Monday, April 28, 2008

4. Research Design

Theory
Based on the findings of Burnside and Dollar (BD) in 2000, and specified by the case-study evidence of countries presently receiving AIDS-aid, I theorize that AIDS-aid will be effective only in countries with “good” policy environments. In the work of three main donors from 1998 – 2003, AIDS-aid has affected the HIV rates in both positive and negative directions. Figure 1 illustrates that while a cluster of decreased rates may be attributed to small amounts of aid, a seemingly equal amount of increases could also be attributed to the provision of aid as well. AIDS-aid appears sufficient but not necessary for a decrease in HIV by the recipient nation.


I adapt the BD model of the effect of aid on growth to suggest a decrease in HIV prevalence is possible but only when the recipient nation upholds a certain institutional and political environment. I modify BD’s Policy Index (P’) and create an interaction term of aid times policy (A*P’) relevant to the AIDS pandemic, to confront the often-overlooked endogeneity issue: does aid create good policy (and ultimately, decreased HIV rates) or does good policy attract donors to give? The interaction term makes it possible to consider the influence of the Policy Index (P’) when measuring aid (A) efficacy on HIV rates (Y):

Δ Y it = a + b (A it × P’ it) + c (A it)+ d (P’ it) +ε it

The effect of aid on the change in HIV rates therefore depends on the Policy Index:

δY it / δ A it = b P’ it + ε it

Although BD do not test for a selection bias in who receives aid, I conduct a treatment effects test because of abundant research warning that aid efficacy is based on both the donor and recipient behavior. In addition to the theories of Mascarenhas and BDM, Bourguignon and Sundberg (2007) emphasize that donor behavior influences national policymakers and Smith (2005) reveals the inadequacies of international organizations like the World Bank, including their failure to institute political reform as a precedent to developmental aid. Milner (2005) echoes these concerns and outlines the realistic costs inflicted on recipient nations involved in funding from international institutions. Their criticisms – for example the use of inputs and not outputs as measurement tools, giving further insight into the logic of corruption (Smith) - suggest that countries may be better off or may have remained at the same level of development without the foreign aid from these organizations.
Much of my theory supports Easterly’s (2002) contention that top-down giving is inherently deficient, both because it prioritizes low-return items like ‘glossy reports’ over legitimate evaluations, and also because it burdens scarce administrative skills of aid-receiving nations. Even more specifically, Bueno de Mesquita and Root (2002) suggest that earmarking money– in this case, for AIDS - causes the problem of recipient nations having no incentive to eliminate AIDS, because if money will consistently be given for AIDS relief, it can consistently be stolen.
The Treatment Effects model to check for selection bias in the distribution of aid legitimates the use of a Policy Index in the performance equation, for if aid is indeed given according to need (to countries with highest HIV prevalence rates) the hypothesis that a Policy Index could determine aid efficacy becomes null due to a selection bias.
Lastly, this study is built to cover the issue of AIDS financing broadly, unlike the more specific efforts of groups like The Center for Global Development (CDG), a team of development experts including Birdsall, Levine, Roodman, Easterly and Radelet who collect research on development, including effectiveness of aid. Their new “HIV/AIDS Monitor” research often features issue- and country-specific case studies on AIDS-aid efficacy, but Bourguignon and Sundberg (2007) warn that such close up examination can make it difficult to establish a counterfactual to findings and leave unanswered major debates over whether or not AIDS-aid in general has any effect on systematically and predictably decreasing HIV rates. Therefore, this study includes a broader base of all countries who have recorded HIV rates by UNAIDS, and who have received aid from one of the three main sources, considering relevant control variables and the possibility of a selection bias.

Core Hypothesis and Methodology
As HIV/AIDS-aid increases, I expect HIV rates to decrease but only under specific circumstances of good institutional and policy environments. I use OLS panel regressions to relate HIV rate as a dependent variable to aid, institutional and policy variables by country-period units.
The model will test the impact of aid (A), a calculated Policy Index (P’), an interaction term of aid and policy (AxP’) and an index of independent control variables (j(x)) on HIV rates in specific country (i)-time period (t) units:
ΔY it = a + b (A it × P it) + c (A it) + d (P it) + Σ e(j(x) it) +ε it

I separate the AIDS-aid by donor and run the same regression, measuring World Bank and Global Fund contributions separately, as the two donors differ in their methods of distribution and separation of the aid sources provide insight into what drives current trends of giving.

