Democracy or Accountability? Governance and Social Spending in Africa

Abstract

In recent years, democracy has often served as shorthand for good governance when considering what facilitates development-friendly public expenditure. While recognising the sufficiency of democracy, we argue that it is accountability, achievable outside full democracy, that is the necessary component of governance. However, vague conceptualisations of accountability as ‘responsiveness’ or ‘answerability’ have prevented empirical work from exploring the relationship between accountability and public spending. In this paper we develop an understanding of accountability as the interaction between opposition, transparency, and enforcement and test its impact on social spending in Africa in both the presence and absence of electoral institutions.

Acknowledgements

The authors thank participants of the College of William & Mary Senior Seminar in Challenges to Development and of the School Political and International Relations Seminar Series, and four anonymous reviewers for comments on the manuscript. All remaining errors and omissions are the authors’ own.

Notes

1. In addition to defining accountability, Rubenstein (2007, pp. 616–632) enumerates the normative benefits of accountability within a state. These benefits include its facilitation of the constitution of non-domination, the increase in rule following by power wielder, the promotion of the accountability holder’s preferences, promotion of valuable substantive and procedural norms, the promotion of civic virtues and self-development as part of both power wielders and accountability wielders, and the provision of useful information to accountability holders. These benefits highlight the importance of accountability mechanism within a state.

2. For example, Diamond (1994, p. 7) highlights the role of civil society as ‘a vital instrument for containing the power of democratic governments, checking their potential abuses and violations of the law, and subjecting them to public scrutiny’.

3. Horizontal accountability can take various forms. Traditional forms of horizontal accountability include checks and balances in institutional settings, characterised by a separation of powers. Other examples include ‘human rights ombudsman, corruption control agencies, legislative investigative committees, and administrative courts’ (Ackerman, 2004, p. 443). Przeworski et al. (1999, p. 50) also advocate the use of accountability agencies as a form of horizontal accountability.

4. Indeed, with respect to Botswana, they argue that Botswana’s success in managing public expenditure cannot be attributed solely to electoral conditions. Instead, they conclude that this success may be attributed to ‘the influence of a deeper culture of societal responsiveness and prudence among politicians’ (Healey & Tordoff, 1995, p. 249).

5. We would note that we would not expect an alternative hypothesis of more accountability decreasing social spending, but we do think that the null hypothesis of no relationship between accountability and social spending is a plausible outcome.

6. Stasavage (2005) uses expenditure as a percentage of GDP. However, both he and Nooruddin and Simmons (2006) also use expenditure as a share of total government expenditure as alternative measures. We do the same, finding no substantive differences.

7. For ease of interpretation we inverted the freedom house metric, such that a score of 100 is ‘free press’ and 0 is ‘not free press’.

8. We would note that the while the latent judicial independence (LJI) measure does incorporate information from the Polity variables that we otherwise use in our model, the LJI measure also incorporates number of other indicators and combines them through a Bayesian random walk prior process that uses a logistic link function, rather than a simple multiplicative process. Accordingly, we think the Polity measure is sufficiently transformed to be simultaneously included as right-hand side variables (Linzer & Staton, 2011). However, we also omit the polity variable from our models in Online Appendix Tables AB4 and AB5.

9. We use the ‘Freedom of Assembly and Association’ variable from the CIRI database to proxy for civil society. This metric takes on values of 0, 1, or 2, and measure the extent of de facto government restrictions or limitations on assembly and association. We recognise that this is a measure of civil society potential rather than activity, but as we would expect those to be highly correlated we fell it is an appropriate proxy.

10. While we recognised the importance of federalism or decentralisation in opposition, we omit incorporating a metric to capture this variable due to the fact that our dependent variable is centralised health and education spending. As discussed at greater length in the conclusions, we see examination of accountability on decentralised government expenditure as an interesting avenue for future research.

11. We also run non-interactive models with each component and a mdoel with all three components included (but no interaction term). The results of these models (available in Online Appendix Table AB1) find that each component separately is positively associated with spending on health.

12. Over 16 years for our 53 countries, we have 779 observations of health spending compared to just 409 for education spending.

13. Lu et al. (2010) find negative relationships between official development assistance (ODA) and partner government health spending and no significant relationship between income levels and health spending.

14. And we further refine this by using aid disbursements, rather than commitments, as a more accurate ‘on the ground’ picture of foreign aid. However, the following results are all largely robust to using the commitment measure.

15. The latter, death by communicable disease, is only available for the year 2008; however, we would expect the differences between countries to remain relatively stable over time and as such we use the 2008 data as a proxy for all years in the dataset. Dropping this variable does not change the other substantive results and it of course drops from the fixed-effects model as a time invariant regressor. We also omit the number of physicians, as this measure is highly correlated with both under-five mortality and death by communicable disease, and the data coverage on the number of physicians measure is substantially less than on the other two metrics.

16. We also run a linear model using the polity measure and omitting the accountability components, finding a positive association between polity scores and public spending, replicating Stasavage’s (2005) results. However, this result, combined with the results from Model II, suggests that while democracy may be sufficient for public spending outcomes it is not necessary.

17. Including a panel correction for autocorrelation and country dummies. As a robustness check we also run FGLS fixed-effects specifications. These results are available in Online Appendix B.

18. We also use minimum, mean, and maximum values, finding no substantive differences.

19. Again, additional results and robustness checks, including FGLS fixed-effects specifications, are available in Online Appendix B.

20. Perhaps starting with Devarajan and Swaroop (2000).

21. Interestingly, when examining two-way interactions in Online Appendix B, we find, if anything, a substitution relationship. This suggests that two out of the three components will not suffice, and, in fact, may not provide any total impact on social spending larger than one of the elements would on its own.