Transcript Economic Shocks and Civil Conflict: An Instrumental Variables

Economic Shocks and Civil Conflict: An
Instrumental Variables Approach
• Civil wars (wars within states) have resulted in three times as many
deaths as wars between states since WWII
• It is important to understand the conditions that lead to civil war in
order to know how we might prevent it
• A growing body of research highlights the associations between
economic conditions and civil conflict, but this literature has not
adequately addressed the endogeneity of economic conditions
• During civil war the economy suffers, so it is not clear whether poor
economic performance causes civil war or the other way around
• In addition, omitted variables such as institutional quality may drive
both economic outcomes and conflict
Sub-Saharan Africa and Instrumental
Variables
• Sub-Saharan Africa has experienced many civil wars
• 29 of 43 sub-Saharan African countries suffered from civil conflict
during the 1980s and 1990s
• The authors focus on this region
• Their focus allows them to use exogenous variation in rainfall as an
instrumental variable for income growth in order to estimate the
impact of economic growth on civil conflict
• Weather shocks are plausible instruments for GDP growth in
economies that rely heavily on rain-fed agriculture (only 1% of
cropland in this area is irrigated and there is little industry)
• Weather shocks are unlikely to be directly related to civil conflict
shocks
Problems with Previous Studies
• Previous studies have recognized the endogeneity problem but they
have used worse instruments
• For example, one study uses lagged values of per capita GDP growth
or levels
• However, this approach implicitly assumes that economic actors do not
anticipate the incidence of civil war and adjust economic activity
accordingly
Results
• GDP growth is significantly negatively related to the incidence of civil
conflict in sub-Saharan Africa during the period 1981-99
• This result holds across a range of regression specifications, including
some with country fixed effects (that control for country-specific
variables such as institutional differences that do not vary over time)
• The relationship between GDP growth and the incidence of civil
conflict is extremely strong: a five-percentage-point drop in annual
economic growth (less than one standard deviation) increases the
likelihood of civil conflict (measured as at least 25 deaths per year) in
the following year by over 12 percentage points, which amounts to an
increase of more than one-half in the likelihood of civil war
• A similar relation holds for major conflicts (over 1000 deaths)
Results
• Other variables such as per capita GDP level, democracy, ethnic
diversity, and oil exporter status do not display similarly robust
relationships
• Further, the impact of income shocks on civil conflict is not
significantly different in richer, more democratic, more ethnically
diverse, or more mountainous African countries or in countries with a
range of different political institutional characteristics
Limitations
• The results do not isolate the path through which poor economic
performance affects civil conflict
• For example, one possible path is through micro-level labor market
considerations: young men face a choice between productive economic
activities (such as farming) and picking up arms; if the returns to
productive activities fall, picking up arms becomes more likely
• Another path is through state military strength: low national income
leads to weaker militaries and worse infrastructure and thus makes it
more difficult for the government to repress insurgencies
• Another path could be that negative income shocks widen income
inequality, which could heighten resentment and generate increased
tension
The Exclusionary Restriction
• In order to be appropriate instruments, rainfall measures must satisfy
the exclusion restriction: weather shocks should affect civil conflict
only through economic growth (and not directly)
• The authors make several arguments in favor of this assumption but
also admit there are some ways rainfall might have a more direct effect
• For example, floods might destroy the road network and make it more
costly for the government troops to contain rebel groups; this is not
serious because high rain is associated with less conflict in their
regressions
• Another example: Rainfall may make it difficult to fight because
transportation is difficult; the authors used data on the usable road
network from the World Bank to rule this out
Data
•
•
1.
2.
