Ruth Björklöv

Introducing women's political representation as an explanatory variable for aid utilization

An analysis of the influence of women's political representation on the utilization of foreign assistance

Ruth Björklöv

Thesis Work, 15.0 c, Fall 2020

Department of Government, Uppsala University Development Studies C

Supervisor: Joakim Kreutz Word count: 12 445

Abstract

This thesis investigates how women's political representation influences foreign aid

utilization. While recent contributions show that the impact of foreign aid is highly dependent on the recipient government, there is still limited research on the relationship between

women's political representation and aid utilization. Existing work within the research field of female political representation suggests that women are more likely to prioritize resource allocation towards healthcare and education and less likely to prioritize the military. Thus, women’s political representation is predicted to work as a moderating effect on aid

utilization, whereby increases in female representation is associated with more aid resources being allocated towards healthcare and education and less to the military. To test the

relationship(s) implied, this thesis employs multiple regression analysis on a time series data set of 102 aid-receiving countries from 2000-2017. The hypothesis that women's political representation has a moderating effect on aid utilization could not be supported by the regression analysis. The results do however indicate that female representation in the recipient countries influence government allocation in general.

Keywords: Women´s political representation, foreign aid, government expenditure, healthcare expenditure, education expenditure, military expenditure

Table of contents

1. Introduction 3

1.1 Main purpose and research question 5

1.2 Disposition 5

2. Previous research and theoretical framework 6

2.1 Aid utilization 6

2.2 Women's political representation and government spending 9

2.2.1 First set of hypotheses 11

2.3 Theoretical argument 12

2.3.1 Causal diagram 13

2.3.2 Second set of hypotheses 13

2.4 The contribution of this study 14

3.1 Operationalization 15

3.1.1 Dependent variable 15

3.1.2 Independent variables 16

3.1.3 Control variables 17

3.2 Specification of the regressions 18

3.3. Reliability and validity 19

4. Regression analysis and discussion 22

4.1 Descriptive statistics 22

4.2 Regression results 24

Table 2. Healthcare Expenditure 24

Table 3. Education Expenditure 25

4.3 Regression analysis 27

4.3.1 Healthcare expenditure 27

4.3.2 Education expenditure 28

4.3.3 Military expenditure 30

4.4 Discussion 31

5. Conclusion 35

6. References 37

1. Introduction

Questions concerning foreign aid and its effectiveness have been debated among scholars since the origin of the development policy field itself. Although researchers are ambiguous in determining the impact of aid, there is growing evidence in the literature that aid efficacy is highly dependent on the conditions of the recipient country. Contemporary scholarship often holds that the recipient countries’ policies and the quality of the local institutions will

determine whether or not aid actually has the intended consequences on the life of the citizens in the recipient country (Burnside and Dollar 2000; the World Bank 1998; Quibria 2014). In countries with low accountability of government officials and high levels of

corruption, the impact of foreign assistance is likely to be low or even negative (Perkins et al.

2012: 534). Thus, the recipient government becomes a crucial actor when determining the expected impacts of foreign aid.

Montinola and Prince (2018) suggest that there is a link between women’s political empowerment and the effectiveness of foreign aid on child development outcomes. Even though women in parliaments remain underrepresented in nearly all countries around the world, women's political representation has increased substantially in recent years. In January 2020, women held 25.2% of parliamentary seats worldwide, compared to 13.9% in 2000 (World bank 2020). Based on the assumption that women in elective office behave differently than their male counterparts, increases in women's political representation¹ is expected to affect the outcomes of collective decision-making. Accordingly, transformations in the gendered composition of the legislative bodies will presumably affect how foreign is used.

Thus, this thesis endeavors to examine if women's political representation influences aid utilization. To test the relationship implied, this study uses time series data analysis on 102 aid-receiving countries from 2000-2017. In this study aid utilization will be understood as the influence of foreign aid on government allocation, measured through healthcare, education and military expenditure².

1 Women's political representation and female representation will be used further in this thesis as the extent to which women is physically represented in the recipient regime, measured at the proportion of seats held by women in national parliaments. This operationalization will be discussed in 3.3.

2 The rationale behind this operationalization is explained further in 3.1.1.

Although research indicates that there might be many implicit reasons behind the allocation of foreign aid, the “official” claim of aid is to improve the life of the recipient citizens (Kaya and Kaya 2020). Given the global development agenda and the Sustainable Development Goals (SDGs)³, in which foreign aid is often aiming to promote healthcare and education, it is arguably of interest for donors (and recipients), to examine when aid resources are likely to be allocated to these sectors. The choice of spending sectors is further motivated by their relevance to women's political representation which will be further explained in 2.2.

To the authors' knowledge, this study contributes with the first cross-national study that investigates the influence of women's political representation on aid utilization⁴. This thesis has implications for two different areas of scholarship. Firstly, it contributes to the research on gender equality and female political representation. Existing research on the issue often employs data on OECD-countries or a combination of high and low-income countries, which makes an analysis on solely aid-receiving countries a valuable contribution. Secondly it builds on and contributes to the literature on aid utilization by introducing women as a moderating effect.

Two sets of hypotheses are tested, in which the first set predicts that female representation is related with greater spending on healthcare and education, and reduced spending on the military. The purpose of testing these first hypotheses is to examine if women's political representation is associated with certain spending patterns in the sample of aid receiving countries. The second set of hypotheses tests the main argument, which is how women's political representation influences how foreign aid is utilized. The regression results provide some support for the first set of hypotheses, as the estimations suggest that women's political representation is positively correlated with healthcare and education expenditure. However, according to the findings of this thesis, there seems to be no joint effect of women's political representation and foreign aid on government allocation. Thus, the second set of hypotheses, predicting that female parliamentarians work as a moderating effect on aid utilization is not

3 The list on the 17 SDGs: https://sdgs.un.org/goals

4 There are studies that investigates how women in donor countries affect aid allocation (Fuchs and Richert 2017) as well as studies on the relationship between foreign aid and gender equality (Bush 2011). Montinola and Prince (2018) examines an area closely related to this thesis; the relationship between women's empowerment and aid effectiveness. Focus has not yet been put on the relationship between women's political representation and aid utilization.

supported by the result. In addition, the estimated results imply that there is a negative relationship between foreign and healthcare expenditure, which indicate that foreign aid is used in a fungible manner⁵. The results on military expenditure are inconclusive. Due to several limitations, the results should be interpreted with caution and further research on the issue is necessary to determine the influence of women's political representation on aid utilization.

