write 5 questions

Please read the lecture note I provided below. After reading, please come up with 5 questions with answers using the lecture note. Answers for the Questions must be short answer (2-3 sentences) or short paragraph (about 5 sentences).

Sample question:

For short paragraph: Why might e-monitoring be more successful than traditional monitoring schemes at improving public service delivery? Give one example from class readings.

For short answer:1. Was the Bjorkman and Svennson study of community health monitoring in Uganda found to be externally valid?2. Does monitoring of teachers in India work without incentives?

LEC 1 COMMUNITY MONITORING

Power to the people...

· Evaluates a randomized field experiment on community-based monitoring of public primary health care providers in Uganda 


· Local NGOs conducted two rounds of village meetings aimed at: • Encouraging communities to monitor public health providers
 • Strengthening their capacity to hold these providers accountable 


· In these meetings they discussed baseline levels of health provider performance relative to other organizations and government standards, and they encouraged community members to make a plan of action. 
• Not much data on these meetings as the researchers didn’t want an external presence. 


· A year after the meetings, treatment communities were more involved in monitoring their health providers.

• And these providers were performing better in terms of real outcomes—reduced child mortality (33%!) and increased child weight.

This has been a very influential study

• It’s the result we all wanted---with a cheap and quick intervention, communities can make a plan to monitor their own service providers and will see huge gains.

· Some thought the results were too good to be true. 


· The authors shared their data and their results held up---no cherry picking and/or p-hacking. But it is a small sample (50 health centers).

· But others tried similar things and didn’t find similar impacts---for example, grassroots monitoring of road building in Indonesia didn’t work when government audits did. 


· Along come Parkerson, Posner, and Raffler... 


Parkerson, Posner, and Raffler (PPR)

• They decide to test the generalizability of Bjorkman and Svensson (BS) by replicating the exact same intervention again in Uganda, but with a larger sample (376 centers) and with greater attention to “considerations”.

Unpacking the intervention

· People have hypothesized that the information given to villagers in BS’s study could be what is driving the results, rather than the community meeting component of the intervention. PPR designed their study to test this directly. 


· They use a “factorial design” = they break the intervention into two factors (information and meetings) and randomly provide one, the other, both, or neither to villages. This allows them to understand which of the two factors is more important for the overall impact. 


· By comparing villages that received both information and meetings to those who received neither, they can also still test the generalizability of BS.

So what do they find?

· Positive impacts on treatment quality and patient satisfaction. 


· No effect on health utilization rates. 


· No effect on health outcomes (including child mortality). 


· It does seem like information was more important than community monitoring when doing factorial analysis. 


Conclusion

· Power to the people---community monitoring of public health centers 


in UG reduced under-5 child mortality rates by 33%!

· But is this result externally valid? 


· Parkerson, Posner, and Raffler implemented the same intervention in Uganda in a larger sample and found no impacts on health. 


• Also seems information was more important than community meetings.
• If the result had replicated, we would have likely stopped and been happy. Since it didn’t, maybe we need more results! • Examples of `more results!’---metaketa, microfinance

LEC 2 E-MONITORING 1

E-monitoring Level 1: Cameras

Incentives Work: Getting Teachers to Come to School by Duflo, Hanna, and Ryan

· 35 percent absence rate of teachers at baseline in Seva Mandir schools in Rajasthan, India. 


· 57 randomly selected schools were given a camera to take pictures of teachers and students in class to prove attendance. 
• Tamper-proof data and time functions
• Had to take opening and closing school day photos 


· Teachers were paid according to number of “valid” days (5 hours+ with two photos). 


• They find monitoring + incentives had large impacts on teacher attendance. • They get fancy (structural model) and determine the effect is driven by the

incentives. Hence the title.
• Note: this study monitors “para-teachers” ---not protected by unions.

E-monitoring Level 2: Smartphones

Data and Policy Decisions by Callen, Gulzar, Hasanain, Khan, and Rezaee

Doctor attendance and politicians

· We measure absence in 850 (34%) of clinics spanning 240 MPA constituencies. 