•
•
•
The authors use the Armed Conflict Data database developed by the
International Peace Research Institute of Oslo, Norway, and the
University of Uppsala, Sweden
This database records all conflicts with:
More than 25 battle deaths per year
More than 1000 battle deaths per year
The database focuses only on politically motivated violence that
involves the government
Like other conflict databases, this data does not provide the exact
number of deaths or disaggregate below the annual level
The authors focus on civil wars (conflict within a country)
Variables
• The unit of observation is a country in a particular year
• The dependent variable takes the value 1 if the country has a civil
conflict in progress with at least 25 deaths (or 1000 deaths) during the
year, and 0 otherwise
• In 1981-99, there was civil conflict in 27% of country-year
observations using the 25 death cutoff, and 17% using the 1000 death
cutoff
• They also examine conflict onset, where onset measures events where
a conflict occurs in a country in year t when there was no conflict in
year t – 1
• 38 separate conflicts began during 1981-99 (this does not count those
that were ongoing in 1981) and 27 ended (at least temporarily)
Rainfall Data and Other Variables
• The Global Precipitation Climatology Project keeps track of monthly
rainfall going back to 1979
• This data is reported at particular latitude and longitude degree
intervals
• They aggregate to the country-year level
• Their main measure of a rainfall shock is the proportional change in
rainfall from the previous year (R(i,t)/R(i,t-1) – 1)
• They include several other control variables: ethnolinguistic
fractionalization, religious fractionalization, measures of democracy,
etc.
Estimation Framework: First Stage
• In the first stage regression, the authors regress per capital economic
growth in each country-year on a country fixed effect, current and
lagged rainfall growth, a vector of control variables, and a countryspecific time trend:
growthit  ai  X b  c0 Rit  c1Rit 1  di yeart  eit
'
it
First Stage Results
• Current and lagged rainfall growth are both significantly related to
income growth at the 5% level, and this relationship is robust to the
inclusion of country controls and fixed effects
• The authors tried several other specifications to confirm that their
results were robust and that they could not come up with better
instruments by including further lags of rainfall growth, exploring
interactions between variables, etc.
Estimation Framework: Second Stage
• In the second stage regression, the authors regress the conflict indicator
variable each country-year on a country fixed effect, current and
lagged growth, a vector of control variables, and a country-specific
time trend:
conflictit  a2i  X it' b2  c20 growthit  c21 growthit 1  d 2i yeart  e2it
Second Stage
• They performed IV-2SLS estimation and a nonlinear two-stage
procedure following Achen (1986) to correct standard errors in the
presence of a dichotomous dependent variable in the second stage
• The IV-2SLS method is typically preferred even in cases in which the
dependent variable is dichotomous (Angrist and Kreuger 2001;
Wooldridge 2002) because strong assumptions are required to justify
the Achen approach
• The authors focus on the IV-2SLS; in any case, the results are similar
with both methods
• Using the linear model in the presence of a dichotomous (0-1)
dependent variable introduces heteroscedasticity
• Thus authors do not discuss this issue explicitly, but in all of their
econometrics they report Huber standard errors (which is another name
for White heteroscedasticity consistent standard errors)
Results
• For purposes of comparison, the authors report results from simple
OLS regressions and a Probit model
• In these cases, the coefficients on contemporaneous and lagged
economic growth rates are negative but small and not statistically
significant
• The IV-2SLS results are not much different for contemporaneous
economic growth, but the effect of lagged economic growth becomes
much stronger and statistically significant at the 5% level
• Control variable effects are mostly insignificant (the lagged logged
national population is the only significant control)
Further Analysis
• The authors interact the IV economic growth rate measures with the
controls to determine whether economic shocks have a greater or lesser
effect on the incidence of civil conflict in countries with different
political characteristics (democracy, ethnicities, oil producing,
mountainous, etc.)
• The interaction terms are not statistically significant
• This suggests that economic factors dominate the others and that
institutional and social characteristics within sub-Saharan Africa are
not strong enough to mitigate the effect of economic shocks
• These results may not generalize to other regions where institutions are
stronger
• Also, alternative or better institutional/social measures might have an
effect
Conflict Onset and Termination
• The IV growth measures are also significantly negatively related to
conflict onset and termination
• High current or lagged growth makes it less likely that conflict starts
and more likely that it ends
• In these regressions, as in essentially all of the others, the effects of
current and lagged growth are not statistically different from each other
• Given this, the authors do not emphasize whether current or lagged
growth is more responsible for the estimated effect on conflict,
although in most cases according to the point estimates one variable’s
effect appears stronger than the other’s (and more significant)