## By Sonya Kalara

Do women win the Hoopes Prize, an undergraduate award for thesis-writing at Harvard College, more often than men? Does the gender makeup of the advising team affect the probability that their student will go on to win the Hoopes?

I break down the timeline between enrollment at the college and Hoopes acquisition into three stages: Pathway, Connection, and Evaluation.

• In the pathway stage, a student is enrolled at Harvard, they choose a concentration, and they choose to write a thesis. Whereas Harvard College at large has a 50/50 gender split, gender representation in different concentrations varies massively.
• In the connection stage, a student decides to write a thesis, and chooses their advising team. The literature on this topic indicates that students search for same-gender advisors and so we’d expect that gender “dyads” (m/f combinations) would exhibit some skew.
• Thirdly, in the evaluation stage, a student is nominated for the prize by at least one member of their advising team. Although the nomination phase is crucial, all information about Hoopes nominees is confidential, therefore this piece isn’t directly included in my analysis. Finally, a student is selected for the Hoopes Prize and we expect that this selection from the pool of nominated writers exhibits variance.

Pathway Stage

I compare the number of graduating seniors in each department to the number of thesis writers. This allows me to calculate the average number of women in a concentration that go on to write a thesis as compared to the average number of men. Surprisingly, for all 9 concentrations that I study, women are more likely to write theses than men. Although this result wasn’t statistically significant in all concentrations, the evidence suggests that the rate at which women write theses is not the broken rung in the Hoopes evaluation process.

Connection Stage

Figure 1 shows all of the different pairs between advising teams and students by gender. A further subdivision by concentration is included in the full thesis; the volume of each pair varies widely by department. However, this graph should demonstrate at a general level that there exists some assortative matching.

After establishing that this pattern of seeking same-gender advisors exist within specific concentrations, I used statistics to randomly assign advisors to students. I then compared this “predicted distribution” to the actual rates of student/advisor matching and I find that students search for same-gender advisors, and women are more likely to seek out female advisors in male-dominated concentrations.

Evaluation Stage

Figure 2 shows the proportion of Hoopes Winners in every concentration by gender. I use OLS to investigate how the probability of winning a Hoopes Prize differs by student and advisor gender. First, in most concentrations, there is some baseline probability that women with at least one female advisor will win the prize. The mathematical evidence is mixed regarding the relative (dis)advantage of being a male student and having an all-male advisor team doesn’t appear to be particularly advantageous for female students. Interestingly, it appears that any advantage male students have is almost entirely quashed if they have an all-male advisor team. This evidence indicates that the most productive pairing is between male students and female advisors.

# By Keshav Rastogi

Offshore Wind Turbines Have Rapidly Grown in Size

In my thesis, I explore the drivers of capital expenditure (CAPEX) reductions in the offshore wind energy sector. Identifying key drivers is important because of how quickly the international and domestic offshore wind industries are projected to grow. By the end of 2018, nearly 23 gigawatts (GW) of offshore wind capacity were installed globally, with 154 to 193 GW projected by 2030.[1]

Specifically, I investigate two hypotheses. One is whether there has been significant learning-by-doing among companies developing (planning and assembling) offshore wind farms and those manufacturing turbines. The other is whether technological innovations among turbine original equipment manufacturers (OEMs) have been primarily responsible for recent reductions in CAPEX. The primary technological advancement has been in the size of turbines, as offshore turbines have become continually larger than their onshore counterparts since the early 2010s. Learning-by-doing is the idea that productivity increases due to accumulated experience.

In order to understand whether there exists learning-by-doing in the offshore wind sector, I model the typical developer’s project design problem, which includes, among others, the variables of interest measuring developer experience, OEM experience, and turbine size. By exploiting the fact that developers frequently have limited agency over deciding the size of their projects, I solve for the developer’s profit maximizing decision and manipulate the resulting equation such that it can be empirically estimated by an OLS regression with fixed effects. I conduct the analysis using data from offshore energy consultancy 4C Offshore covering all fully commissioned, or completed, offshore wind farms. The preferred set of specifications, which control for market selection of low-cost firms, show no statistically significant evidence of learning-by-doing among either developers or turbine OEMs.