· We interview 541 of 560 doctors posted to these clinics 


· We visit each clinic three times (Nov. 2011, Jun. 2012, Oct. 2012) 


· We find, before our intervention: 
 • Doctors are present 1/3 of the time 
 • Attendance falls 40 percent as you move from high to low political competition 
 • Doctors who know their politician show up to work 21% less often. 


Results

• We find large impacts on inspections, but no average impacts on attendance.

• But we do find impacts in certain areas / with certain doctors:
• Monitoring => greater attendance in politically competitive constituencies • Monitoring => greater attendance for doctors who do not know their MPA • Monitoring => greater attendance for doctors with `better’ personality

• What are the channels by which these impacts occur (or not)?...

E-monitoring Level 3: Dashboards

Results

Results

Perhaps the key: this stuff is cheap. So e-monitoring is taking off, research be damned.

In PK alone, there are e-monitoring platforms for health, education, police, judges, vaccinations, trash collection, vets, ...

E-monitoring Level 4: Biometrics

The Devil is in the Details by Dhaliwal and Hanna

Results

· Conducted RCT of IMIDSS in Karnataka, India with 322 primary health centers. 


· Biometric monitoring increased overall health staff attendance from 40 to 49%, though no increase for doctors. 


· This led to a decrease in low birth weight babies in treated areas. 


· Could this have worked better?
 • No punishments for those not showing up. 


LEC 3 CONSIDERATIONS WHEN CODUCTING RCT

Using the livestock project in Pakistan as an example

Choosing the right question to test / Finding a specific context in which to test the question

· We really wanted to evaluate the Citizen Feedback Model but couldn’t. 


· So we pounded the pavement to find a context that was open to a similar evaluation. 


· Punjab’s Livestock and Dairy Development Department had a great Secretary (at the time). 


· So we made the question right for the context. 


Groundwork to hone the intervention / Piloting interventions

· What we had in mind and what worked in practice were two very different things. 


· We spent a lot of time in rural Punjab with veterinarians and farmers talking and testing our ideas. 


· We learned AI is more important than vaccinations for livestock. 


· And vet effort was more important than bribes for AI. 


· Then we had to get farmers to answer the phone. 


Random assignment

• We randomized at the farmer level rather than the vet level for power reasons.

• Power measures the probability that we will fail to reject the null hypothesis of no impact of the intervention when in fact there is a difference between the treatment and comparison groups (Running REs pg. 253).

• But this created a lot of additional considerations. • Information spillovers.

Data collection plan

• We got all this great administrative data from our intervention itself. But we couldn’t trust it---reporting bias.

• So we conducted our own surveys as well.

• A lot went into this.
• How do we select villages for our sample?
• How do we select households in each village to interview?

Implementing the intervention

· A lot went into this too---purchasing equipment, designing apps and websites, training vets...


· Piloting was key. It took a while. 


· And our implementing partner changed their tune mid-way through. This was bad news bears

Analyzing the data

· If you make it this far, you’re in great shape. Most don’t. There’s entire books on failed field experiments---Failing in the Field. 


· Measuring main treatment effects was straightforward. 


· But using analysis to understand why we saw the effect that we did 
was not. This is common. 


· Research transparency is important. 


LEC 4 DEVELOPMENT ENGINEERING

Reading: Toward a new field of development engineering: linking technology design to the demands of the poor

BY Lina Nilsson, Temina Madon, and S Shankar Sastry. Procedia Engineering, 2014.

Development engineering what?

• This article makes the case for a new field of research---Development Engineering.

• Why engineering?

· “Engineering has the potential to accelerate the development of low-income communities by integrating insights from the social sciences along the entire arc of technological innovation, from idea to manufacture at scale.”

• Why development?

· “Engineering innovation must be tightly coupled with the design of interventions that address persistent institutional gaps, market failures, cultural constraints, and behavioral biases—all of which prevent poor people from accessing and adopting technologies that could significantly improve their lives.”

