The MCC Incentive Effect: Quantifying Incentives for Policy Change in an Ex-Post Rewards System

Transcription

1 The MCC Incentive Effect: Quantifying Incentives for Policy Change in an Ex-Post Rewards System Prepared for the Millennium Challenge Corporation By Ingrid Aune Yanyan Chen Christina Miller Joshua Williams October 2013 Workshop in International Public Affairs Spring 2013

2 2013 Board of Regents of the University of Wisconsin System All rights reserved. For additional copies: Publications Office La Follette School of Public Affairs 1225 Observatory Drive, Madison, WI The Robert M. La Follette School of Public Affairs is a teaching and research department of the University of Wisconsin Madison. The school takes no stand on policy issues; opinions expressed in these pages reflect the views of the authors. The University of Wisconsin Madison is an equal opportunity and affirmative-action educator and employer. We promote excellence through diversity in all programs.

3 Table of Contents List of Figures... vii List of Tables... viii Foreword... ix Acknowledgments... x Executive Summary... xi Introduction... 1 I. The MCC and the Incentive Effect... 2 A. Selection Indicators and Incentive Effects... 2 B. Evidence of the MCC Incentive Effect: Status Quo Qualitative Evidence Quantitative Evidence... 4 II. Data... 4 A. Indicator Score: Description of Data Economic Freedom Investing in People Ruling Justly... 6 B. Indicator Rates of Reform... 6 C. Time Correlation... 7 D. Indicator Mobility... 7 E. Correlation between Indicators... 9 III. Which Countries Have Incentives? A. LICs versus LMICs B. Treatment Groups Qualitative MCC Incentive Effect Bordering the Eligibility Threshold Passing Control of Corruption Threshold Partnership Government Expenditures a. Countries in the Middle Range of Government Expenditures b. Countries in the Highest Quartile of Government Expenditures c. Countries in the Lowest Quartile of Government Expenditures Bordering the Eligibility Threshold and Passing Control of Corruption Bordering the Eligibility Threshold or Passing Control of Corruption Passing or Falling Short of the Eligibility Threshold C. Additional Control Variables IV. Bearing Down and Moving Forward V. Preliminary Analyses A. Nonparametric Approach... 16

4 B. Basic Rates of Reform Mean Comparison VI. Models A. Rate of Reform: Difference-in-Differences Model B. Likelihood of Improvement: Difference-in-Differences Model C. Proportion of Indicators Improved: Difference-in-Differences Model D. Assumption and Caveat VII. Difference-in-Differences Analysis for All Sample Countries A. Overview of Results B. Economic Freedom Qualitative MCC Incentive Effect Bordering the Eligibility Threshold Passing Control of Corruption Threshold Partnership Countries in the Middle Range of Government Expenditures C. Investing in People Qualitative MCC Incentive Effect Bordering the Eligibility Threshold Passing Control of Corruption Threshold Partnership Countries in the Middle Range of Government Expenditures D. Ruling Justly Qualitative MCC Incentive Effect Bordering the Eligibility Threshold Passing Control of Corruption Threshold Partnership Countries in the Middle Range of Government Expenditure E. Results from Proportion of Indicators Improved Model VIII. DD Analysis for LICs and LMICs A. LICs Breakdown B. LMICs Breakdown C. Time Effects IX. Conclusions and Recommendations A. Lack of Quantitative Evidence B. How to Improve Analyses: C. How to Improve MCC Incentives Create Aspirations Provide Knowledge and Information Separate Country Group LICs and LMICs D. Final Thoughts... 47

5 Appendix A: Descriptions of Indicators A. Economic Freedom Indicators Credit Depth Legal Rights Cost to Start Business Time to Start Business Fiscal Policy Inflation Time to Register Property Cost to Register Property Regulatory Quality Trade Freedom Gender in the Economy Access to Land B. Investing in People Indicators Girls Primary Education Health Expenditures Immunization Rate Education Expenditures Girls Secondary Education Enrollment Child Health Natural Resource Protection C. Ruling Justly Indicators Civil Liberties Control of Corruption Government Effectiveness Political Rights Rule of Law Voice and Accountability Freedom of Information Appendix B: Indicator Information and Rates of Reform Statistics Appendix C: Autocorrelation within Indicators Appendix D: Correlation between Indicators Appendix E: Average Number of MCC Indicators Passed Appendix F: Treatment Countries Appendix G: Full Regression Results Endnotes Works Cited... 74

6

7 List of Figures Figure 1: Indicator Mobility: Indicator Rank Change... 8 Figure 2. Qualitative MCC Incentive Effect: Bordering the Eligibility Threshold Treatment Rates of Reform for Control of Corruption Figure 3. Mixed or Nonexistent Effect: Qualitative MCC Incentive Effect Treatment Rates of Reform for Control of Corruption Figure 4. Negative Effect: Countries in the Middle Range of Government Expenditures Treatment Rates of Reform for Control of Corruption Figure 5. Qualitative MCC Incentive Effect: Bordering the Eligibility Threshold Treatment Mean Rates of Reform for Control of Corruption Figure 6. Mixed or Nonexistent Effect: Qualitative MCC Incentive Effect Treatment Mean Rates of Reform of Control of Corruption Figure 7. Negative Effect: Countries in the Middle Range of Government Expenditures Treatment Mean Rates of Reform of Control of Corruption Figure 8: Number of Positive and Negative Results in LICs Figure 9: Number of Positive and Negative Results in LMICs Figure 10: Number of Positive and Negative Results in Different Time Periods Figure C1. Fiscal Policy Figure C2. Inflation Figure C3. Regulatory Quality Figure C4. Trade Freedom Figure C5. Girls Primary Education Figure C6. Health Expenditures Figure C7. Immunization Rate Figure C8. Education Expenditures Figure C9. Control of Corruption Figure C10. Government Effectiveness Figure C11. Rule of Law Figure C12. Voice and Accountability Figure D1: Correlation between Economic Freedom Indicators Figure D2: Correlation between Investing in People Indicators Figure D3: Correlation between Ruling Justly Indicators Figure E1: Average Number of MCC Indicators Passed by LICs Figure E2: Average Number of MCC Indicators Passed by LMICs vii

8 List of Tables Table 1. Indicator Mobility: Average of Indicator Rank Change... 9 Table 2. Correlation of Qualitative MCC Inventive Effect to Other Treatments Table 3. Number of Positive and Negative Results of Three Models for Three Categories of Indicators Table 4. Statistically Significant Results for Economic Freedom Table 5. Statistically Significant Results for Investing in People Table 6. Statistically Significant Results for Ruling Justly Table 7: Statistically Significant Results for Proportion of Indicators Improved Model Table B1: Indicator and Rates of Reform Information Table F1: LICs included in Treatments Table F2: LMICs included in Treatments Table G1: Economic Freedom Indicators Rate of Reform Model Table G2: Economic Freedom Indicators Likelihood of Improvement Model Table G3: Investing in People Indicators Table G4: Ruling Justly Indicators Rate of Reform Model Table G5: Ruling Justly Indicators Likelihood of Improvement Model Table G6: Proportion of Indicators Improved Model viii

9 Foreword The La Follette School of Public Affairs at the University of Wisconsin Madison offers a two-year graduate program leading to a Master of Public Affairs or a Master of International Public Affairs degree. In both programs, students develop analytic tools with which to assess policy responses to issues, evaluate implications of policies for efficiency and equity, and interpret and present data relevant to policy considerations. Students in the Master of International Public Affairs program produced this report for the Millennium Challenge Corporation. The students are enrolled in the Workshop in International Public Affairs, the capstone course in their graduate program. The workshop challenges the students to improve their analytical skills by applying them to an issue with a substantial international component and to contribute useful knowledge and recommendations to their client. It provides them with practical experience applying the tools of analysis acquired during three semesters of prior coursework to actual problems clients face in the public, non-governmental, and private sectors. Students work in teams to produce carefully crafted policy reports that meet high professional standards. The reports are research-based, analytical, evaluative, and (where relevant) prescriptive responses for real-world clients. This culminating experience is the ideal equivalent of the thesis for the La Follette School degrees in public affairs. While the acquisition of a set of analytical skills is important, it is no substitute for learning by doing. The opinions and judgments presented in the report do not represent the views, official or unofficial, of the La Follette School or of the client for which the report was prepared. Melanie Frances Manion Professor of Public Affairs and Political Science May 2013 ix

10 Acknowledgments The journey toward the completion of this report has been arduous but rewarding. We would like to thank the following people, without whom we could not have accomplished this task: First and foremost, Professor Melanie Manion, who was understanding and encouraging, and who went beyond the call of duty to ensure we could present the best as possible work; Andria Hayes-Birchler and the people at the Millennium Challenge Corporation, who provided us with the project, gave us useful insight and helpful feedback; Karen Faster, who assisted us with a watchful editorial eye; Shirley Smith, who added an artist s perspective to our tables and graphs; our families for their understanding and support; finally, all of the staff and students at the La Follette School of Public Affairs, who consistently encouraged us as we made the common study space our home. x

11 Executive Summary The Millennium Challenge Corporation (MCC) is an independent U.S. foreign aid agency that uses an ex-post rewards system to determine potential eligibility for funding. When countries seek to change their policies specifically to meet MCC eligibility requirements, the MCC incentive effect is taking place. This report analyzes that effect. We independently compiled data from the original third-party sources the MCC uses for its eligibility indicators. We then defined a series of plausible treatment groups countries we believe are most likely to show signs of the MCC incentive effect. We conducted preliminary analyses to test our hypotheses and models. We then used three difference-in-differences regression models and analyzed the results by indicators, treatment groups, country income groups, and time periods. We conclude no compelling evidence suggests an overall MCC incentive effect. However, we do find subtle evidence of the MCC incentive effect that varies by how we divide our data. When we separate the indicators by the three MCC indictor categories of Economic Freedom, Investing in People, and Ruling Justly, we see differences in our results depending on which of the five of our most plausible treatments we use. For instance, countries that border the MCC eligibility threshold appear to have the incentive to improve on indicators oriented toward good governance. Similarly, when we separate countries by per-capita gross domestic product, we see differences in results that vary by indicator category and time period. For instance, low-income countries appear to have incentives to improve indicators oriented toward good governance. Yet, this evidence may not be strong enough to truly convince skeptics that the MCC incentive effect exists. We further find that two treatment groups that should show the strongest signs of the MCC incentive effect because of their structural biases at best show weak evidence of its existence. We therefore finish the report with recommendations on what the MCC could do to improve research on the topic, as well as what the MCC might do to stimulate a stronger MCC incentive effect. These steps most notably include: creating aspirations in countries to pursue MCC funding; providing more information about MCC criteria; and creating more specified differences in eligibility requirements based on country groups. xi

12 xii

13 Introduction The United States government established the Millennium Challenge Corporation (MCC) in 2004 to reduce poverty and promote sustainable economic growth in other countries. The agency has contributed directly to this development work by having allocated more than $7 billion in poverty-reduction compacts with poor, but well-governed countries as of spring The MCC selects countries for assistance based on past policy performance, an ex-post rewards system. The MCC s foreign aid system may also contribute to fighting poverty indirectly: the possibility of obtaining substantial funding from the MCC may create incentives for governments to change policies based upon MCC criteria. This response is the MCC incentive effect. However, little research demonstrates the quantifiable existence of the MCC incentive effect. Documentation of a quantifiable incentive effect would support the claim that the relatively new ex-post rewards system the MCC applies and promotes is beneficial in the fight against poverty. As the MCC is one of a handful of worldwide agencies using the ex-post foreign aid model, evidence of the MCC incentive effect could have two important implications. First, it could provide a model for change in the international aid community. Second, it could improve the MCC s standing in Congress to help it secure funding to keep fighting poverty around the world. This report seeks to answer two questions: 1. Does rigorous quantitative analysis support MCC anecdotal evidence of the MCC effect? 2. What analytical and policy prescriptions do quantitative findings suggest for the agency? Section I of this report introduces the MCC and research on the MCC incentive effect. Section II describes the data we utilize in the analysis. Section III outlines our conjectures about which countries might show MCC incentive effects. Section IV details how we focus the analysis. Section V presents preliminary analyses of our data to justify our choice of treatment groups and regression models. Section VI details the difference-in-differences regression models we utilize. Section VII provides general analysis of the results of the regression models for all countries. Section VIII looks at differences between low-income countries and low-middleincome countries, as well as differences over time. Section IX provides our conclusions and recommendations to the MCC for action. 1