In contrast, the empirical results suggest that a doubling in average turbine rating, or size, is robustly associated with a 19 percent decrease in total costs, which corroborates the broad academic and industry-based consensus that larger turbines have been a key driver of CAPEX reductions. In my discussion of the results, I demonstrate how the cost-reducing effects of turbine rating nest within the developer’s decision regarding turbine size at a particular site. I then use these estimates to make projections for the costs of offshore wind farms currently under development globally. With regards to policy, the lack of significant learning-by-doing may make it more difficult to justify demand-based policies such as future installation targets, while the cost-reducing impact of larger turbines may rationalize expanded government-funded research and development efforts.

[1] Musial, Walter, Philipp Beiter, Paul Spitsen, Jake Nunemaker, and Vahan Gevorgian. 2019. “2018 Offshore Wind Technologies Market Report.” U.S. Department of Energy.  https://www.energy.gov/eere/wind/downloads/2018-offshore-wind-market-report

# By Dhruv Mohnot

Background

Social scientists have long been interested in voting as a phenomenon. There is a famous Georg Hegel quote as well: “The casting of a single vote is of no significance when there is a multitude of electors.” Why, then, do people vote? Theoretical work has failed to predict the high turnout we see in elections, since each individual vote is almost certainly inconsequential to the outcome. Empirically, several papers have found that social pressure influences voting patterns. By making voting behavior more salient, voting becomes more popular (internal/social benefits rather than external candidate-level benefits).

My research focuses on two major changes that plausibly influenced the 2020 presidential election outcomes. First, Black Lives Matter protests were a major national phenomenon that may have motivated individuals to vote (by, for example, making voting behavior more relevant in social contexts). Second, the expansion of universal mail-in balloting in several states due to the COVID-19 pandemic allows for a natural experiment investigating the effect of the cost of voting.

Using two quasi-experimental research designs (instrumental variables for protests & difference-in-difference for mail-in balloting), I estimate the effect of these changes on 2020 election outcomes (voter turnout and Republican vote margin).

Results

Interestingly, I find that protests had a very limited effect on both turnout and margin. A one-unit per capita increase in protest size increased Republican margin by 0.03 percentage points and turnout by 0.17 percentage points, a statistically significant result robust to many specifications and interactions. The effect was primarily concentrated in Democratic areas, suggesting that increases in Republican margin were primarily driven by a backlash against the movement. Notably, this is a very small effect size relative to prior papers investigating the effect of protests (e.g. Madestam et al. find huge effects of the Tea Party protests in 2009 on 2010 midterm elections).

Mail-in balloting, on the other hand, has a large effect. Counties that expanded to universal mail-in balloting in this election cycle saw increases in turnout by 2.52 percentage points and Republican margin by 0.45 percentage points, relative to counties that did not change mail-in balloting policies. This is primarily driven by higher voting rates by 65+ individuals in counties that had universal mail-in balloting, suggesting that older voters who feared COVID-19 infection only voted (primarily for Republicans) when voting was safe. Both changes together only explain about 9% of the overall change in turnout in 2020 and likely did not affect the overall election outcomes.

Discussion

There are several hypotheses for why protests had such a limited effect. My design estimates the locally specific effect of protesting. It is possible that protests were such a nationalized phenomenon that there was no additional locally specific effect of protesting. For example, protests in Portland were immediately broadcast via news networks and social media to all parts of the country, making the Portland specific effect next to nothing.

Alternatively, political polarization has made protests unlikely to change voters’ minds. Instead, national events are filtered through a partisan lens and only affect voter motivation. Republicans were motivated by seeing protests in their backyard, but voters did not change their minds at all.

Ultimately, my paper shows that making voting easier has a much larger effect on turnout than large-scale protests. Though it is unclear if mail-in balloting expansion will remain in place going forward, it does increase voter turnout, as theory would expect. On the other hand, BLM protests did not have a locally-specific effect in the short-term but may precipitate a longer-term realignment of race discussions in the country, leading to longstanding political changes over the next decade.