The role of missing markets and institutional failures

• Missing markets:

• Poor people are often priced out of the market for new products and services.

· Credit is hard to secure; information is lacking => risky to invest

· This is exacerbated by weak infrastructure.

· So poor people are left out of iterative design and don’t end up with products made for them.

• Institutional failures:

• Many new products and services would be administered by the public sector in developing countries.

Now back up a sec! This is not a new idea.

• People having been thinking about “technology for development” since the 1970s, focused on low cost, simply designs as a way to try to get the products right for folks in developing countries.

• Development engineering is different in two ways: 1. Bring in the economists! (And other social scientists.) We’re good at working with governments and citizens, with large-scale surveys and experiments, etc.

2. “Reconfigure the full arc of technology development”---i.e. bring these social scientists onboard from the beginning and throughout the whole process

How do we do this?

• How do you create and support a new academic discipline? • New engineering tools and approaches (e.g. biomedical engineering)

• Mechanisms for government sponsorship (e.g. USAID -> DIL) • A Whitaker Foundation for Development Engineering (again e.g. bioengineering) • International and multidisciplinary collaboration (e.g. information science) • Career trajectories and performance metrics (e.g. DevEng journal) • Partnerships beyond the ivory tower

A concrete example

Community Cellular Networks in the Philippines

Motivation: The Mobile Revolution

• Much excitement about the “mobile revolution” • In last decade, 2 Billion new subscribers in developing world

Motivation: The Last Mile

• This popular narrative masks an inconvenient truth: the missing “Last Mile” of connectivity

• 800 million people live outside of cellular network coverage • Not profitable for commercial telcos (sparse, rural, poor)

Motivation: Why this matters

• Connectivity matters

• A growing body of research indicates positive social and economic impacts of mobile phone connectivity

Community Cellular Networks: Idea

• Low-cost, locally-operated cellular networks

• Reduce cost of deployment by 10X

Commercial cellular ($250k/tower) Community cellular ($20k/tower)

Community Cellular: Overview

1. Technological innovation

• The “village base station” 2. Community ownership

• Any community can be their own telco

3. Social impact

• Rigorous impact evaluation

Reading: Fighting poverty with data

Joshua Evan Blumenstock. Science, 2016.

Machine learning algorithms measure and target poverty

• Josh Blumenstock offers the “perspective” that

• Cites study combining satellite imagery and machine learning to predict poverty.

Goals of this Talk

1. Convince you that there is exciting work to be done at the intersection of machine learning and economic development

2. Give a (shallow) deep dive into an example application, to illustrate a few “low hanging” opportunities and deeper challenges

AI & Development: Context

• The “Big Data Revolution”

· Mobile phones: 96% globally

· Facebook: 2.2B active monthly users

· Twitter: 330M active users (500M Tweets/day)

· WhatsApp: 1.3B monthly active users

· Sensors: 1000’s of satellites, traffic cams, etc.

“Big Data:” A first world problem?

• Compare successive rounds of Angolan census:

1970census:5.6Million 2014census:24.4Million

Predicting Poverty: Research Question • Can novel data and methods create new options for measuring poverty & vulnerability? obile Phone Data Online Mapping Data

Predicting Poverty: What Data?

• “Big” datasets still rare in poor countries • With some prominent exceptions:

• 3.5 Billion mobile subscribers in developing countries • 1000’s of satellites in earth’s orbit

Predicting Poverty: Our Approach • Supplement digital traces w/ traditional surveys

• Use machine learning to detect signals of poverty in mobile phone usage logs

Data

Predicting Poverty: Who cares?

• But are these methods actually useful?