14 I. The MCC and the Incentive Effect The MCC is an independent U.S. foreign aid agency aimed at fighting global poverty through focus on good policies, country ownership, and results (MCC 2013a). The MCC s board of directors selects countries as eligible for grant assistance based on how they perform on a series of 20 (as of fiscal year 2013) quantitative policy indicators derived from independent third-party sources. The indicators are grouped into three policy categories: Economic Freedom, Investing in People, and Ruling Justly (MCC 2012, 3). Rather than base eligibility for aid on need or strategic interest, the MCC bases awards on past performance an ex-post rewards system. Two main reasons are given for this approach. First, taxpayer money will be better spent because MCC funds will only go to the best-governed poor countries. Second, when aid is based on performance, poor countries should have an incentive to pursue good governance. This latter reason is known as the MCC effect (Johnson and Zajonc 2006, 2-4) or, more precisely, the MCC incentive effect. The MCC Effect is the positive impact that MCC is having on developing countries beyond its direct investments. To date, the most significant impact has been the incentive created for countries to adopt legal, policy, regulatory, and institutional reforms related to the MCC eligibility criteria. (MCC 2008, 1) A. Selection Indicators and Incentive Effects The amount of foreign aid assistance the MCC allocated as of spring 2013 is more than $7 billion in poverty-reduction compacts. Almost $500 million went for policy improvement through its Threshold Programs (MCC 2013a). Dreher, Nunnenkamp, and Öhler (2012) find that countries receiving funding from the MCC are more likely to receive funding from other countries and donor organizations, which may contribute to strong economic benefits and, hence, incentives to improve outcomes measured by the MCC. However, Grabbe (2005) and Öhler, Nunnenkamp, and Dreher (2010) explain that if MCC funding declines or does not reach the level expected by potential partnership countries, the financial limits may constrain potential partnership countries willingness to comply with MCC criteria. The funds allocated to the MCC as of spring 2013 were less than the amount first outlined by former President George W. Bush, and Congress was expected to continue cutting the MCC s funding through sequestration requirements under the Budget Control Act of 2011 (Tarnoff 2013). The MCC only considers distributing aid to a country once that country has passed a threshold on its indicator criteria list. The MCC empirically determines each indicator based on data from third-party institutions. The number of indicators has grown and changed since the MCC s creation (MCC 2012). The MCC s Board of Directors evaluates countries for eligibility by looking at how they perform compared to countries with similar income levels in one of two groups: low-income countries (LICs) or lower-middle-income countries (LMICs). 2

15 For fiscal year 2013, candidate countries must perform better than the median of its peers. To be more specific, the candidate countries need to perform above the median or absolute threshold on at least half of the indicators within their income peer group, as well as above the median on the Control of Corruption indicator[,] and above the absolute threshold on either the Civil Liberties or Political Rights indicators. For the Political Rights, Civil Liberties, Inflation, and Immunization Rate indicators, countries performance is gauged against an absolute threshold as opposed to the median score. The board also considers whether a country passes at least one indicator in each of the three indicator categories Ruling Justly, Investing in People, or Economic Freedom (MCC 2012, 4-5). The selection process for MCC funds is rather complex, which undoubtedly means countries have difficulty understanding how they can improve their MCC scores or which indicators they should improve on to meet the MCC eligibility criteria. Accordingly, the more complex the selection process, the less likely we expect the incentive effect to occur. B. Evidence of the MCC Incentive Effect: Status Quo Previous work finds qualitative and quantitative evidence of the MCC incentive effect. 1. Qualitative Evidence The MCC finds qualitative or anecdotal evidence that suggests the MCC incentive effect exists. In a 2008 report, the MCC explains that third parties, such as academics, journalists, NGOs, investors and donor agencies, have independently recognized the MCC effect (MCC 2008, 3). However, none of the information or authoritative citations presented is quantitative in nature. MCC employees working closely with candidate and potential candidate countries believe that the policy criteria have positive externalities. They report that candidate countries continuously come back to the MCC to see what they can change in order to improve their MCC score (Hayes-Birchler 2013). Governments regularly contact the MCC to ask what kind of reforms and investments can improve their MCC score. Some governments subsequently implement reforms, strengthen institutions and improve data quality (MCC 2013b, 4). Although few studies specifically document the MCC incentive effect, studies of incentive effects examine ex-post policy criteria. The literature on European Union membership criteria widely acknowledges that ex-post conditionality brings about substantial change in domestic policy (Grabbe 2005; Schimmelfennig, Engert, and Knobel 2006; Schimmelfennig and Sedelmeier 2005). 1 Freyburg and Richter (2010) stipulate that incentive-based instruments are only suitable for triggering policy change under certain domestic conditions. 2 Similarly, Öhler, Nunnenkamp, and Dreher (2010, 18-21) anticipate that the MCC 3

16 incentive effect will get weaker with time because aid rewards will not be perceived as sufficiently compensating the costs of reform efforts (i.e., the costs of compliance with criteria that are not primary national interests). 2. Quantitative Evidence Two empirical studies document the MCC incentive effect. The first systematic and quantitative attempt to identify the MCC incentive effect was published less than two years after the MCC was created. Johnson and Zajonc (2006) employ two statistical methods, a difference-in-differences model and a regression discontinuity model. 3 They (2006, 22) find substantial evidence that countries respond to MCC incentives by improving their indicators. However, Johnson and Zajonc s evidence is neither consistent nor overpowering. The findings include few data points and center around the two years prior to and after the announcement of the MCC in 2002, rather than its establishment in If the MCC incentive effect in fact exists, we expect that some time would be required for it to show up in the data. The second and more recent study, by Öhler, Nunnenkamp, and Dreher (2010), also documents the existence of the MCC incentive effect; however, the authors conclude that the effect gets weaker over time. The study uses a difference-indifference method similar to that of Johnson and Zajonc (2006), but it focuses on only one of the criteria for eligibility: the Control of Corruption indicator. Because of this limited scope of analysis, Öhler, Nunnenkamp, and Dreher do not provide a strong empirical analysis of the MCC incentive effect for all MCC indicators. Rather, the study provides another stepping stone for use of a difference-in-differences model in an expanded analysis. It also indicates a need to analyze effects over different time periods. II. Data For our analysis we started with a list of 26 indicators based on the MCC s current and past indicators; a majority of the indicators are the same as the indicators the MCC uses in fiscal year 2013 (MCC 2012, 4-5). However, we adapted some and eliminated others as described in appendix A to come up with 20 indicators for our analysis. Appendix B summarizes the data sources and availability. Drawing directly from the third-party sources the MCC uses, we compiled a single large panel database. Note that when we use the term indicator, we are from this point forth referring to the data indicators we have gathered, unless otherwise specified. Availability of data for each of the 20 indicators is not consistent across countries and years, but overall the panel is mostly balanced. The reason we use these data instead of MCC data is because the third-party sources revise their historical data as new information becomes available, while the MCC does not. Moreover, we have found inconsistences in MCC data. 4 Therefore, our data are more up-to-date and 4

17 consistent than data used in previous research on the MCC incentive effect, which should help our analysis yield more reliable results. We examine indicators in four ways: indicator score, rates of reform, time correlation, and indicator mobility. Our conclusions are based on our data. A. Indicator Score: Description of Data Our list of indicators differs from the indicators the MCC uses in fiscal year 2013 for three reasons: 1. MCC policy inconsistencies over time. We exclude the Freedom of Information indicator, which the MCC added in Instead, we use Voice and Accountability, the predecessor to Freedom of Information, to ensure comparable data over the time periods we analyze. We exclude Girls Secondary Education Enrollment and Gender in Economy indicators, which the MCC also added in fiscal year Lack of data. We drop Child Health and Natural Resource Protection indicators because our models require pre-mcc and post-mcc period data, which these indicators lack. 3. Gaps between the MCC and third-party indicators. The MCC combines thirdparty data to create its Access to Credit, Business Start-Up, and Land Rights and Access indicators (not listed in appendix A). Because we obtain our data from the MCC s third-party sources, we keep the original data as separate indicators in our analysis: Credit Depth, Legal Rights, Cost to Start Business, Time to Start Business, Cost to Register Property, and Time to Register Property, respectively. 5 From this point we only discuss our 20 indicators, which we divide using the MCC s three indicator categories: Economic Freedom, Investing in People, and Ruling Justly (MCC 2012, 4-5). 1. Economic Freedom The Economic Freedom indicators assess a government s commitment to strong economic policies and economic opportunities for its people through market forces (MCC 2012, 29). The indicators in this category we use are: Credit Depth, Legal Rights, Cost to Start Business, Time to Start Business, Fiscal Policy, Inflation, Time to Register Property, Cost to Register Property, Regulatory Quality, and Trade Freedom. An environment where people can access credit (MCC 2012, 33-35), start a business and register property with little time and cost (MCC 2012; World Bank Group 2013c), own property (World Bank Group 2013c), have a government whose policies reduce barriers to trade (Miller, Holmes, and Feulner 2013), and have sound monetary and fiscal policy will promote economic growth through lower consumer prices and increased consumption, production, savings, and investment (MCC 2012). 5

18 2. Investing in People The Investing in People indicators assess the government s commitment to improving the standard of living of its people (MCC 2012, 23). The indicators in this category we use are: Girls Primary Education, Health Expenditures, Immunization Rate, and Education Expenditures. Studies show that investments in health result in increased economic activity because healthier people are more productive (MCC 2012, 24-25). The protection and preservation of the environment is also important to sustain increased economic development without depletion of natural resources (MCC 2012, 29). Additionally, research shows that investment in education contributes to economic growth (MCC 2012, 26). 3. Ruling Justly The Ruling Justly indicators assess a country s level of democratic governance (MCC 2012, 13). The indicators in this category we use are: Civil Liberties, Control of Corruption, Government Effectiveness, Political Rights, Rule of Law, and Voice and Accountability. Countries with the following characteristics have higher rates of economic growth: citizen participation in the selection of government (MCC 2012, 13-14); freedom of expression (World Bank Group 2013g); a perception of low corruption and high ability to combat corruption (MCC 2012, 16-18); a high quality of public and civil services (MCC 2012, 18-20; World Bank Group 2013d); a legitimate legal system (MCC 2012, 20-21; World Bank Group 2013f); democratic governance (MCC 2012, 13-14); and expanded organizational, human, and economic rights (MCC 2012, 15-16). B. Indicator Rates of Reform For each country, we devise the rates of reform in each policy area by calculating two-year differences in indicator scores. A two-year difference helps to remove yearly fluctuations that might be present in single-year differences. Using differences at intervals of three or more years decreases the level of detail by which we can do the analysis, particularly in regard to changes over time. For 15 of our 20 indicators, a positive rate of reform means a country is improving its policies and getting closer to (or is passing) the threshold to become eligible for an MCC poverty-reduction compact. For example, a rate of reform of 0.5 for the Control of Corruption indicator for indicates a positive policy change. For the other five indicators, Cost to Start Business, Time to Start Business, Inflation, Cost to Register Property, and Time to Register Property, reform occurs when the original scores are negative. For instance, a lower number of days to start a business is regarded as better than a higher number. So, we multiply the rates of reform of those five indicators by negative one (-1) for consistent interpretation: the greater the value, the better that country performs on the indicator. Appendix B summarizes statistics of the 20 indicator rates of reform in this study. 6

19 C. Time Correlation Changing and implementing national policies takes time. In our data we look at 98 countries in four time periods ( , , , and ). The presence of autocorrelation within each country must be expected. That is, the policy a country has in one time period is likely to influence policy in the following periods. We examine autocorrelation by calculating the correlation of residuals from the regression of the dependent variable of interest to the time periods. We do this to help understand what exactly is affecting changes in the rates of reform of indicators over time. Appendix C gives an overview of the autocorrelation within 12 of the indicators. The autocorrelation does not seem to be very strong as a whole, meaning a country s policy does not have much influence on later policy. First, for most indicators, we observe that the correlation of the rates of reform of any two periods for a given indicator decreases the further apart in time the two periods are from each other. This decrease in correlation is consistent with real-world experience that impacts of policy reform generally tend to fade with time. Second, we observe a high correlation for a few indicators. For example, for Fiscal Policy (figure C1) we see that the periods and have a correlation as high as 0.85 (where 1 is perfect correlation). This high correlation may be due to a connection between the time needed for policy reform to be implemented and enforced. Yet the high correlation usually lasts no more than three time periods. D. Indicator Mobility To determine how quickly one policy indicator can change compared to others, we examine indicator mobility by calculating indicator rank change. Since each indicator score has its own scale, we standardize by ranking the indicator scores for the 98 countries every two years. For example, for Control of Corruption, we rank the countries according to their score when the MCC was established, We then re-rank all countries every two years thereafter according to how much their ranking has changed. A country that has a relatively larger increase (or decrease) in its Control of Corruption score is ranked higher (or lower) than a country that has not changed its Control of Corruption score much from one year to the next. In figure 1 we present the 10 countries that change the most for each of the 20 indicators. For example, the 10 countries with the highest change in ranking for Control of Corruption are represented, as are the 10 countries that change most in rank for Inflation, and for Education Expenditures, and so forth. These countries are not necessarily the same for each indicator. 7

20 Figure 1: Indicator Mobility: Indicator Rank Change 10 Countries that Change the Most for Each of the 20 Indicator Scores Note: The lighter the color, the faster the indicator score changes Source: Authors calculations The shades in the figure reflects an indicator s mobility, so the lighter the color the more mobile the indicator. The Economic Freedom indicators are the lightest in color, followed by the Investing in People and Ruling Justly indicators, which implies that the Economic Freedom indicators generally have more mobility. This pattern might also mean indicators in the Economic Freedom category are less stable and easier to change. Inflation shows the fastest rate of change, followed by Time to Start Business and Credit Depth. Indicators in the Investing in People and Ruling Justly categories are darker in color, which suggests that these indicators are more difficult to change. Another way of presenting indicator mobility is to average all the changes in ranking for each indicator. An indicator with a higher average score reflects 8