# By Chelsea Vuong

Background and Motivation:

Adolescents engaged in risky sexual behaviors, such as having multiple sex partners or partaking in sexual intercourse without birth control, have an increased risk of negative reproductive health outcomes. These outcomes can range from acquiring a sexually transmitted disease (STD) to an unintended pregnancy. Moreover, the physical, mental, and intellectual abilities compromised through high-risk pregnancies deeply impact a teen’s adulthood productivity. Therefore, by mitigating the factors that contribute to risky teen sexual behaviors, adult outcomes can be greatly improved. This paper studies reproductive health education and contraceptive usage among teens, as these are two effective factors that help mitigate risky sexual behaviors.

Data and Methodology:

State level data within the United States is used for this analysis.  Reproductive health education and publicly funded clinics are the two independent variables of interest, and no condom or any birth control use among teens are the two dependent variables of interest. Data on condom or birth control use, the primary outcomes of interest, comes from the Center for Disease Control and Prevention’s (CDC) High School Youth Risk Behavior Survey (YRBS). The state level data on publicly funded clinics, one of two independent variables of interest, comes the Health Resources & Services Administration (HRSA) for the total number of federally qualified health centers (FQHCs). The state level data on reproductive health curriculum, the second independent variable of interest, comes from the Sexuality Information and Education Council of the United States (SIECUS).

The first part of the study measures the correlation of FQHCs and reproductive health education on risky teen sexual behaviors using an Ordinary Least Squares (OLS) regression model. Then, an instrumental variable (IV) design is used to address the endogeneity issue with the OLS regression model. The two risky teen sexual behaviors analyzed in this paper are teens not using a condom or any method of birth control during their last sexual intercourse.

Key Findings and Implications:

The IV regression results for FQHCs indicate that there is no statistically significant effect of the number of health centers per 1000 teens on risky teen sexual behaviors. These results may hold strength in the overall context of publicly funded clinics, as publicly funded clinics comprise of FQHCs, Title X-funded clinics, and Planned Parenthood clinics. Since this study only analyzed FQHCs, it merely captured the effects of a portion of publicly funded clinics. To better estimate the impact of funding cuts on publicly funded clinics, data on Title X-funded and Planned Parenthood clinics should also be analyzed.

The IV results for reproductive health education show that mandated sex education, marriage education, and abstinence education were significant at reducing no condom use. For the simulated instrument regression, contraceptive education and education that prohibits abortion instruction were significant towards improving no condom and birth control use. While the findings are mixed, it is likely that continuing any type or mixture of education has potential in improving risky teen sexual behaviors.

Overall, my thesis provides a preliminary look at the multifarious dimensions of risky teen sexual behavior factors. Additional research that builds upon the findings of this study is necessary to develop informed school and state policies that better address teen reproductive health.

# By Hannah Jo Ellery

The success of a firm depends upon its workers, and thus upon how workers chose which firm to work for. One of the factors which may play a role in this decision is the reputation of the firm under consideration. Indeed, if reputation does play a role, it may have implications for the structures of labor markets and for markets more broadly, as firms which have had more time to build a reputation and rapport—incumbent firms—may have an entrenched advantage in hiring. In my thesis, I explore this possibility: do workers have a preference for firm reputation? What implications does a preference for reputation have for new firms in the labor market? Could prestige become a barrier to entry?

I first build a theoretical foundation for my hypothesis. In order to do so, I study a framework of worker-firm matching in a two-sided matching model, based in part on models of labor market matching introduced in Kelso and Crawford (1982). In this model, workers choose a single job and salary described by a contract, while firms choose multiple contracts, the union of which describes their total workforce.