Applications: “Band-Aid” statistics

These methods produce statistics roughly as accurate as a 5-year old national survey

100X cheaper, 10X faster

Applications: The Future

• Detecting changes in welfare, rather than levels. If successful, this can enable several exciting applications:

1. Better targeting of aid, insurance

2. Improved crisis response

3. New approaches to impact evaluation

Impact of shocks on “Percent of Calls Initiated”

Predicting positive cash shocks

• Novel data and methods create new options for understanding causes and consequences of poverty

Two points to take away from this talk:

1. Convince you that there is exciting work to be done at the

intersection of machine learning and economic development

2. Work involves plenty of “low hanging” opportunities, as well as many deeper challenges

LEC 5 THEORIES OF SUB-NATIONAL CONFLICTS

The prevalence of conflict

• Sub-state conflict and civil wars are common:
 • 1.5 billion people live in countries affected by political violence (World Bank, 2011).

• Since 1960, one third of all nations have experienced a civil war (Blattman and Miguel, 2010).

The persistence of conflict

• Conflicts are also highly persistent:


· 20 percent of nations have experienced at least 10 years of civil war since 1960 (Blattman and Miguel, 2011).

· Most modern conflicts are in countries have a prior conflict since 1970 (World Bank, 2011). 


· Almost all of the 39 countries experiencing a civil war since 2000 also experienced one in the three prior decades (World Bank, 2011). 


The relevance of conflict for development

• Conflict is empirically related to development (and service delivery) failures. People in conflict-affected countries are:

· More than twice as likely to be malnourished.

· Three times as likely to miss primary school.

· Almost twice as likely to die in infancy as people in other developing countries (World Bank, 2011).

Development spending shifts towards fragile states

UK Department for International Development (DFID) 5-year plan:

· ~$12.5B / year

· 22 of 28 priority countries are “fragile or conflict affected”

Two motivations:

1. Traditional development

2. Development as a tool of counterterrorism and counterinsurgency
note: these are the most common sources of modern warfare,
4x as prevalent as interstate war

Problem:


· People and property are not safe, so 1st welfare theorem doesn’t apply

· In particular, aid may decrease welfare
by increasing conflict (Collier, Nunn & Quian)

Put it all together

• Conflict is an important topic for policy research because:

· It is common.

· It is persistent.

· Empirically it is closely linked to worse service delivery outcomes.

· By revealed preference, these are the countries of most interest to aid agencies.

The neglect of conflict as an issue for development

· In 2007, Blattman and Miguel surveyed 63 development economics course syllabi from top universities. 


· Only 13 percent of undergrad and 24 percent of grad courses mentioned conflict at all. 


· Yet there is considerable work on the causes and economic consequences of development. 


What causes war?

Types of conflict

· Interstate war:
• Focus of much of the research in International Relations
• Many of the ideas developed here apply to subnational conflict. 


· Civil war/conflict---internal conflicts that count more than 1,000/25 battle deaths in a single year 
• The most prominent type of subnational conflict
• Often marked by insurgents battling national governments 


· Irregular war---expression largely applied to foreign involvements in fighting insurgencies 


· Other types---Terrorism, coups, communal violence, crime 


Cross-national evidence

• Collier and Hoeffler, 1998; 2004:


· Grievances against national governments exist everywhere. • Economic incentives to rebel, however, do not.
 • These factors are decisive.

• Fearon and Laitin, 2003:

· Proxies for “grievances" such as ethnic and cultural diversity have little predictive power.

· Rough terrain is a consistent predictor.

· Argue that war is a result of “poverty—which marks financially and bureaucratically weak states and also favors rebel recruitment.“

• Miguel, Satyanath, and Sergenti, 2004:

· Uses rainfall as instrument for income growth.

· A 5 percent drop in income growth increases the likelihood of civil conflict in the following year by up to 10 percentage points. 


· It is not clear what is driving this result (opportunity cost of fighting, crop failure reduces government revenue, state capacity, etc.). 


Rationalist explanations for war (Fearon, 1995)

• The central puzzle about war is that it is extremely costly, but nevertheless incurs frequently.

• Possible explanations:

· National leaders and the state are irrational.

· Leaders who enjoy war enjoy its benefits but do not pay the costs, which are suffered by soldiers and citizens.