21 greater mobility across our data. These average scores are presented in table 1. The expectations of change in ranking for indicators in the Economic Freedom, Investing in People, and Ruling Justly categories are 13.55, 9.60, and 5.97 respectively. This result confirms the impression that the Economic Freedom indicators are more mobile in our data than the other two indicator categories. Economic Freedom Investing in People Ruling Justly Table 1. Indicator Mobility: Average of Indicator Rank Change Average Credit Depth Legal Rights Cost to Start Business Time to Start Business Fiscal Policy Inflation Time to Register Property Cost to Register Property Regulatory Quality Trade Freedom Average Girls Primary Education Health Expenditures Immunization Rate Education Expenditures Average Civil Liberties Control of Corruption Government Effectiveness Source: Authors calculations Political Rights Rule of Law Voice and Accountability E. Correlation between Indicators In appendix D, we plot correlations between indicator scores and between rates of reform for each of our 98 countries. Low correlations would suggest that indicators are independent from each other and are not closely related, and vice versa. The Economic Freedom indicators have low correlations in indicator scores and rates of reform. For the rates of reform, only one correlation of indicator scores is above 0.3; the correlation between Cost to Start Business and Regulatory Quality is For the Investing in People category, the correlations of the indicator scores are above 0.3 for three of the six indicators; however, only the correlation between the rates of reform for Girls Primary Education and Health Expenditures is above 0.3. The Ruling Justly indicators have the highest correlations in terms of both indicator scores and rates of reform. The indicator score correlations for 10 out of 15 indicator combinations are close to, or above 0.3, implying that these indicators have strong linear relationships and are somewhat interdependent. The correlations of the rates of reform for seven out of 15 indicator combinations are also close to, or above

22 These different correlation structures of the indicator categories have three implications. The first is that the movement of an indicator is not always isolated; some indicators move as a cluster. For Ruling Justly, high correlations indicate that the indicators move or change together, meaning improving one indicator results in an improvement in others. Economic Freedom is the opposite. For example, Time to Register Property has an extremely low correlation with Inflation; this relationship is reasonable because we would not expect the time it takes to register property to increase if the inflation rate increases. Second, moving one indicator appears easier than moving a bundle of indicators. This movement of individual indicators is another possible reason for the active ranking change in Economic Freedom indicators we observe. At the same time, this pattern of movement suggests the existence of a potential to manipulate the scores of the indicators in the Economic Freedom category with greater ease, at least in the short run. On the other hand, thorough political reform should create better results in the Ruling Justly category in the long run. The third implication is that the MCC s indicators in the Ruling Justly category may be redundant because of the high correlations across indicators. The Ruling Justly indicators are not likely to change; the high correlations reinforce their immutability. III. Which Countries Have Incentives? In this section we briefly discuss groups of countries. First, we take an initial look at whether an analytical division should be made based on country income levels. Then, we discuss conjectures on reasonable treatment groups to explore the MCC incentive effect. While we are calling these groups of countries treatment groups, we do not believe we have a natural experiment. Rather, we consider our analysis to be a quasi-experiment because we select the treatment groups specifically to test for the MCC incentive effect; the groups are not randomly assigned (Remler and Van Ryzin 2011). Finally, we look at potential control variables. A. LICs versus LMICs We initially planned to analyze LICs and LMICs as one group. 6 However, because the MCC distinguishes between LICs and LMICs, we investigate whether a distinction between the two gives supplementary information about the MCC incentive effect. Based on analysis of the number of individual indicators for which each country meets minimum required scores, we conclude that enough of a difference exists between LICs and LMICs to justify separate analyses of each group of country, in addition to analyzing all countries as one group. See appendix E for graphs showing how the 69 LICs and 29 LMICs performed. 10

23 B. Treatment Groups We investigate the MCC incentive effect in multiple ways. We find it reasonable to assume that all countries do not respond the same way to the same incentives. We posit that some countries have greater incentive than others to change policies in accordance with MCC criteria and selection indicators. Countries that we believe to be more likely to exhibit the MCC incentive effect constitute our treatment groups, and the remaining countries constitute the corresponding control groups. No clear theoretical or empirical guidance favors one treatment as more realistic than any other. We present the following 10 treatments as reasonable conjectures, then analyze five of them in detail. 1. Qualitative MCC Incentive Effect We test the countries that show qualitative signs of the MCC incentive effect to determine if those countries also show quantitative evidence of such an effect. We identify 31 countries in this treatment group (as noted in appendix F) based upon our review of MCC literature and discussions with MCC employees (MCC 2008; Hayes-Birchler 2013). 2. Bordering the Eligibility Threshold We postulate that countries that are bordering the MCC eligibility threshold have stronger incentives to change policies than do countries that are not close. We arbitrarily define this incentivized group as countries that are +/- 3 passing indicators within the threshold. For instance, in 2004, countries must pass minimum required scores on eight indicators to meet the eligibility threshold; therefore, our treatment group consists of countries that pass the required scores on five to 11 indicators. We conjecture that countries reaching required scores for five to seven indicators may have greater incentive to change their policies (compared to other LICs and LMICs) because they have fewer improvements to make to be considered eligible to enter into poverty-reduction compacts with the MCC. Countries that meet required scores on eight to 11 indicators may have greater incentives to change their policies (compared to other LICs and LMICs) because they need to maintain their eligibility. We created this treatment group by compiling a database with the number of MCC indicators meeting required scores (obtained from MCC scorecards that summarize how each country performed on the indicators for a given fiscal year) for each country that the MCC evaluated each year since it began operations in We then used computational software to determine which countries fell within +/- 3 of the eligibility threshold during each year. We also note in our database whether each country passed Control of Corruption each year, which is mandatory to become compact eligible, as we discuss below. 11

24 3. Passing Control of Corruption In addition to passing the eligibility threshold, countries need to fulfill some other compulsory requirements to be truly eligible for MCC aid. Meeting the minimum required score for the Control of Corruption was the only compulsory requirement until 2013, when meeting required scores on either Political Rights or Civil Liberties became compulsory as well (MCC 2012). The MCC requires performance above the median of all countries within a given income group for Control of Corruption for a country to be considered eligible for a compact. Several countries meet required scores on enough indicators, but are ineligible for funds because they fail to meet the MCC s standard for Control of Corruption. We conjecture that a country may lack an incentive to pass other standards if it has not (and potentially cannot) pass the Control of Corruption benchmark. Additionally, Öhler, Nunnenkamp, and Dreher (2010) focus on Control of Corruption in their study of the MCC incentive effect, so we include this treatment to allow for comparisons between our conclusions and theirs. 4. Threshold Partnership Another treatment group we define is countries that have received funding from the MCC as threshold partners. These countries receive MCC funding and assistance to help surmount the final hurdles to become eligible to enter into full poverty-reduction compact. Upon advice from MCC staff, we conjectured that if these partnership programs were efficient, one would expect the countries in a threshold partnership with the MCC to have greater incentives to change policies than other LICs and LMICs. These countries are subject to economic incentives driven by direct cooperation with the MCC (ex-ante rewards) in addition to the MCC incentive effect (ex-post rewards). Since the threshold partnership countries are receiving financial and direct support to improve policies and become eligible for MCC aid, the partnership may contribute to a positive bias in our quantitative evidence of the MCC incentive effect. However, quantitatively distinguishing between the ex-ante rewards incentives and the MCC incentive effect is difficult. To this extent, our analysis considers these countries similar to the Bordering the Eligibility Threshold treatment, and therefore worthy of scrutiny to see if they portray similar signs of the MCC incentive effect. 5. Government Expenditures We can think of government behavior as similar to a human agent whose expenditures have certain financial constraints. Thus we would expect that governments generally want to increase revenue. However, the desire to increase revenue through MCC aid is in turn dictated by a government s overall expenditure and ability to generate revenue through other means. Based on this line of thought, we divided countries into quartiles according to General Government Consumption Expenditure (U.S. Dollars) from World Bank Group (2013a) data and created three treatment groups, listed below. 12

25 5a. Countries in the Middle Range of Government Expenditures We expect the countries in this group, countries in the second and third quartiles, to have incentive to pursue MCC funding. 5b. Countries in the Highest Quartile of Government Expenditures We expect the countries in this group may have so much invested domestically that they are unwilling to seek or unlikely to care about MCC funding. 5c. Countries in the Lowest Quartile of Government Expenditures We expect the countries in this group do not have the ability or do not care to engage in activity that would likely lead to the policy reforms needed to become eligible for MCC aid. 6. Bordering the Eligibility Threshold and Passing Control of Corruption This treatment group includes countries which are bordering the eligibility threshold (treatment 2 group) and meet required scores for Control of Corruption (treatment 3 group). 7. Bordering the Eligibility Threshold or Passing Control of Corruption This treatment group includes countries which are bordering the eligibility threshold (treatment 2 group) or meet required scores for Control of Corruption (treatment 3 group). 8. Passing or Falling Short of the Eligibility Threshold Although the theoretical basis for inclusion as a treatment seems weaker to us, we test treatment groups defined by whether countries have or have not passed the eligibility threshold of meeting required scores on at least eight indicators. Treatment 8a consists of countries with eight or more indicators that meet or exceed required scores. Treatment 8b consists of countries with fewer than eight indicators that meet or exceed required scores. Countries that have not passed the eligibility threshold have different incentives than those that have. As these are essentially opposite groups, we would expect to find essentially opposite results. C. Additional Control Variables We created a short list of control variables to account for factors that may affect a country s performance. The country-level control variables we gathered from the World Bank Group (2013a) include: gross domestic product (GDP), per capita GDP, general government final consumption expenditures, population, and tax revenue. We additionally control for the year for each of these control variables. Through statistical variable selection methods (backward selection and Akaike information criterion), we ultimately decided to control for GDP and population as exogenous variables. We scaled down these two variables to make their effects discernible. Also, we averaged values across the three consecutive years covered 13

26 in each of our rates of reform periods (i.e., for the period we use the average of the values in 2004, 2005, and 2006). We believe this approach better reflects the period s trends than would a snapshot of GDP and population at a certain time point. We chose to control only for GDP and population in our analyses due to concerns of endogeneity: some MCC indicators are built by incorporating variables we considered for potential controls. IV. Bearing Down and Moving Forward Due to MCC s internal information and staff members abundance of experience in dealing with international development, we have sound reason to believe that the countries identified in the Qualitative MCC Incentive Effect treatment (treatment 1) are the most likely to display a quantifiable MCC incentive effect, if it exists. Thus the treatment is an ideal approximation of which countries MCC policy actually incentivizes. This approximation allows us to test our other treatment groups, which we believe might exhibit the MCC incentive effect. We test the validity of our treatment group conjectures by estimating the probability that a country under each treatment is actually being incentivized to change its policy, i.e., falling into treatment 1. The model is as below: 1 Being Incentivized: Being in Treatment 1 α β 1 Being in Treatment Group j c X (j 1) (1) Where 1( is a function of whether the expression in the parentheses is true. is a country time specific variable. The results are shown in table 2: 14

27 Table 2. Correlation of Qualitative MCC Inventive Effect to Other Treatments Dependent Variable: Qualitative MCC Incentive Effect Covariates (2) (3) (4) (5a) (5b) (5c) (6) (7) (8) Treatment *** 0.72*** 0.54*** *** *** 0.62*** (0.14) (0.14) (0.16) (0.17) (0.15) (0.15) (0.17) (0.14) Population (0.13) (0.12) (0.15) (0.07) (0.06) (0.14) (0.14) (0.14) GDP -2.13* -2.28* -2.69* -2.17* -2.33* -2.38* (1.29) (1.26) (1.47) (1.31) (1.33) (1.33) Intercept -0.37*** -0.71*** -0.58*** -0.68*** -0.43*** -0.25*** -0.39*** -0.90*** -0.80*** (0.10) (0.11) (0.08) -0.1 (0.08) (0.08) (0.08) (0.16) (0.12) Observations Note: Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Source: Authors calculations Treatment 2: Bordering the Eligibility Threshold Treatment 3: Passing Control of Corruption Treatment 4: Threshold Partnership Treatment 5a: Countries in the Middle Range of Government Expenditures Treatment 5b: Countries in the Highest Quartile of Government Expenditures Treatment 5c: Countries in the Lowest Quartile of Government Expenditures Treatment 6: Bordering the Eligibility Threshold and Passing Control of Corruption Treatment 7: Bordering the Eligibility Threshold or Passing Control of Corruption Treatment 8: Passing or Falling Short of the Eligibility Threshold The countries that display qualitative evidence of the MCC incentive effect are statistically positively correlated with five treatments: Passing Control of Corruption, Threshold Partnership, Countries in the Middle Range of Government Expenditures, Bordering the Eligibility Threshold or Passing Control of Corruption, and Passing or Falling Short of the Eligibility Threshold. The countries that display qualitative evidence of MCC incentive effects are statistically negatively correlated with one treatment: Countries in the Lowest Quartile of Government Expenditures. Countries in the Middle Range of Government Expenditures and Countries in the Lowest Quartile of Government Expenditures both statistically correlate with the countries that display qualitative evidence of the MCC incentive effect, but with opposite signs, and Countries in the Highest Quartile of Government Expenditures has no statistical correlation. These results coincide with our conjecture that countries with large expenditures are not incentivized, and countries with low government expenditures lack the power or political capacity to comply. Countries with government expenditures in the middle range react to MCC actively. Two other treatment groups have no statistical correlation with the countries that display qualitative evidence: countries in the Bordering the Eligibility Threshold, and countries in the Bordering the Eligibility Threshold and Passing Control of Corruption. 15