Into this framework, I introduce reputation, taking as a base work done in Pycia and Yenmez (2017). Namely, I allow workers to care about the success of a company and what other workers think of that company. Notably, this is different from basic matching models, which avoid structures in which workers’ choices depend on the choices of others. When workers care about the success of a company, which depends upon the workers that are employed at that company, equilibrium matchings may not exist; instead, workers may wish to cycle between firms ad infinitum. Indeed, I show that in the context of a labor matching model with agents and utility functions, we cannot introduce a true preference for reputation. If workers’ choices depend on the current status of the labor market, as they would if the worker wants to be employed at a successful firm, equilibria may not exist. Thus, if workers care about the status of their firm, we may not be able to find a matching of workers to firms for the economy.

In spite of this, where equilibria do exist, I show that not only will markets tend to be more concentrated when workers have a preference for prestige, but it will also be weakly more difficult for a new start-up firm to enter a market where incumbents have already developed a reputation.

Second, I begin to show some evidence for such a preference for reputation in the labor market. In particular, I create a data set which collects the alma mater law schools of the employees of five large law firms, as well as the graduation year, hiring year, and former clerkships for each worker. Fortunately, large law firms are ranked explicitly on reputation and prestige in Vault’s annual Law 100 listing. This enabled me to compare the firm’s annual hire quality with their reputation, showing a strong and significant positive relation between reputation and hire quality both across and within firms.

Overall, this combination of theoretical and empirical evidence indicates that firm reputation may indeed play a role in the functioning of labor markets. Future research can help us to determine what industries feel the effects of this barrier the most, and how reputation impacts entry in practice.

## Econ Thesis 2021

We’re excited to have an incredible cohort of Class of 2021 Ec Thesis Writers this year!  They deserve a special shout-out for taking on the task of thesis research entirely remotely.  We look forward to sharing their work with you on this site and to hosting our annual Ec Thesis Poster Session in April.  More details coming soon!

## Ec Thesis 2020 Virtual Poster Session

Enjoy the posts below and chat with thesis writers via Zoom on Monday, May 25, 1-2pm ET!  Links available in each post and also here.

# By Paula Chappel

### Motivation

In my thesis, I investigate the impact of women’s education on their contraceptive knowledge and intent to use contraceptives. I exploit a Kenyan education policy change in January 2003 that made education free and compulsory. This is an important topic because better educated women are often healthier, and are more likely to be in the formal labor market and earn higher incomes. They are also likely to marry at a later age and, if they choose to become mothers, have fewer, healthier, and more educated children. These outcomes suggest that women’s school education, especially because due to potential effects on fertility, has the power to lift households and communities out of poverty. (World Bank, 2002)
Thus if indeed this primary school education policy is effective, and it increases the number of women attaining primary education then measuring the effect of this increase on women’s contraceptive knowledge helps to explain if some of the effects mentioned above are possible. Much of this work has been done in developed countries and less in developing countries.

### Data

For my main analysis, I use Kenyan Demographic Health Surveys (KDHS) data from 2,931 respondents for the birth cohorts of 1994 and 1996. When the 2003 FPE policy was enacted, the 1996 birth cohort was seven years old and the 1994 birth cohort was nine years old. Since the 1996 birth cohort was exposed to the policy at a younger age, I interpret the 1996 birth cohort to have had more exposure to the policy than the 1994 birth cohort. For additional analysis, I use the Performance and Monitoring Accountability 2020 (PMA2020) data and KDHS data for more birth cohorts.

### Methodology & Key Results

I assign treatment based on exogenous variation between birth cohorts. My findings show that the reform increased women’s educational attainment and literacy rates and led to an increase in knowledge of contraceptives. After including fixed effects for various demographics, I find that women born in 1996 are 4.72 percentage points more likely to obtain some education compared with those born in 1994. They are also 1.22 percentage points less likely to report they have no knowledge of contraceptives. When I assign varying intensities of exposure to the policy according to birth cohorts, I find women exposed to the policy are 8.79 percentage points more likely to get some education than those who are not exposed to the policy. For both analyzes, poor women are more likely to have their education and contraceptive knowledge increase as a result of the policy compared with women from other income brackets. The results of this paper are consistent with the notion that better-educated women have a greater understanding of contraceptives.

Please come chat with me at the Econ Senior Thesis Zoom sessionsMonday, May 25, 1-2pm ET—to learn more or ask any questions about my research or the senior thesis process: https://harvard.zoom.us/j/96528882732.