1.1 Main purpose and research question

This thesis endeavors to integrate the field of gender research with the literature on aid utilization. By drawing on existing work which suggests that the impact of foreign aid is conditional on the recipient government, this thesis introduces women's political

representation as a potential variable explaining aid utilization. Thus, the research question for this thesis is:

How does women's political representation influence foreign aid utilization?

1.2 Disposition

The rest of the thesis is organized as follows. Section 2 presents previous research and the theoretical framework on both aid utilization (2.1) and women's political representation and government spending (2.2). Subsequently, the theoretical argument of this thesis is

introduced (2.3) followed by the empirical and theoretical contribution of this study (2.4). In Section 3 the operationalization and research design are described and discussed. Section 4 provides a regression analysis and a discussion concerning the implications that can be made based on the discoveries. Lastly, in Section 5 there is a conclusion, followed by suggestions for future research.

5 A description of aid fungibility is provided in 2.1.

2. Previous research and theoretical framework

As stated, the influence of women's political representation on aid utilization will be

investigated through the expenditure on three government sectors: healthcare, education, and military. Accordingly, the association between foreign aid and government allocation needs to be examined, but also the relationship between women's political representation and public spending. In order to fully grasp the potential interaction between aid utilization and female political representation, these two issues will first be placed in their respective literature field.

The empirical and theoretical contribution of combining these two is then clarified. The first part of the section, 2.1, focuses on the relationship between foreign aid and government allocation, referred to as aid utilization. Due to the inconclusive findings on the issue, there are no hypotheses on the predicted correlation between foreign aid and government spending.

Section 2.2 provides crucial literature findings on the association between women's political representation and government expenditure. At the end of 2.2 the first set of hypotheses is introduced. Lastly, in 2.3, the theoretical argument of this paper is introduced together with the second set of hypotheses.

2.1 Aid utilization

As the primary purpose of foreign assistance has long been to stimulate economic

development, most policy and scholarly debate have concerned whether or not aid spur the economy of the recipient country. By investigating the influence of foreign aid on

government spending priorities, referred to as aid utilization, this thesis contrasts the

substantial body of research that examines the impact of aid on generic measures of economic growth. The first mentioned literature debate is less mature as compared to the research field that studies aid effectiveness in terms of economic growth. Although the aid effectiveness literature fails to provide any underlying consensus, the role of the recipient countries’

policies and institutions are commonly stressed (Burnside and Dollar 2000; the World Bank 1998; Quibria 2014).

As a large part of aid goes to the public sector (Marć 2016), the recipient government becomes a key-actor when understanding the influence of foreign aid. In a recent study by Tawiah and Karungi (2020) the authors investigate the impacts of differences in political orientation and utilization of aid in African countries. According to their findings, the political orientation of the recipient government has a significant impact on foreign aid

utilization. It appears that leftists’ political parties more often use aid to improve basic

education and public health to reduce poverty. Rightist political parties, on the other hand, are more likely to utilize aid by promoting the free-market and the private sector. The authors do however not examine aid utilization in terms of government expenditure on healthcare, education and the military.

Research on the relationship between aid and public spending to the chosen sectors is often situated within the literature on aid fungibility. Foreign aid is considered to be fungible when the recipient government reduces its own resources on the sector that was purposed by the donor and transfers them to other sectors (World Bank 1998: 61). Aid fungibility is the result of differences in target priorities between the recipient government and the aid donor, which results in that if a dollar of foreign aid is given with the purpose to promote a certain sector, less than a dollar might actually be allocated to that sector. Inflows of foreign aid then substitutes rather than supplements domestic spending. Others have raised doubts about aid being fungible at all. A contrasting perspective predicts that foreign aid will instead enhance domestic spending on the donor targeted sector. Since earmarked aid could contribute to a greater prioritization also by the recipient government, the domestic funding to that sector might increase as a response. The impact of one dollar of aid might instead be greater than a dollar. The latter concept is referred to as the flypaper effect, as the donor funding “sticks where it hits” (Farag et al. 2009).

The fungibility of aid is one of the most contested issues in the aid effectiveness debate.

Recent scholars such as Rana and Koch (2020) do however challenge the premise that aid fungibility is a bad thing. They argue that the negative perception of the concept has roots in a colonial or patronizing perspective, as it implies that the recipient government is less capable of allocating resources to where it is most needed. There is insufficient evidence that non-fungible aid would work better than fungible aid, and some findings even imply high levels of fungibility might be positive (Khilji and Zampelli 1994). When the intended aid funding is reallocated and employed in a manner in which it better meets the SDGs, aid fungibility might not necessarily be harmful from the donors´ point of view. At the same, aid fungibility can be associated with decreases in tax revenues which may reduce the

government accountability to the citizen (Fairbank 2014). It may also lead to increased

spending on other sectors, such as the military (Collier and Hoeffler 2007). From a donor’s view, these latter outcomes are arguably less favorable.

Without addressing whether or not aid fungibility is positive or negative, it is possible to assert that the empirical findings are rather ambiguous. Some results are in line with the flypaper effect (Acosta 2010), while others provide mixed evidence (van de Walle 2007;

Dieleman, Graves and Hanlon 2013) or claim that foreign aid more or less is used in a fungible manner (Chatterjee et al. 2007; Farag et al. 2009). It is reasonable to assume that these inconclusive discoveries partly are due to different employments of methodology and data. Some scholars examine the influence of foreign aid on the total amount of government expenditure (Marć 2016), while others study its influence on specific sectors such as health and education (Farag et al. 2009; Van de Sijpe 2013). Further, some researchers have completed case-studies on specific countries (Acosta 2010) while another group analyzes data on a particular region (Dewortoe and Chui 2020) or a large number of aid-receiving countries (Chatterjee et al. 2007: Farag et al. 2009).

As stated, healthcare and education are both highly prioritized by donors and achievements related to these areas are emphasized within the SDGs. Military, on the other hand, is arguably a sector that most donors would not like to enhance with aid funding. The fact that military support is excluded from the definition of Official development assistance (ODA)⁶, is an indicator that welfare targeted resources categorized as ODA are not intended to promote military spending. Further, many donors such as the EU include conditional clauses in their aid commitments where their funding could be frozen if human rights are abused (Saltnes, J. D 2017). It is therefore expected that donors put implicit pressure on the recipient states to reduce military spending, or at least not increase it following aid disbursements.