28 Partially due to this initial comparison, we selected five treatments for further investigation: Qualitative MCC Incentive Effect, Bordering the Eligibility Threshold, Passing Control of Corruption, Threshold Partnership, and Countries in the Middle Range of Government Expenditures. We keep the Bordering the Eligibility Threshold treatment based on its highly plausible assumptions and client interest. Although we present the Qualitative MCC Incentive Effect group here as an approximation of incentivized countries, we use this group as a treatment (like the others) in further analysis. Appendix F lists the LICs included in each of these five treatments (table F1) and LMICs (table F2). V. Preliminary Analyses In this section, using two nonparametrc estimation approaches for preliminary analysis of our initial sample with all LICs and LMICs, we find some differences among our five treatment groups and their corresponding control groups. All of the treatment groups show distinctively different trends compared to the control groups for the vast majority of the indicators. These results provide further ground for a quantitative analysis of the MCC incentive effect based on our treatment groups using more sophisticated methodology, which we do in Section VII. A. Nonparametric Approach Plotting the data for all 98 countries in our sample by indicator and treatment using locally weighted scatterplot smoothing (LOWESS), we initially compare the treatment and control groups. A merit of LOWESS is that it does not depend heavily on any model assumptions; rather, it allows us to provide a visual presentation of the data (Cleveland 1979). The rates of reform of countries show how much their indicator scores change from one time period to another. If no extraordinary events cause the rates of reform of our treatment and control groups to change from 2002 to 2010, holding other factors equal, one would expect the rates of reform for both groups to follow the same trend. Visually, the trend lines in the LOWESS scatterplot would be parallel to one another. By plotting the indicator rates of reform for all treatment and control groups, we find three trends. Some indicators in some treatment groups show trends that suggest the MCC incentive effect exists and is influencing countries in the treatment groups to perform better on indicators over time than the control groups. For other indicator scatterplots, we find mixed or no evidence of the MCC incentive effect. And yet for others, we find what seems to be a negative effect, where countries in our treatment groups appear to perform worse over time on the indicator than the countries in the control groups. We plotted rates of reform of each country, for all 20 indicators, and for all five treatment groups that display qualitative evidence of the MCC incentive effect 16

29 (100 scatterplots). In figures 2, 3, and 4, we present three scatterplots for Control of Corruption to illustrate the three general trends we see for the treatment groups: the MCC incentive effect, a mixed or nonexistent effect, and a negative effect. In figure 2, the Bordering the Eligibility Threshold treatment shows distinctively different rates of reform for Control of Corruption when compared to its control group. Figure 2. Qualitative MCC Incentive Effect: Bordering the Eligibility Threshold Treatment Rates of Reform for Control of Corruption Indicator Rates of Reform Score Source: Authors calculations Time Period The treatment group (solid diamonds and dashed line) has a lower score on Control of Corruption than does the control group (circles and solid line) prior to the establishment of the MCC. As time passes, treatment group countries seem to improve faster and better on their Control of Corruption indicator rates of reform score than do the control group countries, as indicated by the dashed line 17

30 surpassing the solid line for and diverging further from the solid line for This improvement suggests the MCC incentive effect can be found in countries in the Bordering the Eligibility Threshold treatment group for Control of Corruption. Our data also suggest in this example that prior to the establishment of the MCC, the control group had higher rates of reform on policies related to Control of Corruption than did treatment countries, as shown by the solid line appearing above the dotted line for This initial difference in the rates of reform indicates that policy reform has slowed or worsened in the countries in the control group. In other words, the indicator scores of countries that we do not believe might be subject to the MCC incentive effect are improving slower than are scores of the counterpart countries that border the MCC eligibility threshold. In figure 3, the Qualitative MCC Incentive Effect treatment displays a mixed or nonexistent effect for Control of Corruption when compared to the control group. Figure 3. Mixed or Nonexistent Effect: Qualitative MCC Incentive Effect Treatment Rates of Reform for Control of Corruption Indicator Rate of Reform Score Source: Authors calculations Time Period 18

31 The rates of reform for the countries for the Qualitative MCC Incentive Effect treatment group are higher than the rates of reform for the control group countries in However, the rates of reform for the treatment group decline in , rise above the control group in , and decline again in The overall trend for the treatment countries does not show any consistent direction of change for Control of Corruption, while the control group s consistent near zero scores throughout implies no change. In figure 4, the Countries in the Middle Range of Government Expenditures treatment displays a negative effect for Control of Corruption when compared to the control group. Figure 4. Negative Effect: Countries in the Middle Range of Government Expenditures Treatment Rates of Reform for Control of Corruption Indicator Rates of Reform Score Source: Authors calculations Time Period 19

32 The rates of reform for the Countries in the Middle Range of Government Expenditures treatment group is at first significantly higher than control group countries for Control of Corruption in The rates of reform then decline in the later time periods as they converge with the control group s rates. The different starting points for the lines in figures 2, 3, and 4 suggest that the natural rates of reform, those that would occur without MCC incentives, might differ for the treatment groups compared to their control groups. If we assume that treatment group countries are subject to the MCC incentive effect and that control group countries are not, and the two lines diverge in later time periods, the MCC incentive effect likely exists (as seen in figure 2). If the dashed line and the solid line in later time periods, on average, move together or not at all, there is likely no MCC incentive effect (as seen in figure 3). When the two lines converge (as seen in figure 4), the MCC incentive effect is negative for the treatment group countries in the sense that their rates of reform are slowing down. B. Basic Rates of Reform Mean Comparison In our second nonparametric approach, we compare the expectation of change for the rates of reform of the treatment groups to the control groups through the mean of the rates of reform of all countries for each treatment, respective control, indicator, and time period. If the MCC incentive effect exists for the treatment group, the average rates of reform should increase for treatment groups and remain the same for the control groups. On the contrary, we find that the average rates of reform for the control groups differ in the three periods. In other words, the control groups we established also exhibit different trends for the various indicators. At the same time, however, the variation for the control groups is smaller than the variation for the treatment groups. For the basic rates of reform mean comparison we also find three trends the MCC incentive effect, a nonexistent or mixed effect, and a negative effect (where countries perform worse). Again, we calculate but do not present all 100 comparisons for the 20 indicators and the five treatment groups; instead, we choose the same indicator and treatment group as presented in figures 2, 3, and 4. 20

33 In figure 5, the mean rates of reform of Control of Corruption for the Bordering the Eligibility Threshold treatment show an example of the MCC incentive effect. For the treatment group, the difference in average rates of reform for the second time period after the MCC was established ( ) and the baseline period ( ) is For the control group, the change from to is The difference in these two differences is The positive number implies the MCC incentive effect exists. By similar calculation, the MCC does not provide an incentive in , but does in , as compared to Figure 5. Qualitative MCC Incentive Effect: Bordering the Eligibility Threshold Treatment Mean Rates of Reform for Control of Corruption Time Period Mean Rates of Reform Score Note: Values presented next to each bar in the figure are rounded to the nearest second decimal. Source: Authors calculations 21

34 In figure 6, the mean rates of reform of Control of Corruption for the Qualitative MCC Incentive Effect treatment show an example of a mixed or nonexistent effect. The treatment group illustrates a change of 0.00 from to , while the control group s change is The difference between the two is +0.01, which weakly indicates existence of the MCC incentive effect. However, for the control group s change is from , while the treatment group s change is -0.07, making the difference-in-difference estimate and suggesting the existence of a negative effect. The oscillation between positive and negative results indicates a mixed or nonexistent MCC incentive effect. Figure 6. Mixed or Nonexistent Effect: Qualitative MCC Incentive Effect Treatment Mean Rates of Reform of Control of Corruption Time Period Mean Rates of Reform Note: Values presented next to each bar in the figure are rounded to the nearest second decimal. Source: Authors calculations 22

35 In figure 7, the mean rates of reform of Control of Corruption for the Countries in the Middle Range of Government Expenditures treatment show an example of a negative effect. For the treatment group, the difference between and is The control group experiences a change of The difference in these two differences is -0.06, which implies a negative MCC incentive effect. Similar calculations show the effect to be negative for the other two time periods. Figure 7. Negative Effect: Countries in the Middle Range of Government Expenditures Treatment Mean Rates of Reform of Control of Corruption Time Period Mean Rates of Reform Note: Values presented next to each bar in the figure are rounded to the nearest second decimal. Source: Authors calculations VI. Models In this section we describe three variations of the difference-in-differences (DD) model we use to investigate the MCC incentive effect. A. Rate of Reform: Difference-in-Differences Model 1 We adopt the DD model as applied by Johnson and Zajonc (2006) but with a longer time horizon; we use data from rather than We choose the DD model because of its relative ease of implementation and its ability to control for fixed effects. The DD model is also widely used in estimating causal relationships (Bertrand, Duflo, and Mullainathan 2002). 23

36 This DD model is written as follows: Y Y α β period γ treatment c X δ treatment period (2a) Or: R α β period γ treatment c X δ treatment period (2b) where Y is the score of a particular indicator of a specific country in a given year t, and R is the rate of change (i.e., rate of reform) of a country from year t to year t+2. Period is a categorical variable indicating into which two-year period (between 2002 and 2010) the data fall. Treatment is a categorical variable indicating whether a country is in the treatment group during period t. X stands for country-specific characteristics we believe may influence a country s rate of reform. The coefficient δ, also called the DD estimator, measures the MCC incentive effect. The DD estimator is expressed as: DD E Y Y Y Y Y Y Y Y (3) For our analysis, we set the baseline as year 2002; other years used are 2004, 2006, 2008, and We examine four time periods. The MCC was announced in 2002 but did not begin operations until 2004; therefore, is an appropriate baseline or control period. The other three periods we use are: , , and The differences subtract all fixed effects. The first subtraction of indicator scores of a given country leaves us with the rate of reform ~ Y Y. Our model uses a two-year difference of the indicator scores to reduce the possibility of a single-year endogenous shock affecting our results; for example, a bad growing season may cause the inflation rate to increase and then recover in the following year. The second subtraction removes fixed time effects of the rate of reform for each group R R. This second subtraction is also known as the natural rate of reform; in other words, it is the rate of reform in the absence of the MCC. The third difference is the subtraction of the change of natural rate of reform of the control group from the natural rate of reform change in the treatment group. This subtraction accounts for the possibility that countries in the treatment groups might have differences in their rates of reform compared to countries in the respective control groups. The third difference, if it statistically exists, is where we would find the MCC incentive effect, as seen through the DD ~ estimator. 24

37 B. Likelihood of Improvement: Difference-in-Differences Model 2 We use a probit regression model to estimate the likelihood of a given country to improve on a given indicator. We refer to the results from this model as Likelihood of Improvement. This model s output is originally in terms of the change in the indicator s standard deviation. Our statistical software converts the standard deviation into a probability for easier interpretation. For this model, we adopt an approach similar to Johnson and Zajonc (2006) by defining a DD estimator as: DD E 1 Y Y 1 Y Y 1 Y Y 1 Y Y (4) where 1{ } is a function of whether the expression in the parenthesis is true. Given a particular period, when a country shows a positive change, i.e., improvement in performance on an indicator, a 1 is assigned. A zero is assigned otherwise. Other variables are the same as described in sub-section A. C. Proportion of Indicators Improved: Difference-in-Differences Model 3 In our third model we calculate the number of improved indicators as a proportion of all indicators in a given period (Johnson and Zajonc 2006). If the MCC incentive effect exists, the treatment group should have a larger proportion of improved indicators compared to the control group. By employing the following calculation (equation 5), we combine all of the MCC indicators into one variable. P 1 Y, Y, (5) The DD estimator for this outcome is: DD E P P P P (6) D. Assumption and Caveat We assume that the rates of reform of indicators for one country are independent of the rates of reform for other countries within the same treatment or control group. This assumption is intuitively true; one impoverished country s domestic policy is not likely to be correlated with that of another impoverished country, especially for the good governance indicators. For example, policies to control corruption in China have little or no effect on the same policy in Indonesia. A potential problem would be that some economic indicators, such as inflation rates, might not satisfy this assumption, due to economic globalization. The magnitude and direction of such effects are unknown; therefore, as a simplifying assumption we ignore this possibility in our analysis. A major caveat to our analysis is that, as already stated, we do not have a natural experiment. We consider this a quasi-experiment, because we select the treatment groups to investigate whether the MCC incentive effect exists; our results may contain selection bias (Remler and Van Ryzin 2011). 25

38 VII. Difference-in-Differences Analysis for All Sample Countries In this section we present the results from our three DD models using our initial sample with all LICs and LMICs. In each sub-section we present and discuss statistically significant results and briefly summarize other results. Tables presenting all the results can be found in appendix G. Table 3 summarizes the number of indicators with statistically significant results identifying the MCC incentive effect for all the countries. We disaggregate results for each model by the three indicator categories, the five main treatment groups, and whether the analysis suggests the existence of a positive (the MCC incentive) effect or a negative effect. The total number of positive and negative results by indicator category for each of the models appears at the bottom of the table. 26

39 Treatment Table 3. Number of Positive and Negative Results of Three Models for Three Categories of Indicators Economic Freedom Model 1 Rate of Reform Investing in People Ruling Justly Economic Freedom Model 2 Likelihood of Improvement Investing in People Ruling Justly Economic Freedom Model 3 Proportion of Indicators Improved Investing in People Ruling Justly 1: Qualitative MCC Incentive Effect : Bordering the Eligibility Threshold : Passing Control of Corruption : Threshold Partnership a: Countries in the Middle Range of Government Expenditures Totals of Positive and Negative Results All results: 38+ / Source: Authors Calculations Economic Freedom Indicators: Credit Depth, Legal Rights, Cost to Start Business, Time to Start Business, Fiscal Policy, Inflation, Time to Register Property, Cost to Register Property, Regulatory Quality, Trade Freedom Investing in People Indicators: Girls Primary Education, Health Expenditures, Immunization Rate, Education Expenditures Ruling Justly Indicators: Civil Liberties, Control of Corruption, Government Effectiveness, Political Rights, Rule of Law, Voice and Accountability 27

40 A. Overview of Results We recall here the two main results from the previous quantitative studies of the MCC incentive effect. Johnson and Zajonc (2006) found the MCC incentive effect, but it was not statistically significant. Öhler, Nunnenkamp, and Dreher (2010) find a statistically significant effect on the one indicator they investigated (Control of Corruption), but it diminished over time. The overall results from our quantitative analysis do not provide strong support for the existence of the MCC incentive effect. Of 645 possible instances of statistical significance listed in appendix G, 38 (or 6 percent) exhibit a positive and statistically significant MCC incentive effect. At the same time, there are almost as many statistically significant negative results (32 results, or 5 percent). Negative results (i.e., a negative effect) suggest that the rates of reform are worsening in the treatment group countries compared to the respective control group countries. The negative results are not isolated to a particular model or treatment group in this analysis. In relation to this pattern, three points are worth noting. First, even though we present only the treatment groups that have the strongest qualitative evidence and preliminary support for being affected by MCC incentives, we find statistically significant results that suggest MCC incentives do not always prompt countries to improve their performance on the various indicators. Second, observers should ask whether a country s receipt of MCC funds may affect the presence of the MCC incentive effect. Third, we find an interesting phenomenon in our models: when we exchange the treatment and control groups, the effect often, though not always, switches sign accordingly in a statistically significant way. For example, when we switch treatment and control groups in the Passing or Falling Short of the Eligibility Threshold treatment (that is, observing results from treatment 8a and 8b), almost every DD estimator changed signs. The near symmetry of our model implies that the MCC has a trade-off between its policies. Furthermore, the model can determine which countries the MCC incentive effect will influence. The Qualitative MCC Incentive Effect treatment group shows the most statistically significant results (about one-third of all the significant results we present). Fifty-seven percent of these quantitative results identify the MCC incentive effect. The Countries in the Middle Range of Government Expenditures treatment group has fewer statistically significant results, but of the 16 we find, 62 percent show the MCC incentive effect. The Bordering the Eligibility Threshold treatment group has six significant results (5 percent of all possible results for that treatment), but all of those results show the MCC incentive effect. Countries in the Passing Control of Corruption treatment group do not really show better results on improving policies than countries that are in the Passing Control of Corruption control group. On one hand, the Control of Corruption indicator correlates with other Ruling Justly indicators, so the treatment group might not 28

41 show results that differ from the control group because the treatment group has met minimum requirements on a number of indicators. One could ask why a country should improve a Ruling Justly indicator to get MCC funds if it has already met requirements for Control of Corruption. On the other hand, since Control of Corruption correlates with the other five Ruling Justly indicators, one would assume that it would be easier for this group of countries to improve on these indicators. That is, these countries have already passed the highest hurdle, and the cost of improving the other scores, given that they might not be too far away from the required scores, should therefore be relatively lower. In turn, one might expect these countries to show stronger evidence of the MCC incentive effect in our data, but they do not. The countries in the Threshold Partnership treatment group show the lowest rate of result of the MCC incentive effect, despite these countries being subject to economic incentives as well as the MCC incentive effect. Twenty-nine percent of the statistically significant results we present for Threshold Partnership suggest the MCC incentive effect. Our indicator mobility analysis suggested that the Economic Freedom indicators are the most volatile in our dataset. Table 3 only partially confirms those results. The Economic Freedom indicators provide 33 percent of all significant results, and 32 percent of the MCC incentive effect results. The Investing in People indicators provide 28 percent of all significant results, and 29 percent of the MCC incentive effect results. The Ruling Justly indicators provide 39 percent of all significant results, and 39 percent of the MCC incentive effect results. 7 In the following section we present the results for each indicator category, as well as plausible conjectures for those results when possible. Although the overall results are very mixed, with almost half the regressions with statistically significant results finding a negative effect, we focus more on those results that identify the MCC incentive effect. B. Economic Freedom The Economic Freedom category contains 22 statistically significant results in the Rate of Reform and Likelihood of Improvement models, the most of the three indicator categories. Table 4 displays statistically significant results for each of the five treatment groups separated by the two models. The results are further disaggregated by time period. Appendix G displays all results. (Table G1 shows results for the Rate of Reform model, and table G2 shows results for the Likelihood of Improvement model for the Economic Freedom category.) 29

42 Table 4. Statistically Significant Results for Economic Freedom Treatment Model Indicator : Qualitative MCC Incentive Effect 2: Bordering the Eligibility Threshold 3: Passing Control of Corruption 4: Threshold Partnership 5a: Countries in the Middle Range of Government Expenditures Rate of Reform Cost to Start Business 53.3** Likelihood of Improvement (26.94) Cost to Register Property 1.205** (0.601) Trade Freedom * (3.188) Regulatory Quality 0.429*** 0.456*** (0.146) (0.144) Rate of Reform Trade Freedom 5.808* Rate of Reform Legal Rights 0.509** Likelihood of Improvement (0.235) Inflation 3.714* (2.251) Cost to Start Business -50.2* (26.13) (3.145) Trade Freedom * ** (3.172) (3.180) Cost to Register Property 0.208** (0.0981) Fiscal Policy ** Regulatory Quality ** (0.141) Rate of Reform Time to Register Property * Credit Depth ** (0.330) (14.45) Rate of Reform Time to Start Business 10.73* 11.75** Likelihood of Improvement Source: Authors calculations Trade Freedom 5.291* (5.640) (5.508) (3.138) Inflation * Regulatory Quality * (0.0592) Cost to Start Business 0.229** (0.109) (2.447) (0.215) Standard errors in parentheses ***p<0.01, **p<0.05, *p<

43 1. Qualitative MCC Incentive Effect Qualitative MCC Incentive Effect treatment countries show a statistically significant decrease for Cost to Start Business and Cost to Register Property in the Rate of Reform model. Trade Freedom shows a negative effect, meaning trade barriers have increased or reforms to loosen them have slowed in comparison to the control group. One plausible reason is that the increase of private business investments might have resulted in tightening trade barriers to protect the new producers in the economy. Although consumers enjoy lower prices when trade barriers are lower, many producers are not able to compete with the new world prices of goods, and a decline in business investment is likely. Therefore, this negative result for Trade Freedom is reasonable. The magnitude of the rate of reform for Trade Freedom is notable. The coefficient suggests that countries in this treatment will, on average, have a decrease of 5.7 on their scores compared to countries in the group; this indicator is measured on a scale of 0 to 100. The Likelihood of Improvement model shows statistical significance for the Regulatory Quality indicator in and with coefficients of and respectively, which implies that the countries in this treatment are on average 42.9 percent (45.6 percent for ) more likely to improve their Regulatory Quality score compared to countries in the control group during the respective time periods. 2. Bordering the Eligibility Threshold Bordering the Eligibility Threshold treatment countries show one result that is statistically significant and positive in the Rate of Reform model: Trade Freedom in , with an estimator of 5.8. The result implies that countries in this treatment group, on average, would have an increase of 5.8 on their Trade Freedom score. Also, we notice that the estimator for Trade Freedom has close to the same magnitude as in the Qualitative MCC Incentive Effect treatment, but in the opposite direction: reform is taking place to increase free trade. This change of direction might occur because the countries in the Qualitative MCC Incentive Effect treatment group could be in an earlier stage of economic development and be seeking trade barriers to protect domestic business, while countries in the Bordering the Eligibility Threshold treatment group could be in the process of lowering trade barriers because they are in a later stage of development. Overall, we find few statistically significant results for countries in this treatment group. Our indicator mobility analysis suggested Economic Freedom indicators may be easier to reform, so this lack of results could occur because countries that are bordering the eligibility threshold have, on average, already met more minimum requirements for in the Economic Freedom indicators than indicators in Investing in People or Ruling Justly. 3. Passing Control of Corruption Passing Control of Corruption countries have the greatest number of statistically significant results, but they are mixed. Three indicators identify the MCC incentive effect. Four indicators show a negative effect. Trade Freedom shows a negative result in and , but not ; the rate of 31

44 reform for this indicator is not consistent over time. (The effect of time is discussed in section VIII.B.) The combination of two possible reasons could explain the mix of positive and negative results. First, we expect many reforms to occur in the Economic Freedom category because these indicators are easier to change than indicators in the Ruling Justly and Investing in People categories. The countries in the Passing Control of Corruption treatment group passed the minimum requirements for the Control of Corruption indicator and need to meet the MCC s overall eligibility threshold (at least half of indicators should meet minimum requirements) to be considered compact-eligible. Because Economic Freedom indicators are relatively easy to improve upon, we expect more changes to take place in this category. Second, we believe a trade-off exists between indicators within this category. For example, to encourage domestic business investments, a country may increase trade barriers, which would show up as a negative effect. These two reasons combined may explain the mixed evidence of the MCC incentive effect in the Passing Control of Corruption treatment group. 4. Threshold Partnership Threshold Partnership treatment countries have two statistically significant negative results for Economic Freedom indicators in the Rate of Reform model; Time to Register Property, on average, increases by 24.4 days and the score for Credit Depth, on average, declines by 0.7 points (on a scale of 0 to 6) compared to countries in the control group. No statistically significant results exist for the Likelihood of Improvement model for this treatment. Overall, the MCC incentive effect is not evident for the countries that have been receiving extra assistance from the MCC to become eligible to enter into a poverty-reduction compact. 5. Countries in the Middle Range of Government Expenditures The Countries in the Middle Range of Government Expenditures treatment shows six statistically significant results. The Rate of Reform model shows positive, statistically significant results for Time to Start Business for and , implying that the time to start a business is declining. Similarly, Trade Freedom shows a positive, statistically significant result for , implying trade barriers are decreasing. The other positive, statistically significant result is for the Cost to Start Business in the Likelihood of Improvement model with an estimator of This estimator means that countries in this treatment are 22.9 percent more likely to reduce the cost to start a business compared to the control group countries. C. Investing in People The Investing in People category has 17 statistically significant results, almost as many as Economic Freedom. Table 5 presents results for the three treatment groups with statistically significant results separated by the Rate of Reform model and Likelihood of Improvement model; the results are further disaggregated by time period. Appendix G s table G3 displays all results for Investing in People for both models. 32

45 Table 5. Statistically Significant Results for Investing in People Treatment Model Indicator Rate of Reform Girls' Primary Education 7.215*** 1: Qualitative MCC Incentive Effect 4: Threshold Partnership 5a: Countries in the Middle Range of Government Expenditures Source: Authors calculations Likelihood of Improvement (2.762) Immunization Rate ** (2.224) Girls' Primary Education 0.353** 0.487*** 0.366** (0.166) (0.164) (0.178) Rate of Reform Education Expenditures 0.911** Likelihood of Improvement (0.373) Health Expenditures ** *** (2.127) (2.134) Immunization Rate ** (2.540) Health Expenditures *** ** (0.165) (0.166) Rate of Reform Girls' Primary Education 7.085** 5.815** Likelihood of Improvement Immunization Rate * (2.733) (2.752) (2.125) Girls' Primary Education 0.322* 0.354** 0.454** (0.169) (0.167) (0.174) Standard errors in parentheses ***p<0.01, **p<0.05, *p< Qualitative MCC Incentive Effect Qualitative MCC Incentive Effect treatment countries show a statistically significant, positive result for Girls Primary Education. The Rate of Reform model shows an average increase in girls primary education of 7.2 percent for countries in the Qualitative MCC Incentive Effect treatment group compared to the countries in the control group. The Likelihood of Improvement model has positive, statistically significant results for all three time periods for Girls Primary Education: a 35.3 percent likelihood of improvement in , a 48.7 percent likelihood of improvement for , and a 36.6 percent likelihood of improvement for for countries in the treatment group when compared to countries in the control group. 2. Bordering the Eligibility Threshold Bordering the Eligibility Threshold treatment countries have no statistically significant results for the Rate of Reform or Likelihood of Improvement models. We do not have a plausible conjecture as to why we find no results. 33