· “Even rational leaders who consider the risks and costs of war may end up fighting nonetheless.”

· We will focus on these rationalist explanations.

Rationalist explanations

• Private information:


· Leaders have private information about their military capabilities and incentives to

· misrepresent those capabilities.

· Misrepresentation may lead to a better deal, but might also cause negotiation to break down.

• Commitment problems:
 • Cannot attain an outcome that both would prefer because, at some stage, one side has incentives to renege on the deal.
 • Think of de-escalation/disarmament. This means making yourself vulnerable.

• Issue indivisibilities:

· Some issues, by their very nature, do not admit comprise. If an issue allows only a finite number of resolutions, it may be that none falls within the range that both parties prefer to fighting.

· Fearon argues this is a less likely explanation because bargains are generally over many dimensions.

Theories of sub-national conflict

Understanding the logic of rebellion

· It is important to understand the decision to rebel (and to support rebels). 


· This is the main type of conflict affecting weak states. 


· The challenge is to understand how state fragility is related to insurgency. 


Main theories of insurgency

1. Opportunity cost---insurgency is a job like any other. A better economy means more jobs and fewer insurgents and less violence. 


2. Appropriation/rent capture---in countries where property rights are weakly enforced, it is easy to steal using violent means. 


3. Hearts and Minds---civilians need services. Conflict reflects a violent contest to gain the legitimate right to govern territory. 


Theory 1---Opportunity cost

• Based on the Becker-Stigler model of crime.

• Insurgents are insurgents because the benefits outweigh the costs.

• Increasing employment should reduce violence:

· Insurgency is a full-time occupation.

· And it is low-skilled, so creating jobs for the marginally unemployed reduces recruits.

· Less recruits means less violence.


· (This assumes the supply of labor is a binding constraint for insurgencies.)

Theory 2---Appropriation/rent capture

• Violence is directed at the capture of economic rents.

• Dube and Vargas (2013) find evidence of this with conflict in Colombia:

· Increase in leading labor-intensive export (coffee) reduces violence in coffee- growing regions.

· Increase in capital-intensive good (petroleum) increases conflict in petroleum producing regions.

Theory 3---Hearts and Minds

Reading:Can Hearts and Minds be Bought? The Economics of Counterinsurgency in Iraq

Eli Berman, Jacob N. Shapiro, and Joseph H. Felter. Journal of Political Economy. 2011.

Introduction

· Same motivation as earlier in this lecture---there is lots of conflict; huge amounts of aid have nbeen directed at it. 


· Yet little empirical research has evaluated the impact of aid in these settings “to see where, when, and how efforts to improve material conditions in conflict zones actually enhance social and economic order.” 


· At the same time, conflict is shifting away from force-on-force battles and more towards various forms of insurgency and irregular warfare. 


· The US Army’s counterinsurgency doctrine “places a heavy emphasis on influencing “human factors,” for example, the population’s tolerance for insurgent activities, by combining benign measures such as economic reconstruction with carefully targeted strikes against violent actors.” 


· But there is little evidence on when and why benign measures work. 


· Possible mechanisms by which benign measures (reconstruction) could decrease insurgency: 1. Addressing grievances 2. Raising the opportunity cost of rebellion by improving labor market opportunities 
 3. Buttressing local allies 


· But there is little systematic evidence for any of these hypotheses. 


· And none match with counterinsurgency doctrine. 


· So the authors develop a fourth approach to understanding the impact of reconstruction: 


“Hearts and Minds”

· Three-way game between insurgents, government, and noncombatants 


· Noncombatants make the consequential choice of whether to share information with government about insurgents, which decides who controls their neighborhood. 


· Insurgents choose level of violence 


· Government chooses level of coercive force, and public good provision 


· Model is a variation on Akerloff and Yellen (1994) on gangs 


· Tested with Iraqi data on reconstruction spending and violent incidents, 2004-2008 


Predictions of the theory

Data

· “The attack data are based on 193,264 “significant activity” (SIGACT) reports by Coalition forces that capture a wide variety of information about enemy attacks that target Coalition forces, Iraqi forces, civilians, infrastructure, and government occurring from February 2004 through December 2008.” 