# By Aakanksha Vora

Banking serves as a gateway into the formal financial system, which can decrease inequality of outcomes and opportunity for the 1.7 billion unbanked adults globally. Building a financially included economy, by banking the unbanked, can alleviate poverty and reduce wealth disparities. However, there are several barriers to access, and ownership of a bank account does not always result in long-term, active use.

To study the factors that influence financial inclusion of the unbanked, I estimate the effects of policy exposure to India’s 2016 demonetization on access and usage of bank accounts. In November 2016, the Indian government invalidated 86% of currency overnight to combat endemic corruption, which resulted in asymmetrical cash contractions across districts. Using a differences-in-differences strategy, I exploit spatial variation in the severity of the shock to estimate the effects on household-level bank account and credit card ownership, as well as district-level deposits. For households in treated districts, I find an immediate and persistent 5 p.p. increase in bank account ownership, a 2% increase in savings deposits, and a lagged but persistent increase in credit card adoption.

Despite the benefits of using a bank account, a majority of the banked population has inactive accounts, which is very costly for the banking system. The second half of my thesis studies the factors that influence active bank account ownership to find tractable ways to help individuals accrue the benefits of financial inclusion. Although most studies in the field use randomized controlled trials, I chose to take a different approach by using publicly available data and designing a novel empirical test. To determine the effects of increased financial access on household savings in response to stochastic income shocks, I test the direct effects of income on consumption and the exogenous effects of income, through rainfall shocks, on consumption. I find that households in treated districts are able to offset the changes in consumption due to increased access to savings instruments.

My findings are evidence that small incentives to deposit cash into an account resulted in significantly higher rates of bank account ownership. Moreover, nudging households to operationalize their accounts by making deposits and withdrawals resulted in persistently higher savings and financial resilience. These results provide new insights for policymakers in India and other developing countries with low financial inclusion. My findings lay out solutions that will allow millions of people to escape poverty by increasing their ability to receive direct government transfers and manage unexpected changes in their income; and by giving them an opportunity to save and invest in their health, education, and businesses.

# By Wesley Cash

The recent proliferation of Chinese foreign aid in Africa has raised many concerns about China’s potentially exploitative economic and political intentions. This paper empirically tests the effectiveness of Chinese foreign aid by measuring its effect on child labor participation in developing countries. Using a difference-in-differences-like identification strategy at the subnational level, I compare child labor participation in regions that receive Chinese aid prior to being surveyed with child labor participation in regions that receive aid within the sample period after being surveyed. Using this method, I find a significant negative causal effect of Chinese aid on child labor participation. A conservative interpretation indicates that the presence of at least one aid project reduces regional child labor participation rates by 26.3%. This paper additionally investigates the potential role of education and wealth as mechanisms and finds mixed results. I then replicate the main analysis for aid from the World Bank and find insignificant results. This paper suggests that Chinese aid is not necessarily exploitative and that non-traditional aid practices may have as positive, if not more positive, humanitarian consequences on the developing world as traditional Western aid practices do.

Please come chat with me at the Econ Senior Thesis Zoom sessionsMonday, May 25, 1-2pm ET—to learn more or ask any questions about my research or the senior thesis process: https://harvard.zoom.us/j/94597824270.

Additionally, feel free to direct any questions to cash.wesley@gmail.com.

# By Juan Crestanello

My thesis revisits the relationship between a country’s domestic saving and its investment, a link first explored by Feldstein and Horioka (1980). I show that the once strong relationship between domestic saving and investment waned over time and essentially disappeared in the years before the financial crisis, followed by some reemergence more recently. I find that the countries that opened themselves most to trade saw the greatest decline in the saving-investment relationship. I also present evidence from an instrumental variable strategy that suggests that saving-investment relationship is causal. All told, the evidence in this paper suggests a general decline in the saving-investment relationship indicative of the strong forces of globalization at play and the recent uptick relates to a pull-back of cross-border capital investment following the Great Recession, which is even more severe after the COVID-19 crisis.

