from Figure 3.1 of James et al. (2013)

X

Y

This lecture is based on the following academic articles:

Mayshar, Joram, Omer Moav, Zvika Neeman, and Luigi Pascali. 2015.

"Cereals, Appropriability and Hierarchy."

CEPR Discussion Paper, no.10742.

(Non-technical summary)

Alsan, Marcella. 2015.

"The Effect of the TseTse Fly on African Development."

American Economic Review, 105(1): 382–410.

(Summary by The Economist magazine)

Michalopoulos, Stelios. 2012.

"The Origins of Ethnolinguistic Diversity."

American Economic Review, 102(4): 1508–1539.

Tutorials on

computer programming for data science

Python

R

The Owen Guide (for free)

Economic Analysis of Natural Science Spatial Data

By Masayuki Kudamatsu

Economic Analysis of Natural Science Spatial Data

  • 1,517