· “Our key independent variable is spending on reconstruction projects. Data were compiled from the U.S. Army Corps of Engineers Gulf Region Division’s Iraq Reconstruction Management System (IRMS). These data are unclassified and include the start date, end date, project description, funding source, type of project, and amount spent for 62,628 reconstruction projects active from March 2003 through December 2008. They cover over $25.3 billion in projects...” 


· (The military collects good data.) 


Results

Conclusion

· CERP-spending supported decreased violence. 


· The decrease in violence was consistent with the Hearts and Minds 
theory of insurgency. 


· At the same time, the vast majority of reconstruction spending (CERP was only 10%) had no violence-reducing effect. 


· A word of caution---an observed positive trend between service provision and violence does not imply service provision makes things worse. 


Reading:Will There Be Blood?

Oeindrila Dube and Juan F Vargas. Commodity price shocks and civil conflict: Evidence from Colombia. The Review of Economic Studies, 80(4):1384-1421, 2013.

The relationship between income and violence

· On one hand, a rise in income may reduce conflict by increasing wages and decreasing labor supplied to conflict (opportunity cost theory). 


· On the other hand, “more income means there is more to fight over” (appropriation/rent capture/predation theory). 


· Maybe it’s not always one or the other, but that the type of income shock matters. 


· The authors study income shocks driven by commodity price shocks for two commodities in Colombia: 
 • Coffee • Oil 


What do Dube and Vargas find?

• For coffee, “a sharp drop in the price of coffee during the 1990s increased violence disproportionately in municipalities cultivating more coffee.”

· “The 68% fall in coffee prices over 1997 to 2003 resulted in 18% more guerrilla attacks, 31% more paramilitary attacks, 22% more clashes, and 14% more casualties in the average coffee municipality, relative to non-coffee areas.”

· Wages and work hours decreased in the same areas violence went up.

• For oil, “a rise in oil prices led to a differential increase in conflict in the oil municipalities.”

· “The 137% increase in oil prices over 1998 to 2005 led paramilitary attacks to increase by an additional 14% in the average oil producing municipality. The oil shock also increased municipal revenue generated from taxing natural resources, and kidnapping of politicians and leaders.”

These results generalize to other commodities

• They find a negative relationship between agricultural prices and conflict in the case of sugar, banana, palm, and tobacco.

• They find a positive relationship between natural resource price shocks and conflict for coal and gold.

How can we reconcile different relationships for different commodities?

· Let’s assume price increases both lead to decreased conflict from increased wages and increased conflict from having more to fight over simultaneously. 


· For any given commodity, one effect will be larger than the other. 


· For commodities that use relatively more labor, the wage effect (decreased conflict) will be greater. 


Reading: What if the “commodity” is aid?

Aid for Peace?

Eli Berman, Joseph H. Felter, and Jacob N. Shapiro. In Foreign Affairs. 21 January 2015.

This summarizes 10 years of ESOC research

· Focuses on lessons from the Philippines 


· “On the broadest level, we found that the type of the aid program 
matters greatly.” 


· Certain kinds of assistance—targeted, low-profile, conditional transfers delivered directly to needy families—appear to decrease conflict.” 


· By contrast, large and more high-profile projects, such as initiatives to improve infrastructure, may empower insurgents and exacerbate hostilities.” 


Do you think this is consistent with Dube and Vargas?

Finding one---Nathan Nunn and Nancy Qian

· The authors study the affect of food aid on violence cross-nationally. 


· Increases in US food aid (because of changes in US wheat production that should not be correlated with violence in other countries) increase both the incidence and the duration of civil conflicts. 


· This is consistent with the appropriation theory. 


· No labor required to “make” food aid = support for Dube and Vargas? 


· [Dube and Naidu find US military aid increased paramilitary attacks in Colombia around elections]. 