There is insufficient evidence on whether or not there is any correlation between foreign aid and military spending. A few empirical results suggest that aid may actually increase military spending (Collier and Hoeffler 2007; Khilji and Zampelli 1994). According to the findings of Kono and Monitola (2012), autocracies in particular are more likely to divert foreign aid to

6 ODA is short for Official Development Assistance and will be used synonymous with foreign aid since the variable is measured on data according to its definition. That operationalization and its advantages and disadvantages will be discussed in Section 3.

military expenditure. They support their findings with the selectorate model developed by Bueno de Mesquita et al. (2003). According to such theory, the selectorate (S) is the set of people who can participate in choosing a leader and the quantity of the people whose support the leader must retain to remain in office is the winning coalition (W). Since democratic regimes need support from at least half of the population, their W is bigger as compared to autocracies. The size of the W is expected to influence the composition of public goods and private benefits produced by the regime; whereby autocratic regimes will use foreign aid to restrict civil liberties (private benefits) rather than invest in welfare (public goods). Thus, autocratic regimes are more likely to use foreign aid to strengthen its military capacity. There is however an assumption made here that military spending is used to enhance the repressive capacity, which is debatable since the military sector also is meant for national security. In addition, as stated, there is a lack of empirical results that support any relationship between foreign aid and military expenditure.

To conclude, the aid effectiveness literature is inconclusive on whether or not aid fungibility occurs, and if its potential occurrence is harmful or not. It is however possible to assume that donors find it problematic when aid targeted healthcare and education are displaced to the military sector.

2.2 Women's political representation and government spending

Responding to the last decades significant increase in women's political representation, a large body of empirical research on women's substantive representation has emerged.

A strand of research has paid attention to the impact of women’s political representation in terms of priorities within public spending (Chattopadhyay et al. 2004; Clayton and Zätterberg 2019; Mavisakalyan 2014). Yet, these scholars often face difficulties in determining what policy areas are believed to be more or less in the interest of women.

Some researchers conceptualize gender-related policy simply as areas linked to issues that directly affect women's welfare, such as maternal health or domestic violence (Clayton, Josefsson, and Wang 2017; Schwindt-Bayer 2006). Other scholars examine policy areas that affect women more indirectly based on traditional gender norms, for instance, by studying policies related to child health and family assistance (Besley and Case 2003; Miller 2008).

Another, contextual, approach is first to recognize gendered differences in policy priorities

among the politicians and citizens, and then investigate how women's political representation relates to these preferences in public spending priorities (Clayton et al. 2018; Svaleryd 2009).

Contemporary scholarship often holds that there is a positive relationship between women's political representation and spending on healthcare (Funk and Philips 2018; Mavisakalyan 2013). Regarding education, a few scholars imply that men are more likely to invest in education (Chattopadhyay and Duflo 2004). Yet, most research suggests that spending on education is positively related to increases in women's political representation (Funk and Philips 2018; Mavisakalyan 2013; Svaleryd, 2009). Lastly, the literature strongly indicates that excessive spending on the military is related to male-dominated parliaments (Fay 2020;

Koch and Fulton 2011).

According to Clayton and Zetterberg (2018), women's political representation can contribute to both formal and informal changes in parliamentary institutions and accordingly influence the budgetary process in different phases. Increases in the number of women in the

parliament might influence voting outcomes in chamber wide legislative, budgetary processes and when budget proposals are reviewed within committees. Female legislators may also affect spending outcomes informally when negotiating and bargaining with other

representatives. If women prioritize other issues compared to their male counterparts, a greater share of women in the legislative assembly might cause changes in the aggregated priorities and thus in the political agenda. If a greater entering of women further has a positive impact on the ability of women to achieve higher and more influential political positions, the gendered impact on policy outcomes is expected to be even more significant (O’Brien and Rickne 2016). Further, according to Wängnerud (2009), the issue concerning the influence of women's presence within the political sphere is multi-layered. The women's presence might also affect how male representatives behave and what decisions they make, which she argues, could either modify gendered disparities or reinforce them.

Notwithstanding, when expecting women's political representation to impact policy outcomes, an implicit assumption is made. Women and men are believed to have different political preferences and they are expected to behave differently when elected. Such reasoning is rooted within feminist political theory in which women’s presence in political decision-making will contribute with different values and experiences (Phillips 1995: 65).

Women are in general, due to their traditional role as caregivers and subordinated roles in the

public sphere, more dependent on state-provided goods, it is reasonable to expect that women will prioritize resource allocation welfare-related issues. Women live in poverty at higher rates and more often qualify for and receive welfare benefits (Harvey 2011; Hildebrandt and Stevens 2009). Further, women more often work through the public sector with healthcare or educational services, and they are less likely to hold high-waged jobs within the private sector (Gangl and Ziefle 2009; Guy and Newman 2004). Thus, women are in general more dependent on the public sector both as a provider of goods and support but also as an employer. Accordingly, based on the assumption that women electives will make decisions based on their experiences, they are expected to prioritize resource allocation towards healthcare and education.

However, all female parliamentarians may not share the same experiences and priorities. For instance, women who have managed to obtain positions within governments presumably come from more privileged backgrounds than the average citizen woman (or average citizen).

Nonetheless, earlier findings indicate that female politicians tend to act according to other women´s interests (Bauer 2012). Thus, even though a female parliamentarian might not have personal experiences of being dependent on state support, or have worked within the public sector herself, she may still feel obligated to make decisions for other women in these positions.

If the proportions spent on healthcare and education increases following greater female representation, there must be decreases to other sectors. In addition to the empirical findings that strongly imply that women's political representation is associated with reduced military spending (Fay 2020; Koch and Fulton 2011), two further discoveries by Eichenberg and Stoll (2012) support such a negative correlation. According to the authors findings on individual citizen support for defense spending, women are on average less supportive of military spending and more adverse towards armed conflict. With the presumption that female parliamentarians have the same preferences as female citizens, alters in women's political representation is expected to decrease the military expenditure.