World Bank: ΔY it = a + b (A (WB) it × P it) + c (A (WB) it) + d (P it) + Σ e( j(x) it) +ε it
Global Fund: ΔY it = a + b (A (GF) it × P it) + c (A (GF) it) + d (P it) + Σ e( j(x) it) +ε it

I hypothesize that b > 0, or that an increase in AIDS-aid in countries with specific policy conditions results in decreased rates of HIV. A reminder that Policy Index is always a negative value, so a decrease in HIV associated with more AIDS-Aid in the Interaction term requires b to be positive. My null hypothesis states that AIDS-aid has no impact on HIV rates, even in countries with specific policy conditions, therefore b=0 and c=0. For the two source-specific models the null and alternative hypotheses remain the same.
The Treatment Effects model measures the binary variable of aid giving as a driver of a selection bias, and is more appropriate for this study than the traditional Heckman selection model, which involves missing dependent variables. In the Treatment Effects model, the continuous variable AIDS-aid from the OLS regression becomes an endogenously chosen binary term, so I expand my performance equation dataset to 500 observations, including all nations that have reported national surveillance data by the UNAIDS, not just aid recipients, from 1999-2005. The effect of aid of AIDS-aid on HIV rates becomes conditional on receiving AIDS-aid, and therefore requires a selection equation and an outcome equation. The selection equation uses one set of independent variables that determine how aid is distributed, unique from the variables that determined the change in HIV rate. The outcome equation then predicts the effect of having undergone ‘treatment’, or having received aid, on the change in HIV rate considering the effect of independent variables from the previous performance equation. I hypothesize that there is a selection bias in determining who receives AIDS-aid. The null hypothesis I attempt to reject is that the error terms in the selection and outcome equations are uncorrelated, or that there is no selection bias in the giving of AIDS-aid.

Data Description
Restricted by the availability of HIV prevalence data, the performance equation study includes 103 countries that experienced recorded changes in HIV rate during three periods: (1) 1999-2001, (2) 2001-2003, and (3) 2003-2005. Units of analysis are country-periods that measure how much the aid received at the beginning of a 2-year period affected the change in HIV rate over those two years. A decrease in HIV rate suggests effective aid while an increase in HIV rate suggests ineffective aid.
I deliberately extracted national HIV prevalence rates from four separate sources, the UNAIDS 2006 Report on the Global AIDS Epidemic (2005 rate), UNAIDS 2004 Report on the Global AIDS Epidemic (2003 rate), Report on the Global HIV/AIDS Epidemic 2002 (2001 rate), and the Common Database of the UN Statistics Division (1999 rate). New estimates are annually reported and figures for previous years are revised, however I use rates from the year in which they were published to control if the decision to give aid was based at all on the recipient nation’s HIV rate. All independent control variables are matched to the three periods of AIDS-aid.

Bearce and Tyrone (2007) emphasize that foreign aid with a lag is considered a more effective measurement of aid efficacy, so AIDS-aid is measured by the amount a nation receives over a 2-year period, lagged behind periods of HIV change one-period to account for the time it takes for aid to be spent on programs and services. Periods of aid correspond with periods of HIV change as (1) 1998-1999, (2) 2000-2001, and (3) 2002-2003. PEPFAR funding is included only in the selection equation because it begins in 2004 and HIV rates for 2007 are yet unpublished, so there are no appropriately lagged periods against which to match the impact of PEPFAR funds in the performance equation. Because PEPFAR funds were excluded I have distinguished the 15 focus countries with a dummy variable.
Three variables are used to construct the Policy Index, and are adopted largely from academic theory and policy recommendations on the efficacy of aid. Corruption is considered an immense obstacle to effective use of foreign aid, and Kurtz and Shrank (2005) emphasize that institutional quality may be even more important to effective healthcare provision than expenditure variables. Smith (2006) suggests that skimming funds off foreign aid for personal use is a rational decision for leaders of countries with a small coalition of people who determine their political survival, because foreign aid is ‘unearned revenue’ that will not cause the same economic damage that results from suppression of public goods. In Ethiopia, Liberia and Malawi, AIDS-aid has more than doubled the national health budget in recent years, creating immense pressure to spend large amounts at the risk of not receiving more, and therefore “recipient nations will find ways to absorb the funds, whether legally or illegally.” In Chad, both the World Bank and the Global Fund suspended their AIDS specific funding in 2006 over concerns that the government was managing funds with a deliberate lack of transparency. In Kenya, 1.4 million people were in need of ARVs in 2004, and it is estimated that the amount of foreign funds donated to the National AIDS and STD Control Programme were enough to provide anti-retrovirals to 200,000. By Nov. 2004, only 24,000 Kenyans were reported to be on ARVs. Lastly in Malawi, a recent report revealed healthcare workers demanding sexual, monetary and material favors in return for medical treatment, giving no care or sub-standard care to those who refused. Control of Corruption is one of six dimensions of governance measured by the World Bank and I have adapted the -2.5 to 2.5 scale to a 0.5 to 5.5 scale. Higher scores indicate greater control of corruption so I expect higher scores to correlate with greater decreases in HIV.
The second Policy Index variable measures the state of national health systems. Health systems – including service delivery, health workforce, physical infrastructure, insurance systems, and regulation and licensing of pharmaceuticals – are necessary for AIDS-aid to have any effect. Health expenditure can measure the government’s intended commitment to improving public health systems, and indicates if there is sufficient capacity to support a national public health approach to the pandemic. I measure infrastructure capacity as Public Health Expenditure as a Percent of GDP (Health Exp. % GDP), using World Health Organization data from years 2000, 2002, and 2004. I expect that higher percentages of public health expenditure will correlate with larger decreases in HIV.
If corruption is controlled and health is a priority, the programs that address AIDS still must be strategically devised to address those who are both most at risk and most able to decrease transmission: women. In a wider study on public spending of health, Pritchett and Filmer (1999) find that child mortality has far less to do with money spent, and more to do with factors like level of female education. Women in sub-Saharan Africa make up 61% of HIV infections and Dr. Farmer et al (1996) suggest that this discrepancy goes beyond the fact that women are anatomically twice as susceptible to infection than men through heterosexual intercourse. They suggest that women are endangered due to the very political powerlessness embedded within institutions worldwide, and are necessarily at greater risk for greater HIV infection when left without a voice. A measure of Women’s Political Rights comes from the Cingranelli-Richards (CIRI) Human Rights Indicators Database, and includes suffrage, the right to run for and hold political office, and to petition government officials. Scores estimate women’s political rights as not legally guaranteed (1), legally guaranteed but severely prohibited in practice (2), legally guaranteed and moderately prohibited (3), or legally and practically guaranteed (4). I expect a negative relationship in which more rights for women leads to decrease in HIV.
The Policy Index is constructed by weighting each of the three variables according to their coefficient in an initial regression of independent variables (excluding aid) on the change in HIV Rate (Table 5, Regression 3):