46 3. Passing Control of Corruption Passing Control of Corruption treatment countries do not display any statistically significant results in the Rate of Reform or Likelihood of Improvement models. Again, we do not have a plausible conjecture as to why we find no results. 4. Threshold Partnership Threshold Partnership treatment countries have statistically significant results; however, all but one implies a negative effect. The only positive result is Girls Primary Education for in the Rate of Reform model. 5. Countries in the Middle Range of Government Expenditures The Countries in the Middle Range of Government Expenditures treatment group has positive, statistically significant results for Girls Primary Education in the Rate of Reform model for and , and in the Likelihood of Improvement model for all three time periods. These results imply that the Countries in the Middle Range of Government Expenditure group are improving the primary education completion rates for girls more than the control group countries. The Immunization Rate indicator is statistically significant and negative for The estimator of suggests that countries in the middle range of government expenditure, on average, decrease immunization coverage by 3.88 percent compared to the countries in the control group. A 3.88 percent drop in immunization rate coverage may not be much of a change in practical terms. D. Ruling Justly The Ruling Justly category has 26 statistically significant results for the Rate of Reform and Likelihood of Improvement models. Table 6 displays the statistically significant results for each of the five treatment groups separated by the two models; the results are further disaggregated by time period. Appendix G displays all results. (Table G4 shows results for the Rate of Reform model, and table G5 shows results for the Likelihood of Improvement model for the Ruling Justly category.) 34

47 Table 6. Statistically Significant Results for Ruling Justly Treatment Model Indicator : Qualitative MCC Incentive Effect 2: Bordering the Eligibility Threshold 3: Control of Corruption 4: Threshold Partnership 5a: Countries in the Middle Range of Government Expenditures Rate of Reform Civil Liberties 1.566* Likelihood of Improvement (0.878) Political Rights 2.398** (1.146) Voice and Accountability ** *** *** Political Rights 0.366** (0.0575) (0.0579) (0.0579) (0.139) Control of Corruption 0.278* (0.148) Voice and Accountability ** *** ** (0.145) (0.142) (0.145) Rate of Reform Control of Corruption 0.112* Likelihood of Improvement Rate of Reform (0.059) Voice and Accountability 0.102* (0.0553) Control of Corruption 0.255* 0.322** 0.374*** Government Effectiveness Rate of Reform Civil Liberties 2.307** Likelihood of Improvement (0.139) (0.138) (0.138) 0.138** 0.103* (0.0553) (0.0556) (1.006) Political Rights 2.443* (1.335) Rule of Law ** (0.0611) Control of Corruption 0.427** (0.170) Political Rights 0.284* Rule of Law ** (0.177) Rate of Reform Rule of Law * Likelihood of Improvement Source: Authors calculations (0.0519) (0.168) Voice and Accountability *** (0.0559) Voice and Accountability ** (0.374) Standard errors in parentheses ***p<0.01, **p<0.05, *p<

48 1. Qualitative MCC Incentive Effect Qualitative MCC Incentive Effect treatment countries have positive, statistically significant results for Civil Liberties and Political Rights for the Rate of Reform model in While these results are statistically significant, the magnitude of the change in rate of reform may not be practically significant because Civil Liberties is measured on a scale of 0 to 60, and Political Rights is measured on a scale of 0 to 40. The estimators on these are and 2.398, implying that countries in this treatment group have, on average, an increase in their scores of and 2.398, compared to the countries in the control group. 2. Bordering the Eligibility Threshold Bordering the Eligibility Threshold treatment countries have three statistically significant results, all of which are positive. Voice and Accountability, and Control of Corruption are statistically significant and positive for in the Rate of Reform model. Control of Corruption is statistically significant and positive for all three time periods in the Likelihood of Improvement model. This last result could be because the countries in this treatment group are close to or have met the MCC s eligibility threshold and may need to only meet the minimum requirements for Control of Corruption to be considered eligible for a poverty-reduction compact. 3. Passing Control of Corruption Passing Control of Corruption treatment countries have only one indicator that is statistically significant, Government Effectiveness. It is positive in two time periods, and Countries in this treatment group should have a higher score on the Ruling Justly indicators than the control countries because Control of Corruption correlates strongly with most of the other Ruling Justly indicators. Also, because of the high correlation among the Ruling Justly indicators, one would assume that it would be easier, or less costly, for a country that has already passed the Control of Corruption benchmark to improve and pass other Ruling Justly thresholds. So, to see such limited results is somewhat surprising. Yet, due to the correlation, some problems with endogeneity exist within this treatment group. 4. Threshold Partnership Threshold Partnership treatment countries have very similar results to countries in the Qualitative MCC Incentive Effect treatment group. Threshold Partnership countries have received economic support from the MCC to help surmount the final hurdles to be considered eligible for a poverty-reduction compact. Most of these programs are targeted toward improving Ruling Justly indicators (MCC 2013c), most notably Control of Corruption, which, again, is highly correlated with other Ruling Justly indicators. In addition to any MCC incentives, these countries receive economic incentives and receive direct moral support and help from the MCC to improve their policy indicators. However, our data do not provide quantitative documentation of the effects of these programs. Threshold Partnership countries are exposed to several incentive effects to improve their 36

49 MCC selection indicators and are relatively close to passing the MCC eligibility threshold; therefore, they should show evidence of the MCC incentive effect if the effect exists. As we are not quantitatively able in our analysis to subtract the economic incentives this group of countries is subject to, our numbers might be positively biased. Yet, statistically these countries are not different from the control countries. 5. Countries in the Middle Range of Government Expenditure Countries in the Middle Range of Government Expenditures treatment group have statistically significant results, however they are all negative. We do not conjecture as to why here. E. Results from Proportion of Indicators Improved Model The Proportion of Indicators Improved model measures whether the MCC incentive effect influences the proportion of indicators improved. Statistically significant results are presented in table 7. A full summary of the results is listed in appendix G, table G6. The merit of this method is that aggregation of indicators allows us to locate the MCC incentive effect in a bigger framework, not only indicator by indicator as in the Rate of Reform, or Likelihood to Improve models. In this section, we delve into the Proportion of Indicators Improved model for four groups of indicators: all indicators together and the three indicator categories: Economic Freedom, Investing in People, and Ruling Justly. Because aggregating 20 indicators results in the largest number of missing values, a missing value for any one of the 20 indicators would contaminate the aggregated results by causing a loss of a fraction of information. Thus we treat missing values conservatively: whenever a missing value appears, we assume the indicator has not improved. This procedure comes at the cost of a possible downwardly biased MCC incentive effect estimator. Also, for the dependent variable, we apply a logarithmic transformation to get a better linear relationship. (The interpretation of coefficient is, therefore, the percentage change of the proportion of indicators improved.) Table 7: Statistically Significant Results for Proportion of Indicators Improved Model Treatment Model Indicator Category : Qualitative MCC Incentive Effect 4: Threshold Partnership 5a: Countries in the Middle Range of Government Expenditures Source: Authors calculations Proportion of Indicators Improved Proportion of Indicators Improved Proportion of Indicators Improved Ruling Justly * All Indicators ** Economic Freedom ** Investing in People * *** Investing in People ** Standard errors in parentheses ***p<0.01, **p<0.05, *p<

50 The results are concentrated in two groups: negative results for the Threshold Partnership treatment group and a positive result for Countries in the Middle Range of Government Expenditures. For the Threshold Partnership treatment, all the statistical significant results we find are negative. For all indicators in there is a decrease of 5 percent, for the Economic Freedom indicators there is a decrease of 0.8 percent, and for the Investing in People indicators there is a decrease of 13 percent. For the Countries in the Middle Range of Government Expenditures treatment, although there are no statistically significant results for any of the time periods, the joint F test yields positive results that are significant at the 0.1 level, suggesting the MCC incentive effect exists. Although the direction of the MCC incentive effect is hard to discern without significance at the individual indicator level, considering that we have awarded all the missing values no indicator performance improvement, our data is likely to be negatively biased: the MCC incentive effect may be positive for the Countries in the Middle Range of Government Expenditures treatment group. VIII. DD Analysis for LICs and LMICs Up until this point we have treated LICs and LMICs as essentially the same. Yet, it is more intuitive that countries with different levels of economic activity will act in different ways. Preliminary evidence also indicated the possibility of a difference. For these reasons, we decide to investigate whether statistically evident differences exist between LICs and LMICs. A. LICs Breakdown In figure 8 we break down statistically significant results by individual indicator and indicator category (as well as model) for LICs. Most evident is that the statistical significance is not distributed evenly among the 20 indicators or among the three indicator categories. In the Economic Freedom category, we can see relatively little activity. We find an average of 2.0 statistically significant results per indicator, with 40 percent of those results positive in orientation. In comparison, Investing in People shows an average of 5.8 statistically significant results per indicator, with 48 percent of those results positive in orientation, while Ruling Justly shows an average of 3.3 statistically significant results per indicator, with 70 percent positive in orientation. The strongest individual indicator to show the MCC incentive effect in the Economic Freedom category is Cost to Start Business, with three positive and one negative result. The Investing in People category shows two stronger indicators: Girls Primary Education and Education Expenditures. The Ruling Justly category shows four strong indicators: Civil Liberties, Control of Corruption, Government Effectiveness, and Political Rights. 38

51 These results all suggest that LICs may tend to focus more on policies that affect the general governance of people, rather than policies that affect economic activity; this finding is intuitive given that these countries have little wealth and are under pressure from the wealthier countries to improve policies toward citizens. Regardless, 9.8 percent of all possible results for the MCC incentive effect in these countries show statistically significant results; of those, 52 percent is positive. B. LMICs Breakdown In figure 9, we break down statistically significant results by individual indicator and indicator category (as well as model) for LMICs. We see a different pattern than with LICs, most notably the lack of statistically significant positive results. Overall, 7.4 percent of all results are statistically significant, with 40 percent of them positive. More specifically, the Economic Freedom category averages 1.8 statistically significant results per indicator, with 44 percent positive. The Investing in People category averages 3.3 results per indicator, with 54 percent positive. The Ruling Justly category averages 2.8 results per indicator, with 24 percent positive. Of the individual indicators, we find the most evidence for the MCC incentive effect for Inflation, Girls Primary Education, and Control of Corruption one indicator from each of the three indicator groups. In comparison to the results from LICs, these results for LMICs could imply that LICs have stronger incentives to improve their policies to become eligible for MCC funding. One could expect a poorer country with lower GDP to have stronger incentives to attract aid. The results in part support the MCC distinction between the two income levels because they have different incentives. The major qualification, however, is that the LMIC group sample size of 29 countries is small, which could adversely affect the potential for statistically significant results. 39

52 Figure 8: Number of Positive and Negative Results in LICs Figure 8 depicts the cumulative number of statistically significant positive and negative results from the three models for LICs. We break down the results by indicators and group the indicators under their respective MCC category Economic Freedom, Investing in People, and Ruling Justly. The orientation of the bars is purely for visualization and has no other intended meaning. 15 Number of Results Statistically Significant Positive Results Rate of Reform Likelihood of Improvement Proportion of Indicators Improved Statistically Significant Negative Results Rate of Reform Likelihood of Improvement Proportion of Indicators Improved Economic Freedom Investing in People Ruling Justly Indicators Source: Authors calculations 40

53 Figure 9: Number of Positive and Negative Results in LMICs Figure 9 depicts the cumulative number of statistically significant positive and negative results from the three models for LMICs. We break down the results by indicators and group the indicators under their respective MCC category Economic Freedom, Investing in People, and Ruling Justly. The orientation of the bars is purely for visualization and has no other intended meaning. 15 Number of Results Statistically Significant Positive Results Rate of Reform Likelihood of Improvement Proportion of Indicators Improved Statistically Significant Negative Results Rate of Reform Likelihood of Improvement Proportion of Indicators Improved Economic Freedom Investing Indicators in People Ruling Justly Indicators Source: Authors Calculations 41

54 C. Time Effects We turn briefly to the issue of time periods. Given that time is needed for policies to be implemented and for results to be recorded, it is reasonable to assume that the MCC incentive effect will show up stronger in our data in time periods after the initial period from when the MCC was established. At the same time, Öhler, Nunnenkamp, and Dreher (2010) suggest diminishing returns as time passes. Our results are not conclusive about either argument, as seen in figures 10a, 10b, and 10c. Figures 10a, 10b, and 10c depict the cumulative number of statistically significant positive and negative results from the three indicator categories (Economic Freedom, Investing in People, and Ruling Justly) separated by LICs and LMICs. The orientation of the bars is purely for visualization and has no other intended meaning. Looking strictly at positive results in LICs, we see results somewhat similar to Öhler, Nunnenkamp, and Dreher (2010) in Economic Freedom (figure 10a) and Ruling Justly (figure 10c) an increase in results from the to , then a decrease in In Investing in People (figure 10b), we see an increase in results from to , followed by another increase in Positive results for LMICs show different patterns. In Economic Freedom (figure 10a), we see an increase in results from to , followed by stagnation in Investing in People (figure 10b) shows stagnation from to , followed by an increase in Ruling Justly (figure 10c) shows continuing decreases as time progresses, with no results for These results are mixed. In the best scenario, we see an increase in results over time. In the worst, the results started to decrease after the initial surge. This latter scenario could be due, for instance, to a general cooling off of interest by countries in the potential for MCC aid. Limited results from the current models restrict the possibility for definite conclusions. 42