2.2.1 First set of hypotheses

When combining the predicted associations between female political representation and healthcare and education spending together with these findings, it is possible to expect that

the first-mentioned increases will be at the expense of the military. Building upon these empirical and theoretical findings, the first set of hypotheses is introduced:

H1a: All else equal, there is a positive correlation between women's political representation and government spending on healthcare.

H1b: All else equal, there is a positive correlation between women's political representation and government spending on education.

H1c: All else equal, there is a negative correlation between women's political representation and government spending on the military.

2.3 Theoretical argument

Foreign aid is an external source of income to the recipient government, which, like any other resource, is found to be used according to the political agenda of the regime (e.g. Tawiah and Karungi 2020). Even though foreign assistance sometimes is channeled through non-

governmental organizations, the greatest share is transferred from the donor to the recipient government (Marć 2016). Thus, the recipient regime become a key-actor when studying the impacts of foreign aid. Building on these insights, this paper investigates how the gender composition of the recipient governments influence aid utilization. If women hold a

substantial number of positions in the government, these are predicted to influence how aid resources are employed.

When elected, women are expected to behave according to their own experiences (Phillips 1995: 65) as well in the interest of other women (Bauer 2012; Clayton et al. 2018). Women tend to be more dependent on the public sector, both as a provider of support (Harvey 2011;

Hildebrandt and Stevens 2009) and as an employer (Gangl and Ziefle 2009; Guy and Newman 2004). Thus, female parliamentarians are predicted to prioritize funding to healthcare and education rather than non-welfare related targets. Further, women are in general more adverse to armed conflicts and military expenditure (Eichenberg and Stoll 2012). Based on the predicted formal and informal parliamentary influence of female electives (Clayton and Zetterberg 2018; Wägnerud 2009) increases in women's political representation will affect the budgetary process and thus government expenditure.

Theoretically, increases in female representation will then contribute to greater proportions spent on healthcare and education, and less resources allocated to the military sector.

Moreover, it is possible to assert that the predicted interests of female parliamentarians are similar to the targets of the SDGs and the global development agenda. Accordingly, the likelihood that aid resources purposed to promote the welfare of the recipient citizens are used in a fungible manner, is expected to decrease with greater female representation. The recipient government´s potential reduction of self-funded means to a certain sector in response to foreign aid, is predicted to be less substantial in the presence of women.

Therefore, women’s political representation is suggested to work as moderator in the

relationship between foreign aid (IV) and government allocation (DV), whereby increases in female representation is associated with more aid resources being allocated towards

healthcare and education and less to the military.

2.3.1 Causal diagram

Figure 1. The predicted moderating effect of women's political representation on the relationship between foreign aid and government allocation.

2.3.2 Second set of hypotheses

When deriving the second set of hypotheses, the expected influence of women's political representation on government allocation is combined with the empirical findings that support the influence of aid as dependent on the recipient government. Thus, the second set of

hypotheses is:

H2a: All else equal, women's political representation has a positive influence on the correlation between foreign aid and healthcare spending.

H2b: All else equal, women's political representation has a positive influence on the correlation between foreign aid and education spending.

H2c: All else equal, women's political representation has a negative influence on the correlation between foreign aid and military spending.

2.4 The contribution of this study

This thesis endeavors to incorporate the findings of the aid utilization literature with the discoveries of the of previous scholars within the women's political representation field.

Combining these two elements is arguably contributing to the existing research within both fields.

Firstly, this paper contributes to the aid literature by introducing women's political

representation as an explanatory variable for aid utilization. Although research has been done on the relationship between aid and women's political empowerment (Montinola and Price 2018), focus has not yet been put on women´s influence on the utilization foreign aid in terms of government allocation. Hence, recognizing the conditional nature of aid utilization, while addressing the last decades substantial increase in women's political representation, an important gap is bridged in the aid debate.

Secondly, it builds on the literature that examines the influence of women's political representation on public spending. Most research on the issue employs data on wealthier countries or a combination of wealthy countries and less developed countries (Clayton and Zätterberg 2018; Schwindt-Bayer 2006). There is still limited evidence on the relationship between women's political representation and policy outcomes in developing countries (Montinola and Prince 2018). It is therefore advantageous to use data on only ODA-recipient countries, as these estimations will reflect the influence of women's political representation in this particular group of countries. As stated, the actual ability for female legislators to

influence decision making may be highly limited. According to indicators on gender equality, such as the Gender Inequality Index (GII), gender inequality is on average greater in ODA- receiving countries as compared to wealthier countries (UNDP 2020). The influence of women's political representation may therefore differentiate between these groups of countries.

3. Operationalization and research design

As this thesis intends to examine generalizable patterns of covariation, a quantitative method is suitable. More specifically, six different Ordinary Least Square (OLS)⁷ regressions are applied on the time series data. The dependent variable (DV) in these regressions is

government spending priorities, where each spending sector (healthcare, education and the military) will be tested separately. The independent variables (IVs) are foreign aid and women's political representation. To further analyze if women's political representation has a moderating effect on the direction and/or strength of the relationship between foreign aid and government spending, an interaction variable is created: women's political

representation*foreign aid. The interaction variable allows the aid effect on government spending to depend on the level of women's political representation (Scott and Watson 2020:

298). The body of data contains information from a total of 102 ODA-recipient countries (see Appendix for the complete list of countries) from 2000-2017. Each country i, each year t, is assigned an individual observation number i,t. Thus, the total number of units of analysis is 102*18 = 1836. The sample is primarily determined by data accessibility on the main variables. The estimations are made in Stata 15.1.

3.1 Operationalization

This section begins with a presentation of the operationalization and data of the variables included in the panel data regression analysis. Subsequently, the different regression models that will be used are shortly explained, followed by a discussion concerning reliability and validity.

3.1.1 Dependent variable

For the purpose of this study, aid utilization is operationalized as government allocation in the aid-receiving countries. Based on the notion that foreign aid, regardless of its purpose, is an external source of income to the recipient country and thus increases the total amount of resources, it is likely to affect government priorities. By examining the correlation between foreign aid and government allocation, it is therefore to some extent, possible to estimate how foreign aid is utilized. Spending priorities are measured as yearly expenditures on healthcare,

7 The OLS-estimator chooses the coefficients in which the estimated regression line is as close as possible to its observed data. Closeness is measured by minimizing the sum of squares of the differences between the observed DV and those predicted by the linear function (Stock and Watson 2020: 148).

education and military as a share of the total government spending. As compared to studying the total amounts in US$, using data on the proportion allocated to each sector makes it easier to conduct comparative research between different countries and over time. In addition, budgetary allocations provide a comparable description of the trade-offs and priorities done by the politicians. The data for healthcare expenditure is retrieved at the World Health Organization, and the data for education expenditure and military expenditure is retrieved at the World Bank database.