Please come chat with me at the Econ Senior Thesis Zoom sessionsMonday, May 25, 1-2pm ET—to learn more or ask any questions about my research or the senior thesis process: https://harvard.zoom.us/j/95544475337.

# By Elizabeth Herington

#### -Malala Yousafzai, global education activist and survivor of Taliban violence, 2013

Civil violence has become the most prevalent form of political violence since World War II; more than half of all countries have experienced a form of civil conflict since 1960 (Blattman and Miguel 2010). In the post-Cold War era, and especially since 9/11, developed countries have seen security problems in developing countries as threats to their own national security. This has led policymakers in developed countries to intervene in conflict-affected developing countries; in 2016, 65% of all development aid went to countries identified as fragile by the OECD (OECD 2018 p. 11). Much of the Economic literature on foreign aid in conflict zones focuses on the effect of humanitarian aid on conflict in these regions, and many scholars argue that aid can prolong or provoke conflict. However, a significant portion of the aid provided in these areas is not humanitarian aid, but instead falls under a category which I call security assistance. Security assistance is aid provided to governments to help them provide essential services like military and police, effective justice systems, and critical infrastructure to their citizens. For my thesis, I wanted to explore the relative efficacy of these different kinds of interventions. Do they accomplish their stated goals of strengthening states, ending conflicts, and reducing suffering? Which interventions are most effective?

I explore the role of foreign aid in conflict beginnings and endings by developing a theory of the impact of aid on state legitimacy and of state legitimacy on civil conflict. State legitimacy is the citizens’ belief that their government’s laws should be obeyed. I argue that security assistance will increase state legitimacy and that humanitarian aid has an ambiguous effect on state legitimacy. I then argue that aid which increases state legitimacy will both prevent conflicts from beginning and end existing conflicts faster.

I first test the hypothesis that security assistance will increase state legitimacy using survey data on citizen perceptions of their government from Afrobarometer and subnational data on various sectors of foreign aid from AidData. Using an ordered logistic regression with province fixed effects, I find that increases in security assistance are associated with increased state legitimacy.

### Table 1: Effect of Aid on State Legitimacy

 VARIABLES Courts Make Binding Decisions People Must Obey the Law People Must Pay Taxes Security Assistance (Mlns \$) 1.00046* 1.00059*** 1.00082 (0.00027) (0.00021) (0.00061) Humanitarian Aid (Mlns \$) 0.99963 0.99426*** 0.99627 (0.00209) (0.00184) (0.00402) Other Aid (Mlns \$) 0.99174* 1.00306 1.00997** (0.00421) (0.00992) (0.00472)

Subnational unit dummies and standard controls are used in all specifications. Coefficients reported are odds ratios. Odds ratios show whether a one unit increase in the independent variable is associated with an individual being more or less likely to answer in a category of stronger agreement with the statement. Standard errors clustered at the province level are in parentheses.

*** p<0.01, ** p<0.05, * p<0.1

Next, I test the theory that aid which increases state capacity will prevent and end conflict using the same subnational AidData dataset and geocoded data on conflict from the Uppsala Conflict Data Program.  Using a hazard model with province fixed effects to estimate the probability of conflict ending, I find that both security assistance and humanitarian aid are associated with decreased length of conflict, and that government capacity building aid, a type of security assistance, is the most effective. Government capacity building aid includes projects like audits to reduce corruption, training court employees to better provide effective justice and democracy promotion efforts. I also find that security assistance policy instruments are more effective at ending conflict when combined for a Whole of Government approach than when employed alone, and that military aid alone may increase the length of conflict.

### Table 2: Effect of Aid on Length of Conflict

 VARIABLES Conflict End Conflict End Conflict End Conflict End Mil Aid x Whole of Gvt 0.00362* (0.00200) Cap Aid x Whole of Gvt 0.02434* (0.01338) Whole of Gvt 0.01506 -0.03956 (0.07461) (0.08339) Military Aid (Mlns \$) 0.00043 -0.00194 0.00028 (0.00155) (0.00174) (0.00169) Capacity Aid (Mlns \$) 0.00772* 0.00721* 0.00488 (0.00414) (0.00415) (0.00473) Infrastructure Aid (Mlns \$) 0.00119 0.00117 0.00119 (0.00080) (0.00079) (0.00080) Security Assistance (Mlns \$) 0.00122* (0.00071) Humanitarian Aid (Mlns \$) 0.00640*** 0.00636*** 0.00626*** 0.00600*** (0.00196) (0.00198) (0.00197) (0.00192) Other Aid (Mlns \$) 0.00104 0.00080 0.00162 0.00123 (0.00334) (0.00348) (0.00366) (0.00400)

Subnational unit dummies and standard controls are used in all specifications. Columns 1 and 2 present results from my preferred specification with different kinds of aid. Columns 3 and 4 present results from a regression estimating the effects of a Whole of Government approach. Conflict End is a dummy set =1 if an ongoing conflict ended in a given year, and Whole of Gvt is a dummy set =1 if all three kinds of security assistance were used. Coefficients from logistic regression are reported as marginal effects. Standard errors clustered at the province-level and determined by the delta method, are in parentheses.

*** p<0.01, ** p<0.05, * p<0.1

Using the same hazard model with province fixed effects, I further find that government capacity building aid is the only intervention associated with a decrease in the likelihood of civil conflict.

 VARIABLES Conflict Start Conflict Start Conflict Start Conflict Start Cap Aid x Whole of Gvt 0.00597 (0.01151) Infr Aid x Whole of Gvt -0.00077 (0.00068) Whole of Government -0.06367 -0.03760 (0.04947) (0.04183) Military Aid (Mlns \$) -0.00294 -0.00148 -0.00160 (0.00797) (0.00711) (0.00731) Capacity Aid (Mlns \$) -0.01113* -0.01089 -0.00919 (0.00644) (0.00744) (0.00560) Infrastructure Aid (Mlns \$) -0.00025 -0.00022 -0.00018 (0.00019) (0.00017) (0.00014) Humanitarian Aid (Mlns \$) -0.00093 -0.00068 -0.00056 -0.00056 (0.00085) (0.00083) (0.00084) (0.00085) Security Assistance (Mlns \$) -0.00028 (0.00027) Other Aid (Mlns \$) -0.01035* -0.00884* -0.00864* -0.00869* (0.00559) (0.00504) (0.00503) (0.00508)

Subnational unit dummies and standard controls are used in all specifications. Columns 1 and 2 present results from my preferred specification with different kinds of aid. Columns 3 and 4 present results from a regression estimating the effects of a Whole of Government approach.  Conflict Start is a dummy set =1 if a conflict began in a given year, and Whole of Gvt is a dummy set =1 if all three kinds of security assistance were used. Coefficients from logistic regression are reported as marginal effects. Standard errors clustered at the province-level and determined by the delta method are in parentheses.

*** p<0.01, ** p<0.05, * p<0.1

I conclude that the kinds of aid which are effective in fragile states are different from, and less varied than, the kinds of aid which are effective in conflict states, and that policymakers should be careful to take initial conditions into account before designing interventions to prevent and end conflict in unstable states.

The OECD identifies state weakness as the single biggest spoiler for the achievement of the UN Sustainable Development Goals (OECD 2018 p. 6). In order to increase security and prosperity in the developing world, it is important for policymakers to understand the relative efficacy of different intervention strategies. Policymakers who are planning to intervene in an ongoing conflict should design comprehensive Whole of Government interventions, while policymakers planning to intervene in fragile states should focus their efforts on programs which will increase the local government’s bureaucratic, regulatory, and democratic capabilities and reduce corruption.

Works Cited:

Blattman, C., & Miguel, E. (2010). Civil War. Journal of Economic Literature, 48(1), 3–57. https://doi.org/10.1257/jel.48.1.3

OECD. (2018). States of Fragility 2018. OECD. https://doi.org/10.1787/9789264302075-en

Data Sets Used:

https://www.aiddata.org/datasets (AIMS datasets)