55 8 Figure 10: Number of Positive and Negative Results in Different Time Periods (a) Economic Freedom Number of Results to to to 2010 Time Period Statistically Significant Positive Results Low-Income Countries Lower-Middle-Income Countries Statistically Significant Negative Results Low-Income Countries Lower-Middle-Income Countries Source: Authors calculations (b) Investing in People 8 Number of Results to to to 2010 Time Period Statistically Significant Positive Results Low-Income Countries Lower-Middle-Income Countries Statistically Significant Negative Results Low-Income Countries Lower-Middle-Income Countries Source: Authors calculations 43

56 (c) Ruling Justly 8 Number of Results to to to 2010 Time Period Statistically Significant Positive Results Low-Income Countries Lower-Middle-Income Countries Statistically Significant Negative Results Low-Income Countries Lower-Middle-Income Countries Source: Authors calculations IX. Conclusions and Recommendations We have quantitatively assessed the MCC incentive effect. Building on limited documentation, we identified five groups of countries that are the most likely to be incentivized to work to meet MCC eligibility criteria. We find the five country groups do not exhibit a strong, statistically significant MCC incentive effect and conclude that not enough quantitative evidence supports the argument that an overall MCC incentive effect exists. Our conclusion should come as no surprise. Explaining that changes in world politics and countries rationales for policies as being due to the existence of a foreign aid agency is a courageous task. What is interesting then is that we do find some positive and statistically significant results throughout our array of analyses in support of the MCC incentive effect. When we break down our analysis by country income level, we find statistically significant (albeit limited) quantitative evidence of the MCC incentive effect. In particular, LICs show evidence of the MCC incentive effect for indicators in the Ruling Justly category. Accordingly, we cannot rule out that the MCC incentive effect indeed exists at some level. We finish by discussing two of the treatment groups, by recommending ways to improve similar analyses, and by providing ways the MCC could improve its policies to foster the MCC incentive effect. 44

57 A. Lack of Quantitative Evidence Two of the treatment groups presented in this study, the Qualitative MCC Incentive Effect and the Threshold Partnership, should by the nature of their composition represent the countries that are most likely to be incentivized by the MCC. Therefore, these treatment groups should be the most likely to show a quantifiable MCC incentive effect. In this context, we are disappointed that we did not find a stronger MCC incentive effect for these two treatment groups when analyzing all countries in our sample together. The Qualitative MCC Incentive Effect treatment countries are the countries that the MCC itself believes it incentivizes to change their policies. In our study this treatment group is also the group that reveals the most statistically significant results throughout our models, with 21 significant results. However, just 57 percent of those rather limited results showed a positive effect. For the other 43 percent, the control countries did better than the countries that MCC staff assumes would show the MCC incentive effect. The Threshold Partnership countries are countries that, in addition to being subject to any MCC incentive effect, receive direct MCC support and ex-ante rewards to improve their MCC policy indicator scores. Since we cannot quantitatively distinguish these various incentives for this group of countries in our models, the additional Threshold Partnership incentives should add a positive bias to the MCC incentive effect. Therefore, the expectation should be that if the MCC incentive effect can be quantified, it should show up in our data in a more pronounced manner for this treatment group. However, in our overall results the Threshold Partnership treatment group is where we find the highest ratio of negative MCC incentive effects indicating that countries that do not belong to this treatment group in fact improve their policies more than those countries which do. In sum, we do not find strong evidence of a quantifiable MCC incentive effect. If one existed, we would have liked to see more positive results for these two treatment groups. Nonetheless, certain modifications to the models and the MCC selection process may allow these results to appear if they do exist. B. How to Improve Analyses: For future studies, we recommend the following: 1. Run randomized samples in treatment groups to get an empirical distribution to check for false positives. Although we believe we have analyzed the strongest possibilities for the MCC incentive effect, given the low number of positive results, we may be seeing false positives in our results. 2. Reduce the endogeneity inherent in the data and models. We recommend, for example, employing an instrument variable, or combining a regression discontinuity design with the DD model (fixed around countries with similar per-capita GDP). 45

58 3. Use multivariate longitudinal data analysis, which allows for more than one response variable to be analyzed simultaneously over time. 4. Add random effects for each country to adjust for some of the inherent heterogeneity that exists in countries decision-making structures related to policy change. A mixed effects model also would help expand the predictive ability of the model to each country. C. How to Improve MCC Incentives As explained in this report, recent research shows that countries that meet the MCC eligibility criteria are also more likely to receive foreign aid from other agencies (Dreher, Nunnenkamp, and Öhler 2012). Hence, there seems to be a tempting apple hanging in the tree. But are countries interested in or capable of harvesting the fruit? Or do they even know that the golden opportunity of MCC funding exists? Three recommendations that the MCC should take into consideration when working to improve the strength of the MCC incentive effect are: 1. Create Aspirations The first, and most obvious, is that the MCC must encourage countries aspirations to get MCC funding. Countries in need of or with a national interest in increasing the aid they receive should be encouraged more to comply with MCC eligibility criteria. For instance, countries where aid is a relatively large part of their GDP might be easier to incentivize and may provide a target group for the MCC, should the organization seek to strengthen the MCC incentive effect. 2. Provide Knowledge and Information The second criteria for an efficient MCC incentive is that countries with some interest in receiving aid must have better knowledge and information about MCC funding opportunities and, more importantly, about how they can comply with the criteria to meet the eligibility threshold. This point is perhaps the biggest Achilles heel of the MCC today. Throughout the MCC s nine years of existence, the number of policy indicators a country has to improve to achieve eligibility for MCC poverty-reduction compacts has changed on multiple occasions. Moreover, the requirements for complying with criteria have changed. (For example, in addition to the absolute threshold for Control of Corruption, after 2012, countries need to get a certain score on the Political Rights or Civil Liberties). Indeed, the indicator criteria are quite a jungle for a third party, such as the authors of this report, to navigate. These issues with criteria have forced certain restrictions on this quantitative analysis; more importantly, the issues most certainly impede countries interested in obtaining eligibility for MCC support in learning what to do or which policies to change. Actually knowing how one needs to improve on a policy to pass an indicator requirement is difficult. In sum, the road to meeting the MCC eligibility criteria is filled with unpredictable turns, and the map the countries are equipped 46

59 with does not provide them with the necessary motivation to get to the final destination. Although the annual MCC scorecards for each country provide some useful information for potential applicants, up until now there has been no official tote board for MCC scores. Since the MCC eligibility criteria are relative (for instance, the MCC board considers how countries perform relative to the median score of their income peer group), the actual status a country has relative to meeting the eligibility criteria is not transparent. We suggest three ways to deal with this problem: 1. Publish the ranking each country has for each indicator. Tote board diplomacy can be a cheap and effective way to promote interest, knowledge, and competition. 2. Reduce the number of indicators, particularly in the Ruling Justly category. Since the Ruling Justly indicators are highly correlated and change very slowly (both for treatment and control countries), cutting the number of these indicators has some incentivizing advantages, such as making the way forward for countries more clear cut regarding exactly which policies to change Set five-year plans for selection indicators. Maintaining consistency with criteria over a longer period of time will help improve transparency. If these three recommendations are combined, despite some potential problems, 9 the results will allow each country to better know how much is demanded to reach a certain MCC score. In addition, these actions may help certain countries that today do not see MCC as a potential source of aid to realize that they are actually closer to the eligibility threshold than they thought, thus incentivizing them to pursue eligibility. The result will strengthen the MCC incentive effect. 3. Separate Country Group LICs and LMICs Through our various analyses, we have found that some groups of countries appear to have more difficulty complying with some selection indicators than do others. Those countries might be limited by the current selection system. Although slight differences exist between the LICs and LMICs in terms of eligibility requirements, setting more distinct individual goals for the two income groups toward indicators they respectively show ability to improve on may help with overall performance. At the same time, this recommendation would come at a cost of some consistency and transparency, that cost is likely to be minimal if done in combination with the other recommendations. D. Final Thoughts Our report is at best inconclusive as to determining the existence of the MCC incentive effect. In comparison, Johnson and Zajonc (2006) conclude that the 47

60 MCC incentive effect exists, but they did not provide statistically significant data. Öhler, Nunnenkamp, and Dreher (2010) find a statistical significant effect on the Control of Corruption indicator, but said that it diminished over time. Our results do not confirm or refute either argument. At the same time, we have conducted a much more comprehensive study of the MCC incentive effect. We also transparently present both the positive and negative statistically significant results in the data. This report provides new perspectives on how the MCC incentive effect should be studied. We hope our work will lead to research and actions by the MCC that will more solidly point to and foster the MCC incentive effect. Such research and actions will, at the same time, be important in providing quantitative evidence of the MCC incentive effect and quantitative evidence of incentive effects in ex-post rewards systems. 48

61 Appendix A: Descriptions of Indicators This appendix describes 26 MCC indicators we initially considered and notes which six we eliminated for our analysis. A. Economic Freedom Indicators Indicators in the Economic Freedom category assess a government s commitment to strong economic policies and economic opportunities for its people through market forces (MCC 2012, 29). We draw on 10 out of 12 indicators described below in our analysis. 1. Credit Depth Credit Depth measures rules and practices affecting the coverage, scope and accessibility of credit information available through either a public credit registry or a private credit bureau. A country receives a score of 0 or 1 for each of six features available in those credit institutions. The final score is the summation of the six scores. The data come from the International Finance Corporation s Doing Business Survey. The MCC averages Credit Depth with Legal Rights to craft its Access to Credit indicator (MCC 2012, 33-35). We keep Credit Depth and Legal Rights separate in our analysis. 2. Legal Rights Legal Rights measures the extent to which bankruptcy and collateral laws protect the rights of borrowers and lenders to facilitate lending. It contains 10 aspects in collateral and bankruptcy law. A score of 0 or 1 for each of the 10 features of the laws, and the final score is the sum. The data come from the International Finance Corporation s Doing Business Survey. The MCC averages Legal Rights with Credit Depth to create its Access to Credit indicator (MCC 2012, 33-35). We keep Legal Rights and Credit Depth separate for our analysis. 3. Cost to Start Business Cost to Start Business measures how much it costs to start an industrial or commercial business while following all the required procedures. This indicator is measured as a percentage of the country s per-capita income. A smaller number for this indicator implies less cost to start a business, which is more desirable. The data come from the International Finance Corporation s Doing Business Survey. The MCC normalizes and averages Time to Start Business with Cost to Start Business to calculate its Business Start-Up indicator (MCC 2012, 35-37). We keep Time to Start Business and Cost to Start Business separate for our analysis. 4. Time to Start Business Time to Start Business measures the number of days it takes to start an industrial or commercial business while following all the required procedures. The data come from the International Finance Corporation s Doing Business Survey. The 49

62 MCC normalizes and averages Time to Start Business with Cost to Start Business to calculate its Business Start-Up indicator (MCC 2012, 35-37). We keep Time to Start Business and Cost to Start Business separate for our analysis. 5. Fiscal Policy The Fiscal Policy indicator provides insight to how the government uses spending and taxation to influence the economy (Horton and El-Ganainy 2012). The International Monetary Fund compiles data that measure general government net lending and borrowing (revenue minus total expenditure) as a percentage of GDP averaged over the three previous years (MCC 2012). 6. Inflation Inflation harms a country s economic growth (De Gregorio 1993). The Inflation indicator provides insight to a country s monetary policy because these policies influence the rate of inflation. The data come from the International Monetary Fund s World Economic Outlook Database and are calculated as the percentage change in consumer prices averaged throughout the year. The MCC requires that countries perform below the absolute threshold of 15 percent to pass this indicator s median (MCC 2012, 38-39). 7. Time to Register Property Time to Register Property indicates the number of days that are required for a business to purchase a property from another business and to formally transfer the property title to the buyer s name. Smaller Time to Register Property indicates less time to register properties, which is more desirable. The data come from the International Finance Corporation s Doing Business Survey. The MCC uses Time to Register Property as one of three parts of its Land Rights and Access indicator, the other two parts being Cost to Register Property and Access to Land (MCC 2012, 31-33). We keep the three parts separate for our analysis. 8. Cost to Register Property Cost of Register Property records the cost required to register property measured as a percentage of the value of the property necessary for a business to purchase a property from another business and to formally transfer the property title to the buyer s name. This indicator draws on data from the International Finance Corporation. The MCC uses Cost to Register Property as one of three parts of its Land Rights and Access indicator, the other two parts being Time to Register Property and Access to Land (MCC 2012, 31-33). We keep the three parts separate for our analysis. 9. Regulatory Quality Regulatory Quality measures the perception of policy formation and implementation in regard to promoting private sector development (MCC 2012, 30-31). The data come in the form of an index compiled by the World Bank that uses a scoring range from -2.5 to +2.5, where -2.5 represents poor regulatory 50

63 quality and +2.5 represents excellent regulatory quality (World Bank Group 2013e). 10. Trade Freedom Trade Freedom measures the barriers to trade that affect imports and exports. The Heritage Foundation compiles the data as part of its Index of Economic Freedom. Trade Freedom scores range between 0 and 100, where 0 represents high barriers to trade and 100 represents low barriers to trade. The calculation takes the weighted average of the tariff rates. Then a penalty can be imposed for non-tariff barriers determined by the pervasiveness of quantity, price, regulatory, investment, and customs restrictions, as well as direct government intervention. Depending on the pervasiveness, the score is reduced 5, 10, 15, or 20 percentage points (MCC 2012, 37-38). 11. Gender in the Economy Gender in the Economy measures the legal ability of women to interact with the private and public sector. Research suggests that the transition of women from the primary caretakers of the home to members of the workforce increases household income and GDP. Data for the indicator come from the World Bank and the International Finance Corporation (MCC 2012, 39-40). The indicator was added in year 2012 (Butts 2011); data are only available for 2010 and 2012, and the format has not been indexed (World Bank Group 2013b). In that regard, the MCC indicates it combines 20 data points to make its indicator, but it is not absolutely clear which data points are used and how the agency compiled them (MCC 2012, 39-40). Because of these limitations, we do not use this indicator in our analysis. 12. Access to Land Access to Land is one of three parts of the MCC s Land Rights and Access indicator, the other two parts being Time to Register Property and Cost to Register Property. The MCC collects the data from the International Fund for Agricultural Development (MCC 2012, 31-33). We were not able to collect data on this indicator; therefore, we exclude it from our analysis. B. Investing in People Indicators The Investing in People indicators assess the government s commitment to improving the standard of living of its people (MCC 2012, 23). We draw on four out of seven indicators as described below in our analysis. 1. Girls Primary Education Girls Primary Education uses the gross intake ratio in the last grade of primary school. It is calculated as the total number of girls enrolled in the last grade of primary school minus the number of female students repeating the last grade of primary, school divided by the total population of females that are of theoretical entrance age of the last grade of primary school. The data come from UNESCO s 51

64 Institute of Statistics to measure girls primary education completion. The MCC only uses the Girls Primary Education indicator for LICs (MCC 2012, 26-27). 2. Health Expenditures Health Expenditures measures a government s commitment to investing in the health and well-being of its people. Increased spending on health, when combined with good policies and good governance, can promote growth, reduce poverty, and trigger declines in infant, child, and maternal mortality. The MCC has chosen an indicator that is an input for the well-being of citizens because it may take time to observe the outcome of healthcare interventions. The data come from the World Health Organization and measure general government health expenditures as a percentage of GDP (MCC 2012, 24-25). 3. Immunization Rate The Immunization Rate of a country is used as a proxy to measure overall public health care. Immunization Rate is the rate of vaccinations in 1-year-olds, specifically for diphtheria-pertussis-tetanus and for measles. The indicator uses the simple average of the two vaccination rates based on a combination of health administrative data and coverage surveys collected by the World Health Organization and the United Nations Children s Fund. For fiscal year 2013, the MCC s requirements differ between LICs and LMICs. LICs need to perform at or above the median of the rate of their peers, while LMICs must obtain an average of at least 90 percent (MCC 2012, 23-24). 4. Education Expenditures Education Expenditures is the amount a government spends on primary education as a percentage of GDP. It is based on UNESCO s Institute of Statistics as its primary source of data. The MCC uses self-reported data from national governments as a secondary source. The country GDP is crosschecked with World Bank and International Monetary Fund s GDP estimates (MCC 2012, 25-26). We use only data from UNESCO in our analysis. Additionally, the MCC uses a measure of primary education expenditures as a percentage of GDP. We could not find this exact measure directly from UNESCO, so we calculated the indicator by taking UNESCO data measuring the total expenditures on education as a percentage of GDP and multiplying it by the percentage of expenditures on primary education. 5. Girls Secondary Education Enrollment Girls Secondary Education Enrollment measures the female lower secondary education enrollment ratio based on data from UNESCO s Institute of Statistics. The ratio is calculated as the number of females enrolled in lower secondary divided by the total population of females that are the standard age of lower secondary enrollment. The MCC uses the Girls Secondary Education Enrollment indicator only for LMICs. (MCC 2012, 27-28). However, due to lack of data availability, we do not include this indicator in our analysis. 52

65 6. Child Health Child Health is based on data from the Center for International Earth Science Information Network and calculates child health as the average of three equally weighted indicators that improve child health. The first, Access to Improved Sanitation, is measured as the percentage of the population with access to adequate sanitation. The second, Access to Improved Water, is measured as the percentage of the population with access to improved water supply. The third, Child Mortality, is measured as the probability of children dying between the ages of 1 and 5 years (MCC 2012, 28-29). However, due to lack of available data, we do not include this indicator in our analysis. 7. Natural Resource Protection The Natural Resource Protection indicator is calculated as the percentage of biome area protected within a country s land area based on data from the Center for International Earth Science Information. The indicator first appeared on MCC scorecards in 2006 as an index on a scale of 0 to 100, with 0 being no protection and 100 being fully protected. The number is weighted by the share of the biome s area and then multiplied by 10 (MCC 2012, 29). However, due to lack of data, we do not include this indicator in our analysis. C. Ruling Justly Indicators The Ruling Justly indicators assess a country s level of democratic governance (MCC 2012, 13). We draw on six out of seven of these indicators as described below in our analysis. 1. Civil Liberties Civil Liberties measures a variety of factors relating to personal freedoms. The data come from the Freedom House in the form of an index. Country evaluation is on a scale of 0 to 60, where 0 represents the least amount of freedom and 60 represents the most amount of freedom (MCC 2012, 15-16). 2. Control of Corruption Control of Corruption measures the perception of corruption in government in both petty and grand forms, as well as a country s ability to combat corruption. The MCC considers corruption to impede the fight against poverty. As a result, the MCC requires that countries indicators must meet the minimum requirements, regardless of their overall scores on other indicators, to be considered eligible to enter into a poverty-reduction compact (MCC 2012, 16-18). The data come in the form of an index score compiled by the World Bank and range in score from -2.5 to +2.5, where -2.5 represents weak control of corruption and +2.5 represents strong control of corruption (World Bank Group 2013h). 3. Government Effectiveness Government Effectiveness largely measures the quality of public service, civil service, and policy formation (MCC 2012, 18-19; World Bank Group 2013d). The data come in the form of an index score compiled by the World Bank and range in 53

66 score from -2.5 to +2.5, where -2.5 represents a government that has low effectiveness and +2.5 represents a government that has high effectiveness (World Bank Group 2013h). 4. Political Rights Political Rights measures a variety of factors relating to the political system. The Political Rights indicator comes in the form of an index compiled by Freedom House and is based on 10 questions from three subcategories: Electoral Process, Political Pluralism and Participation, and Functioning of Government. Points are awarded to each question on a scale of 0 to 4, where 0 points represents the fewest rights and 4 represents the most rights. It includes two additional questions related to political rights, for which 1 to 4 points are subtracted depending on the severity of the situation. The overall scale is 0 to 40, with 0 being least free and 40 being most free (MCC 2012, 13-14; Freedom House 2013b). 5. Rule of Law Rule of Law measures the perception of the legitimacy of national law by its citizens, focusing on the justice system (MCC 2012, 20-21; World Bank Group 2013f). The data come in the form of an index score compiled by the World Bank and range in score from -2.5 to +2.5, where -2.5 represents an illegitimate perception and +2.5 represents a legitimate perception (World Bank Group 2013h). 6. Voice and Accountability Voice and Accountability measures the extent to which a country s citizens are able to participate in selecting their government, as well as freedom of expression, freedom of association, and media independence (World Bank Group 2013g). The data come in the form of an index score compiled by the World Bank and range in score from -2.5 to +2.5, where -2.5 represents the weakest and +2.5 represents the strongest respectively in this policy area (World Bank Group2013h). The MCC dropped this indicator in 2013 and added a new indicator, Freedom of Information. 7. Freedom of Information We do not use the Freedom of Information indicator in our analysis due to data limitations. Instead, we substitute the older Voice and Accountability indicator. 54

67 Appendix B: Indicator Information and Rates of Reform Statistics Table B1 presents initial data for the 20 indicators that we use in our analysis, along with summary statistics for rates of reform values. Category Economic Freedom Indicator Credit Depth Legal Rights Cost to Start Business Time to Start Business Fiscal Policy Inflation Time to Register Property Cost to Register Property Indicator Information Equivalent MCC Data Source Indicator Category Access to Credit Access to Credit Business Start-Up Business Start-Up Fiscal Policy Inflation Land Rights and Access Land Rights and Access Table B1: Indicator and Rates of Reform Information International Finance Corporation (2013) International Finance Corporation (2013) International Finance Corporation (2013) International Finance Corporation (2013) International Monetary Fund (2013) International Monetary Fund (2013) International Finance Corporation (2013) International Fund for Agricultural Development (2013) Years Available Score Bounds Direction of Improvement* Rates of Reform Summary Statistics Number of Observations Mean Standard Deviation [0,6] [0,10] [0, infinite] [0,infinite] [0,100] [-infinite, infinite] Min Max [0,infinite] [0,infinite] Regulatory Quality Regulatory Quality World Bank Group (2013h) [-2.5,2.5] Trade Freedom Trade Policy Heritage Foundation (2013) [0,100]

68 Category Investing in People Ruling Justly Indicator Girls Primary Education Health Expenditures Equivalent MCC Indicator Category Girls Primary Education Completion Rate Health Expenditures Immunization Rate Immunization Rates Education Expenditures Primary Education Expenditures Indicator Information Data Source Years Available Score Bounds Direction of Improvement* Rates of Reform Summary Statistics Number of Observations Mean Standard Deviation UNESCO (2013) [0,>100] World Health Organization (2013) World Health Organization (2013) UNESCO (2013) / Authors Calculations [0,100] [0,100] Civil Liberties Civil Liberties Freedom House (2013a) [0,60] Control of Corruption Government Effectiveness Control of Corruption Government Effectiveness World Bank Group(2013h) [-2.5,2.5] World Bank Group (2013h) [-2.5,2.5] Political Rights Political Rights Freedom House (2013a) [-4,40] Rule of Law Rule of Law World Bank Group (2013h) [-2.5,2.5] Voice and Voice and World Bank Group (2013a) [-2.5,+2.5] Accountability Accountability *Notes: In our analyses, we multiply the five negative direction of improvement by negative one (-1) for consistent interpretation: the greater the value, the better that country performs on the indicator. The Rates of Reform summary statistics reflect this change. Sources: As noted and authors calculations Min Max 56

69 Appendix C: Autocorrelation within Indicators Figures C1 to C12 show the autocorrelation for the 12 individual indicators for which data were available for all four time periods. The plot is the correlation of residuals from the rates of reform among of any two periods. The lower-left panels are scatterplots, the top-left to bottom-right diagonal panels are histograms, and the upper-right panels are the correlation coefficients with proportional font size. Figure C1. Fiscal Policy Figure C2. Inflation lmres lmres lmres lmres lmres lmres lmres lmres Figure C3. Regulatory Quality Figure C4. Trade Freedom lmres lmres lmres lmres lmres lmres lmres lmres

70 Figure C5. Girls Primary Education Figure C6. Health Expenditures lmres lmres lmres lmres lmres lmres lmres lmres Figure C7. Immunization Rate Figure C8. Education Expenditures lmres lmres lmres lmres lmres lmres lmres lmres Figure C9. Control of Corruption Figure C10. Government Effectiveness lmres lmres lmres lmres lmres lmres lmres lmres

71 Figure C11. Rule of Law Figure C12. Voice and Accountability lmres lmres lmres lmres.2006 lmres lmres lmres lmres Source: Authors calculations 59

72 Appendix D: Correlation between Indicators In the three figures below, we plot the correlation between indicators in similar categories. The upper-right half of the figure presents the correlation of indicator scores between each two indicators; the lower-left half presents the correlation of the rates of reform of the indicators. The type size is proportioned to the correlation if a correlation is legible, then it is high. As a rule of thumb, a regression between two variables whose correlation is higher than 0.3 would result in a very significant linear regression coefficient. Figure D1 depicts the correlation between each of six Economic Freedom indicators; the other four Economic Freedom indicators are excluded due to missing values. Figure D2 depicts the correlation between each of the four Investing in People indicators. Figure D3 depicts the correlation between each of the six Ruling Justly indicators. Figure D1: Correlation between Economic Freedom Indicators Cost to Start Business Time to Start Business Fiscal Policy Inflation Time to Register Property Regulation Source: Authors calculations 60

73 Figure D2: Correlation between Investing in People Indicators Girl's Primary Education Health Expenditures Immunization Rate Education Expenditures Source: Authors calculations Figure D3: Correlation between Ruling Justly Indicators Civil Liberties Control of Corruption Government Effectiveness Political Rights Rule of Law Voice and Accountability Source: Authors calculations 61

74 Appendix E: Average Number of MCC Indicators Passed The figures below depict the trends in the number of MCC indicator that individual countries have passed (met minimum MCC requirements) over time. The thick solid black line at the top of the figures indicates the maximum possible number of MCC indicators a country could pass. The thin solid black line in the middle is the average over number of indicators passed by all the countries. The other lines represent individual countries. Figure E1 depicts the results for the 69 LICs. Figure E2 depicts the results for the 29 LMICs. Figure E1: Average Number of MCC Indicators Passed by LICs 62