Findings two – like 10 from the Philippines

· #2: violence increased with employment (this one is really a puzzle). 


· #3: Increased local investment via building permits led to greater 
violence (appropriation?) 


· #4: CDD => significant increases in casualty rates =(. 
• Seems to be more about insurgent groups wanted to sabotage state- sponsored projects. Thus another theory---violence to stop economic development. 


· #5: World Bank CCTs to households decreased violence (something worked!) and decreased insurgent group influence. 


· Oh and Hearts and Minds. 


Reading:Do working men rebel? Insurgency and unemployment in Afghanistan, Iraq, and the Philippines

Eli Berman, Michael Callen, Joseph H. Felter, and Jacob N. Shapiro. Journal of Conflict Resolution, 2011.

This is really testing a theory of insurgency

· This paper can be thought of a companion piece to the Hearts and Minds paper. 


· In that paper, Berman et al argue that their Hearts and Minds model is supported with reconstruction spending on public goods in Afghanistan. 


· At the same time, one of the de-facto models of insurgency--- opportunity costs---is not supported by the data. 


Why might we be skeptical about the opportunity cost theory of insurgency?

• Implicit assumptions in the theory:


• Participation in insurgency is a full-time occupation.
 • Insurgency is a low-skilled occupation.
 • The supply of labor is a binding constraint on insurgent organizations.

• We might have reason to be skeptical of these.


• At the same time, we can think of reasons why there may be a negative relationship between opportunity cost and insurgency:


• It’s cheaper for the government to buy information from noncombatants. • Security efforts like checkpoints could increase unemployment.

• Or what if we think of insurgency as a normal good? People would buy less when they have lower opportunity costs.

And now there’s a model

• They present Becker’s (very simple) 1968 theory of crime.

• They then talk through how alternative explanations would bias the coefficient in:

• 𝑣𝑣 = 𝛼𝛼 + 𝛽𝛽𝑢𝑢 + 𝜀𝜀 𝑟𝑟𝑟𝑟𝑟𝑟

· For example, if violence increases unemployment, then the estimate of 𝛽𝛽 will be biased upward. 


· Other potential OVBs:
 • Economic rents (negative bias) 
 • Security measures (negative bias)
 • Information costs decrease with unemployment (negative bias) 


Data from three countries

• It was clearly a heavy lift to get this much data. Sources are different for every country and it’s hard to find good geographic measures of unemployment.

Results

Mo’ data, mo’ problems

• What do you make of these results? Most coefficients are negative, most significant, but not all and the controls and regions studied matter quite a bit.

• In their words---

Let’s formalize this “the power lies in their combination” argument

· They jointly test the null that 


· How they think about it: 
 • The probability of falsely rejecting the null in each country is simply the product of the probability of the individual events occur. And we know these probabilities---they’re the p-values from the results table. 
So the joint probability is 1 - .1*.01*.1 = 99.99 percent. 
This assumes each event is independent, but they may be correlated since in all cases the US was involved, for example. But accounting for correlation only moves the joint probability to 97.8 percent. 


· They’re conclusion: “The likelihood of falsely rejecting the null of a nonnegative relationship between unemployment and violence, when these results are combined across countries, is very small.” 


This means increasing employment increases violence. That’s not good!

Not so fast.

What matters is what caused unemployment rates to change.

· “These results do not imply that policies that increase employment rates cause violence, since the variation in unemployment rates that is negatively correlated with violence is not necessarily due to exogenous changes in labor demand.” 


· The enhanced intrusive security efforts argument is a good example here. 


· But we still should not assume that job creation policies actually decrease violence given the results here. 


· So what’s going on?!? 


Now let’s try to explain these results

· These results are inconsistent with opportunity cost being the dominant mechanism at play. 


· They are consistent with: • Predation 
• Security effects
• Information (Hearts and Minds) 


· The authors then show that when unemployment is high, insurgents switch tactics to those that reveal less information but are less precise (and kill more civilians). 
 • This is consistent with security effects or information and neutral to