3.1.2 Independent variables

Flows of foreign aid are measured as the official development assistance, ODA, that a country received in year t. ODA consists of government aid having the purpose to promote the

economic development and welfare of the countries that are listed as ODA recipients⁸ by the Development Assistance Committee (DAC). In this definition, loans and military support are not included⁹. ODA can be provided bilaterally, from one of the 29 DAC-members, or it can go through multilateral development agencies (e.g. the United Nations or the European Union). All data on ODA is measured in US$ million per capita. Dividing the amount of aid with the recipient population makes it possible to relativize flows of aid between countries.

Further, the data used for this study are disbursements and not commitments, as the latter provide an indication of future aid flows rather than the actual payments to the recipient country, which makes disbursements a more suitable measure when analyzing the

relationship between inflows of external funding and the spending priorities made that year.

Since ODA that is not specifically targeting a certain sector still may increase the proportion of resources allocated to that sector, the total amount of aid is used rather than sector-specific aid. It should be addressed that the ODA statistics suffer from several limitations, these will be further discussed in 4.3.

Women's political representation is measured as the proportion of seats held by women in national parliaments in a single or lower chamber. Its data is retrieved at the World Bank database. In model 5 and 6, which both includes an interaction variable, women's political representation is converted to a binary variable. The observations where women's political

8 see DAC List of ODA Recipients: https://oe.cd/dac-list

9 ODA does however include so-called soft loans (loans where the grant element is more than 25%) and provision of technical assistance.

representation is greater than the sample-median¹⁰ (12.9%) are coded with a 1, and the observations below this threshold receive a 0. The intention behind converting the variable to binary variable is to simplify the interpretation of the interaction variable, and the reason for not using the dichotomous variable in all six regressions is due to the loss of nuance that follows the binary categorization.

Figure 2. The sample variation in women's political representation.

3.1.3 Control variables

To mitigate the risk of omitted variable bias (OVB)¹¹, the regressions contain a number of control variables that previous scholars have suggested might affect the DV and at least one of the IVs, and thus influence the estimated relationship if not included (Stock and Watson, 2020: 231-233). Firstly, a dichotomous measure of whether the current ruling party can be characterized as leftist (1) or otherwise (0) is added to the regressions. Leftist parties are expected to 1) spend more on healthcare and education and 2) have greater female political representation (Erzeel and Celis 2016). Data on the political position is retrieved at Inter- American Development Bank (IADB). Secondly, two binary variables that measure if the country currently is regarded as democracy (Polity V score higher than 6) or an autocracy (score lower than -6) are added¹². Previous research suggests that female parliamentarians within nondemocratic states are less likely to influence decision making (e.g. Burnet 2008).

Moreover, democratization is predicted to influence the size and composition of the public

10 As illustrated in figure 2, there are some outliers that drive up the sample mean, thus the median is used as a threshold to divide the observations in two equally sized groups.

11 Regressions with a single regressor are vulnerable to OVB if an omitted variable Z is correlated with the IV and is a determinant of the DV. If Z is not controlled for, the OLS estimator of the causal effect between IV and DV is biased, as it will capture the effect of the IV and the effect of Z (Stock and Watson, 2020: 234).

12 Countries categorized as anocracies (receiving polity IV-values between -5 to 5) obtain the value 0 in both measures. Further, there was no polity-V data on the following countries: Belize, Eswatini, Timor-leste and Sudan. The annual democratic status of these countries has been collected at the Freedom House and coded according to their definition of whether the country is regarded to be democratic (free) or autocratic (not free).

sector (Lindert 2004). Thirdly, scholars have found a positive correlation between increases in income and government spending on healthcare and education (Glewwe and Jacoby 2004;

Newhouse 1977). Thus, a binary variable is included for countries that are currently categorized by the UN as a Low-Income country¹³. Fourthly, a binary conflict-variable is added, since the degree of military spending in particular, but also government spending in general, is presumably related to whether or not the recipient government is participating in an armed conflict. Further earlier scholars have also argued that ODA seems to be positively linked with occurrence of conflict (Al Daïaa, Van Der Molen and Nader 2014). The conflict data is retrieved at The Uppsala Conflict Data Program (UCDP). Finally, to account for country-specific traits that are time-invariant (e.g. political history and social attributes) that might affect women's political representation, received amount of foreign aid and government spending, country fixed effects are included in the last regressions applied on each spending sector.

3.2 Specification of the regressions

To examine the relationship between women's political representation and aid utilization, seven different regression models are applied on a time series data set, in which women's political representation and foreign aid are the IVs, and government spending priorities is the DV (each spending category will be tested separately). Model 1 and Model 2 are bivariate regression on the relationship between each government sector and ODA respectively

women´s political representation. In the third model, control variables are included, followed by fixed effects in the fourth. In Model 5 the interaction variable (Women's political

representation*ODA) is included. Model 6 include the interaction variable and fixed effects.

The last model is the same as Model 6 but with a two-year lag instead of one. The regression model when all variables are included can be formally written as:

Government allocation⁽𝑖,𝑡) = 𝛽1ODA ⁽𝑖,𝑡) + 𝛽2Women's political representation⁽𝑖,𝑡) + (1) 𝛽3Women's political representation* ODA (𝑖,𝑡) +𝛽4Control variables(𝑖,𝑡)+ αi + 𝜀(𝑖,𝑡)

13 Low-income countries are nations that have a per capita gross national income (GNI) of less than $1,026 (World Bank 2020).

In which i is entity (country), t is time (year), 𝛽1- 𝛽4 are regression coefficients, αi is an unobserved fixed effect¹⁴ that captures everything that is constant over time for the entity (country) and 𝜀(𝑖,𝑡) is an unobserved time-variant and individual error term¹⁵. 𝛽3 is the regression coefficient for the interaction variable (Women's political representation*ODA).

Lastly, addressing the possibility that variation in foreign aid and women's political

representation might not cause changes in public spending immediately, the IV variables are lagged by one year (t-1) in the first six models, and two years (t-2) in the last model. Lagged variables could also mitigate the of reversed causality, which will be further discussed in

3.3. Reliability and validity

This section intends to discuss the research design and operationalization in terms of

reliability and validity. Out of these two, issues related to the latter is supposedly the greatest concern for this study.

Beginning with reliability, this study strives for high reliability through transparency.

Meaning that, even though one might not agree on the choice of methodology and

operationalization applied in this thesis, one should still reach a similar result when reusing the research design. There is always a risk of unsystematic errors during the coding process of a new dataset. Yet, none of the data for the main variables of interest; government

spending, foreign aid and women's political representation, have been coded manually. These are retrieved and converted directly from the World Bank and OECD to Stata. The

categorization of the dichotomous control variables is mainly based on numeric thresholds and leave no room for subjective coding.

To determine the level of validity, most commonly understood as the extent to which a concept accurately measures what it is supposed to measure, the operationalization of the main variables requires further attention. As stated, aid utilization is measured as the correlation between government allocation and inflows of foreign aid. There are however a

14 Using fixed effects is a plausible method for panel data regressions when the OVB might vary across entities (countries in this case) but not over time. The fixed effects model has n different intercepts (one for each country) which can be understood as a set of binary variables that absorbs the influence of all OVB that differ between countries yet is time consistent (Stock and Watson 2020: 367).

15 Due to the risk that unobserved errors within a country might be correlated over time and thus not be i.i.d.

(independently and identically distributed), the regressions have country adjusted standard errors.

few issues with this operationalization. For instance, if it appears that a recipient government spent a greater share on healthcare over time, it does not show what sector they have

allocated less resources to. The proportion spent on healthcare is relative resources spent on all other sectors aggregated. As only three spending sectors are included in this analysis, the influence of women´s political representation on aid funding on all other areas is not

observed. In order to discover exactly how each dollar of foreign aid is utilized, more specific data on the recipient government´s incomes and expenditures is needed. Such a detailed procedure is arguably more suited for a case study. Thus, when interpreting the regression results, variation in the DV should be understood as an approximate indicator of variation in aid utilization.

Regarding the operationalization of foreign aid, using ODA as a proxy for aid has several limitations. To begin with, its data do not include funds provided by non-DAC governments such as India or wealthy Middle Eastern countries. Further, grants by private entities such as Bill & Melinda Gates Foundation are also excluded in the ODA definition (Piva and Todd 2009). According to some estimates, funds provided by private donors may be as much as one half or even two thirds of the official aid data (Perkins et al., 2009: 502-503). As a consequence, the impact of aid might be underestimated. Nevertheless, even though there might be validity concerns with using ODA as a proxy for aid, the ODA-statistics is by far the most comprehensive source of data on foreign aid flows and it is therefore commonly applied by scholars within the development field. Thus, the results of this study will be comparable to the findings of others who have used the same definition.

Furthermore, the validity of the chosen measurement of women's political representation as the proportion of seats held by women in national parliaments, is also open to critique. First and foremost, it builds on the assumption that female electives are able to participate in the decision making. There are several hindrances for female parliamentarians, limiting their chances to influence government decisions. The political arena is characterized by a deeply rooted culture of masculinity that causes gender bias in all levels of the political sphere (Lovenduski 2005: 48-52). Literature has found that female legislators do not experience the same access as men to those political positions and assignments that are associated with influence over agenda setting (Verge and Claveria 2018). The understanding of an “ideal politician” is not compatible with the image of a typical woman, and female politicians who

act according to such ideals are often accused of behaving inappropriately (Bjarnegård 2018).

Moreover, since pursuing a political career is time-consuming, it is often regarded as an unsuitable occupation for a woman whose main responsibilities lie within the household (Kenny and Verge 2016). To examine the extent to which women in different parliaments are actually able to influence budgetary outcomes, again, a case study would be a better option.

However, to enable large scale comparisons, which is the aim of this thesis, the proportion of seats held by women in national parliaments is arguably a valid enough indicator of women's political representation¹⁶.

16 There are other, more inclusive, measurements of women's political empowerment, such as the Women's Political Empowerment Index provided by V-dem. However, an index that aims to describe the overall conditions for women's political empowerment within a country is not directly linked with outcomes of decision-making within governments. Yet, as stated, the ability for a woman to influence decision-making within the government is presumably greatly dependent on her social and political empowerment.

4. Regression analysis and discussion

In this section, the descriptive statistics is presented, followed by the results from the bivariate and multivariate regression analysis. The results will then be interpreted in relation to the hypotheses. For clarity, each spending sector will be presented and discussed

separately in this section. Note that there are seven different regression models employed for each spending sector, making it a total of 21 equations.

4.1 Descriptive statistics Table 1. Descriptive statistics

The statistics in Table 1 reveals a great width in government expenditure priorities across the observations. Based on maximum and minimum values, primarily for healthcare spending, there seems to be outliers who spend far more on healthcare than the mean value. Further, as the minimum value is zero for military expenditure, there are countries that have not

allocated any government resources to the military. Since the variable women's political representation also has a minimum value of zero, it appears that there are recipient

governments who have had no female representation at all during the time period. The high maximum ODA value compared to its mean shows that there is great variety in how much aid the countries have received. Note that the negative minimum value on the ODA variable is caused by an analysis unit in which the country that year paid back more ODA then they received. The mean values for the binary control variables provide insights on the sample countries in terms of other relevant conditions. For instance, approximately 43 percent of the observations were categorized as democracies according to the Polity V index. It is also possible to assert that the variable with least data accessibility is education expenditure.

4.2 Regression results Table 2. Healthcare Expenditure

Notes: cluster adjusted standard errors on country level in parentheses.

Statistical significance is denoted: *** p<0.01, ** p<0.05, * p<0.1.

Low-income is omitted in 4, 6 and 7 due to collinearity.

Table 3. Education Expenditure

Notes: cluster adjusted standard errors on country level in parentheses.

Statistical significance is denoted: *** p<0.01, ** p<0.05, * p<0.1.

Low-income is omitted in 4, 6 and 7 due to collinearity.

Table 4. Military Expenditure

Notes: cluster adjusted standard errors on country level in parentheses.

Statistical significance is denoted: *** p<0.01, ** p<0.05, * p<0.1.

Low-income is omitted in 4, 6 and 7 due to collinearity.

4.3 Regression analysis 4.3.1 Healthcare expenditure

H1a: All else equal, there is a positive correlation between women's political representation and government spending on healthcare.

H2a: All else equal, women's political representation has a positive influence on the correlation between foreign aid and healthcare spending.

As presented in Table 2, the estimated regression results on healthcare spending imply that there is a small positive correlation between women's political representation and healthcare expenditure. Substantially, Equation 2, 3, and 4 suggests that a one percent increase in women's political representation is associated with approximately 0.03 percent more spending on healthcare. These results are similar to findings of Mavisakalyan (2014)¹⁷.

Equations 5-7 in which women's political representation is converted to a binary variable also indicate that governments with greater representation of women spend more on healthcare.

The coefficients in Equations 5 and 7 do however lack statistical significance and as

illustrated in Figure 3, in which the observations are scattered, the association appears to be rather weak. Based on these results, there is a small indication that H1a has support.

Figure 3. Healthcare expenditure on y-axis and women's Figure 4. Healthcare expenditure on y-axis political representation on x-axis. and ODA/ capita on x-axis.

17 Mavisakalyan (2014) assert that the OLS coefficient for female representation is 0.03 and when including fixed effects it is 0.01. His dataset includes a mixture of 120 low and high-income countries in the year 2000.

Moreover, the coefficient for the interacted variable (female representation*ODA) is not significant at any level. Thus, the results do not support H2a, as there is no indication that women's political representation would have a moderating effect on the relationship between ODA and healthcare spending. However, as stated above, the results do suggest that

healthcare expenditure is partly dependent on women's political representation and inflows of ODA, but the combination of the two does not appear to have any additional effect.

As for the relationship between foreign aid and healthcare expenditure. Since the ODA coefficient is negative and statistically significant at the 95 % confidence level throughout equation 2-7¹⁸, it appears that an increased inflow of ODA causes a small decrease in healthcare spending¹⁹. The negative association between the two variables is illustrated in Figure 4. These results are consistent with the presented theory as aid as fungible.

Lastly, the negative and statistically significant coefficients indicate the low-income countries spend a smaller proportion on healthcare. The coefficients in Equation 5 and 7 suggest that there is a negative correlation between healthcare expenditure and democracy and autocracy, which imply that anocracies (receiving polity V-values between -5 to 5) allocate funding to the healthcare sector to a greater extent than these two categories.

4.3.2 Education expenditure

H1b: All else equal, there is a positive correlation between women's political representation and government spending on education.

H2b: All else equal, women's political representation has a positive influence on the correlation between foreign aid and education spending.

According to the regression results, illustrated in Table 3, all five regressions testing the relationship between women's political representation and education expenditure imply, with statistically significance at the 95 % confidence level, that there is a positive correlation between these two variables. In the first regression in which women's political representation

18 The coefficient in Equation 3 reaches statistical significance at the 99 % confidence level.

19 Given that the mean spent on healthcare is approximately 8.7% (see Table 1), a decrease with 0.004 percentage units does not have a substantial impact on the total government spending on healthcare.

is a continuous variable, the coefficient value is on average 0.02, implying that a one percent increase in women's political representation alters the education expenditure with 0.02 percent unit. Although the number is small, the mean value for education spending (4.1 %) is less as compared to healthcare spending (8.7 %). Meaning, the estimated increases in

education spending are relatively greater. Hypothetically, if the percentage of women’s political representation would increase with the difference between its minimum and maximum value (63.75 percentage units), the spending on healthcare would increase by 30%²⁰. According to these estimates, hypothesis H1b is supported. The association between female representation and education expenditure is illustrated in Figure 5.

Figure 5: Education expenditure on y-axis and women's Figure 6: Education expenditure on y-axis and political representation on x-axis. ODA/ capita on x-axis.

As regards to the interaction variable, in the two last Equations (13 and 14) the coefficients are negatively signed, and they reach statistical significance at the 99 % confidence level.

This indicates that governments with great female representation are less likely to utilize aid resources through the education sector. However, since the coefficient does not reach statistical significance in one out of three equations, the null hypothesis, being that the coefficient value is in fact zero, cannot be rejected. At the same, these estimations do not in any way support the hypothesis that women´s political representation would have a positive moderating effect on the relationship between ODA and education expenditure. H2b is thus not supported by the results.

20 Calculation: (Coefficient value* (max-min value of women's political representation)) / mean spending on education = 0.02*63.5/4.1435 = 0.3 = 30 %

Similar to the ODA coefficient in Table 2, its effect appears to be negative, which is an indicator of aid fungibility, however these estimations are not statistically significant. Figure 6 provides an indication that the relationship between the variables is week.

Out of the control variables, the democracy coefficient is only one that reaches statistical significance throughout the equations. Its value indicates that democracies on average spend more on education.

4.3.3 Military expenditure

H1c: All else equal, there is a negative correlation between women's political representation and government spending on the military.

H2c: All else equal, women's political representation has a negative influence on the correlation between foreign aid and military spending.

In relation to these hypotheses, the results illustrated in Table 4 are rather ambiguous. The results from the two last regressions, in which women's political representation is a

dichotomous variable, indicate that there is a positive correlation between women's political representation and military spending. The estimated coefficients for women's political representation reaches statistical significance at the 99 % and 95 % confidence level in equations 20 and 21. According to estimated coefficients in these two regressions, the group of countries that have the highest level of women's political representation spend on average 0.2 percent units more on the military. Since the mean value for military expenditure is 2 % (see Table 1), an increase with 0.2 percent units increases the total military spending with 10

%. The regression line illustrated in Figure 7 does however not provide any clear indication that there is a correlation of any sign between female representation and military spending.

Hence, hypothesis H1c, which predicted that women's political representation would have a negative impact on military expenditure, is not supported. These findings on the relationship between female representation and military spending are not in line with result of Koch and Fulton (2011) who discovered a significant and negative correlation between the proportion of women in parliament and spending on the military.

Figure 7: Military expenditure on y-axis and women's Figure 8: Military expenditure on y-axis and political representation on x-axis. ODA/ capita on x-axis.

Moreover, as for the coefficient for the interaction term, it does not reach statistical

significance in any of the models. Hence, it does not appear as if female representation would have a moderating effect on the relationship between ODA and military spending, H2c is therefore not supported based on these estimations.

The estimations imply that ODA has a small negative effect on military spending, however, the coefficient is not significant throughout the equations. Figure 8 illustrates that the

regression slope is almost completely flat. These ambivalent results are in line with previous research on the issue. Regarding the control variables, the results suggest that low-income countries spend a smaller proportion on the military compared to countries with greater income, which invites to further investigation. As expected, there seems to be a positive relationship between participation in armed conflicts and military expenditure.

4.4 Discussion

The first part of the discussion intends to further analyze the regression results and to what extent these are generalizable. More specifically, the aim is to answer what this research can actually tell us about the influence of women's political representation on foreign aid

utilization. Subsequently, a discussion on potential explanations behind the regression results and its implications is provided.

To be able to establish if there is a causal inference in which women in parliaments effect the way in which aid is utilized, the four causal criteria²¹; causal mechanism, time order,

covariation and isolation proposed by Kellstedt and Whitten(2018: 60-63) will be reviewed relative to the research design of this thesis.

Regarding the causal mechanism, although a plausible explanation for the expected relationship is theoretically and empirically motivated, the predicted interaction between women's political representation and aid utilization has no support in the regression analysis.

Meaning, according to the estimated results, the joint effect of women's political

representation and foreign aid on government allocation is not statistically higher than the sum of both effects individually.

Notwithstanding, this does not necessarily mean that there is no interaction in practice. It is worth considering that women in parliament might have a moderating effect on aid

utilization, but this analysis did not find it. For instance, as stated, this study only investigates spending outcomes in three sectors. It is possible that women have been influencing aid utilization within other policy areas. Further, as regards aid fungibility, and the possibility that women's political representation would reduce the extent to which aid is used in a

fungible manner, there are several limitations for what conclusions that can be made based on the regression results. The fact that this study examines the total amount of ODA, and not aid targeted specifically to healthcare and education might have contributed to the results. The regression analysis could have been improved by utilizing disaggregated data on ODA, in which the influence of aid targeting healthcare and education on their respective spending sector was analyzed separately.

As regards the criterion of time order, applying regression analysis comes with an inherent issue. OLS models estimate whether or not there is a relationship between the variables and the strength of a potential correlation, they do not ensure that the variables do not affect each other simultaneously, or that the relationship in fact is reversed. Thus, the estimated

regression coefficients presented above, might be due to simultaneous or reversed causality.

For instance, the correlation between healthcare expenditure and ODA, might be due to

21 The authors call it “Four hurdles along the route to establishing causal relationships” (Kelldstedt and Whitten 2018: 60).

government spending priorities influencing inflows of foreign aid. At the same time, the proposed chain of order suggested in this thesis is arguably more believable. In addition, the usage of lagged effects could mitigate the issue of reverse causality²².

Further, in terms of covariation, running multiple regressions do provide an indicator of the covariation, namely, the coefficient and its degree of statistical significance. Yet, previous issues related to reliability and validity may impact the substantial meaning of these estimations.

Finally, isolation can partly be ensured, as potential confounders are controlled for. On the one hand, the usage of fixed effects could mitigate the risk of spuriosity to some extent as it absorbs OVB that differs between countries that is time consistent. On the other hand, the risk of OVB is still prevalent as there might be excluded and unobservable factors that could affect the estimated relationships. In addition, the multiple regression models applied in this thesis do not include time fixed effects. Trends that vary over time but are constant

across countries are therefore not controlled for. It is possible that regional or global trends such as pandemics or military tension have affected government allocation to the chosen sectors. In addition, the last decades´ changes in the global development agenda could have contributed to greater spending on healthcare and education. Meaning, the global increase in women´s political representation might have occurred in parallel with other driving factors of government allocation.

Due to previously mentioned concerns of reliability and validity, together with these

additional potential sources of bias, there are limitations in what can be asserted based on the regression analysis. In order to verify the causal mechanism of aid utilization and to

understand exactly how women's political representation affects aid utilization, a case study or process-tracing is required. At the same time, the quantitative method applied in this enabled more general comparisons between countries and over time.

22 The rationale here is that the received amount of foreign aid in year t-1 as well as women's political representation in t-1, could not be a result of the outcome of government spending in year t. There is however lack of empirical evidence supporting that lagged variables are efficient in responding to endogeneity concerns (Bellmare, Masaiki and Pepinsky 2017).

After these critical limitations have been addressed, a brief discussion of the results is still of interest. As the regression results imply that women's political representation is correlated with certain spending patterns, it appears that even though women in these parliaments generally have an impact on spending decisions, they do not influence aid utilization in particular. A potential explanation for this result could be that foreign aid to a great extent is earmarked, and that there is little room for the recipient government to influence its direction in general, regardless of the gender composition. At the same time, the results, particularly for healthcare expenditure, indicate that the recipient government decreases its spending in response to aid. Increased female representation was expected to mitigate such circumvention of resources, which is not supported by the results.

The modest extent of aid fungibility suggested by the results may however not be in contrast to the prioritization of female parliamentarians or the donors. As stated, the fact that aid is used in a fungible manner is only problematic if the resources instead are used for an end that is not consistent with the purpose of the donor. It is possible that a decrease in spending in healthcare or education releases resources to a sector/issue that is more crucial for the

recipient citizens´ welfare and perhaps even more consistent with the SDGs. Albeit, based on mentioned literature findings, women are more likely to prioritize healthcare, education and the welfare of women, which are all highly prioritized on the global development agenda. On that note, it would be more favorable if the results indicated that women do have a

moderating effect on aid utilization.

Lastly, it should not be forgotten that the regression results imply that women's political representation does influence the overall government allocation in the recipient countries.

Although donor funding sometimes is a sizable proportion of the recipient’s government expenditure, it is often only a small share (Perkins et al. 2012: 512-513). The explanatory factors behind the overall government allocation is thus more crucial to understand as compared to aid utilization. Further, since most empirical studies on the issue of women's political representation and government allocation has been done on OECD-countries or a combination of high and low-income countries, it is fruitful for the gender debate to discover to what extent women in low-income countries impact government spending outcomes. As addressed in 4.3.1, the regression results on the association between female representation and healthcare expenditure are similar to the discoveries by Mavisakalyan (2014), who