Policy Index = -.253(Control of Corruption) - 2.71(Health Exp. As % of GDP) - .166(Women’s Pol. Rights)

The remaining independent variables control for conditions that could decrease the HIV rate other than AIDS-aid. Women’s Economics Rights measures women’s accessibility into the labor force, including equal pay, right to gain employment without male relative consent, and job security. Because economic rights also include rights to work at night, to work in occupations classified as dangerous, and to work in the military and police force, the variable measure says less about freedom from and more about exposure to sexual harassment or violation, or opportunities for HIV to be transmitted. The scale ranges from economic rights for women (0), to nearly all economic rights are guaranteed and enforced by law (4). I also measure the condition of national HIV Surveillance Systems constructed from US Census Bureau’s HIV/AIDS Surveillance Database, the Euro HIV Database, and countries’ national HIV surveillance reports made available to WHO/UNAIDS. The quality of systems is scored according to the level of surveillance implementation: poor (1), partial (2), or full (3), and I expect high surveillance scores to correlate with decreases in HIV.
Based on Bueno de Mesquita’s conclusion that leaders of small winning coalitions will give less public goods I have included Freedom of Speech. This value fits BDM and Downs’ (2005) definition of a ‘coordination good,’, or a good that leaders are especially weary of offering because its provision empowers citizens in ways that increase potential for regime overthrow. This ranking measures freedoms of speech and press, from complete government censorship (0) to none (3). I expect that as freedoms increase, HIV rates will decrease, because the freedom is indicative of a government prepared to give public goods of even the riskiest sort to their populace.
I also include a measure of successful Directly Observed Therapy, Short Course (DOTS) Tuberculosis treatments. Because the DOTS treatment requires sponsorship and commitment from the national government, trained and dedicated medical personnel, access to functioning hospital or clinic and laboratory facilities, and availability of pharmaceutical drugs, and a standardized recording and reporting system, the scores indicate the quality of and access to health care, and gives insight into actual health outcomes. Considering that TB is the top cause of death for people with HIV/AIDS, I suggest that an increase in treatment success will equal an increase in HIV prevalence as more HIV positive people will live longer. GDP and GDP Per Capita are included to test contentions that more wealth creates longer personal time horizons and encourages safer behavior that will diminish the spread of HIV.
Multiplying the Policy Index by the amount of aid received creates the Interaction Term, and it is important to note that all values of Policy Index and of the Interaction Term are negative, where values farthest away from zero are considered the “good” policies. I therefore expect both coefficients to be significantly positive in order to support the hypothesis that aid will decrease HIV rates in countries with “good” policies.

However even in countries with incredibly strong controls on corruption, women’s political rights, and commitment to public health, the behavior of donors may still overwhelm and determine the efficacy of aid. BDM et al and Mascarenhas influence the unique set of independent variables required for the selection equation. The variables consider a donor’s desire for policy concessions with terrorist nations, control over natural resources, ease of dealing with new regimes and with autocrats, and rewarding good behavior on human rights. I include the baseline HIV rate to capture the potential that donors might give altruistically.
Terror Rating is taken from the Political Terror Scale, which averages all available annual scores from Amnesty International and State Department reports on a 1-5 scale from lowest to highest human insecurity. Energy Production is the Energy Information Administration measure of World Primary Energy Production in quadrillions of British Thermal Units. Regime Type, taken from the Polity IV dataset, uses the Polity score of difference between democracy and autocracy scores, with -10 representing a strong autocratic regime and 10 representing a strong democratic regime. Regime Stability, also from the Polity IV dataset, measures the number of years since the most recent regime change. Oil production is a measure of national annual oil production, and the Physical Humans Rights score represents an additive index of measures of instances of torture, extrajudicial killing, political imprisonment, and disappearance, ranging from no government respect for freedom from these punishments (1) to full government respect of rights (9).

No comments: