¹ Prominent examples include Barber, Benjamin, Strong Democracy: Participatory Politicsfor a New Age (Berkeley: University of California Press, 1984)Google Scholar; Corry, Dan and Stoker, Gerry, New Localism: Refashioning the Centre-Local Relationship (London: New Local Government Network, 2002)Google Scholar; and Ackerman, Bruce and Fishkin, James. S., “Deliberation Day,” Journal of Political Philosophy 10 (June 2002)CrossRefGoogle Scholar. Trends in developing areas are discussed in more detail below.

² As an analogy, a study showing systematic differences between majoritarian versus PR systems could establish that institutions matter even in the absence of a no-institution benchmark. The converse is not true, however. If there is no variation in outcomes associated with different types of institutions, this does not establish that institutions do not matter; for such a claim a no-institution benchmark would be needed. We return to this point below.

³ Henkel, Heiko and Stirrat, Roderick, “Participation as Spiritual Duty: Empowerment as Secular Subjection,” in Cooke, Bill and Kothari, Uma, eds., Participation: The New Tyranny (London: Zed Press, 2001), 168Google Scholar.

⁴ Chambers, Robert, “The Origins and Practice of Participatory Rural Appraisal,” World Development 22, no. 7 (1994), 1253CrossRefGoogle Scholar.

⁵ Richards, Paul, “Participatory Rural Appraisal: A Quick-and-Dirty Critique,” PLA Notes 24 (1995)Google Scholar.

⁶ International Monetary Fund, 2005 Review ofthe Poverty Reduction Strategy Approac—Balancing Accountabilities and Scaling Up Results—Synthesis, http://www.imf.org/external/np/pp/eng/2005/091905s.htm (accessed September 10, 2006).

⁷ Sen, Amartya, Development as Freedom (New York: Anchor Books, 1999), 291Google Scholar.

⁸ Habermas, Jürgen, Theory of Communicative Action, vol. 1, Reason and the Rationalization of Society (Boston: Beacon Press, 1984)Google Scholar; idem, “Three Normative Models of Democracy,” in Benhabib, Sheyla, ed., Democracy and Difference (Princeton: Princeton University Press, 1996)Google Scholar.

⁹ Stiglitz, Joseph E., “Participation and Development: Perspectives from the Comprehensive Development Paradigm,” Review of Development Economics 6, no. 2 (2002)CrossRefGoogle Scholar

¹⁰ Ibid., 168.

¹¹ Andrew Norton, “Approach Paper: Politics and the PRSP Approach,” Working Paper, PRSP Monitoring and Synthesis Project (London: Overseas Development Institute, 2002).

¹² Laure-Helene Piron and Alison Evans, “Politics and the PRSP Approach: Synthesis Paper,” ODI Working Paper 237 (London: Overseas Development Institute, 2004).

¹³ Ackerman and Fishkin (fn. 1)

¹⁴ Hicks, Darrin, “The Promise(s) of Deliberative Democracy,” Rhetoric and Public Affairs 5 (Summer 2002)CrossRefGoogle Scholar.

¹⁵ Gibbard, Allan, “Manipulation of Voting Schemes: A General Result,” Econometrica 4, no. 4 (1973)Google Scholar; Mark Satterthwaite, “Strategy-Proofness and Arrow's Conditions: Existence and Correspondence Theorems for Voting Procedures and Social Welfare Functions,” Journal of Economic Theory (April 1975).

¹⁶ List, Christian and Dryzek, John S., “Social Choice Theory and Deliberative Democracy: A Reconciliation,” British Journal of Political Science 33, no. 1 (2003)Google Scholar.

¹⁷ Hanson, Robert and Marks, Eli, “Influence of the Interviewer on the Accuracy of Survey Results,” Journal of the American Statistical Association 53 (September 1958)CrossRefGoogle Scholar; O'Muircheartaigh, Colm and Campanelli, Pamela, “The Relative Impact of Interviewer Effects and Sample Design Effects on Survey Precision,” Journal of the Royal Statistical Society Series A (Statistics in Society) 161, no. 1 (1998)CrossRefGoogle Scholar.

¹⁸ Plutarch, 75 c.E., Aristides http://classics.mit.edu/Plutarch/aristide.html (accessed September 10, 2006).

¹⁹ Bardhan, Pranab and Mookherjee, Dilip. “Capture and Governance at Local and National Levels,” American Economic Review, Papers and Proceedings 90 (May 2000)CrossRefGoogle Scholar; idem, “Pro-Poor Targeting and Accountability of Local Governments in West Bengal,” Journal of Development Economics 79 (April 2006)Google Scholar.

²⁰ Lukes, Steven, Power:A Radical View (New York: Macmillan, 1974)CrossRefGoogle Scholar; Bachrach, Peter and Baratz, Morton S., “The Two Faces of Power,” American Political Science Review 56 (December 1962)CrossRefGoogle Scholar.

²¹ Pelletier, David, Kraak, Vivika, McCullum, Christine, Uusitalo, Uula, and Rich, Robert, “The Shaping of Collective Values through Deliberative Democracy: An Empirical Study from New York's North Country,” Policy Sciences 32, no. 2 (1999)CrossRefGoogle Scholar.

²² Dryzek, John S. and Torgerson, Douglas, “Democracy and the Policy Sciences: A Progress Report,” Policy Sciences 26, no. 3 (1993)Google Scholar; Kapoor, Ilan, “The Devil's in the Theory: A Critical Assessment of Robert Chambers' Work on Participatory Development,” Third World Quarterly 23, no. 1 (2002)CrossRefGoogle Scholar.

²³ Elster, Jon, “Deliberation and Constitution-Making,” in Elster, Jon, ed., Deliberative Democracy (New York: Cambridge University Press, 1998)CrossRefGoogle Scholar.

²⁴ Mohan, Giles and Stokke, Kristian, “Participatory Development and Empowerment: The Dangers of Localism,” Third World Quarterly 21, no. 2 (2000)CrossRefGoogle Scholar.

²⁵ Douglas, Mary, Natural Symbols: Explorations in Cosmology (London: Barrie and Rockliff, 1970)Google Scholar.

²⁶ Murphy, William P., “Creating the Appearance of Consensus in Mende Political Discourse,” American Anthropologist 92, no. 1 (1990)CrossRefGoogle Scholar.

²⁷ Ibid.

²⁸ Gerhard Seibert, “Coup d'Etat in Sāo Tomé and Principe: Domestic Causes, the Role of Oil and Former ‘Buffalo’ Battalion Soldiers,” in African Security Analysis Programme, Occasional Paper, Institute for Security Studies (October 10, 2003), 2.

²⁹ Available on-line at http://www.earthinstitute.columbia.edu/cgsd/STP/documents/Questionnaire P_final_000.pdf.

³⁰ Mouffe, Chantal, “Deliberative Democracy or Agonistic Pluralism?” Social Research 66, no. 3 (1999)Google Scholar.

³¹ Fishkin, James S., Democracy and Deliberation:New Directionsfor Democratic Reform (New Haven: Yale University Press, 1991)Google Scholar.

³² For an argument against such structures, see Chambers (fn. 4), 1441.

³³ The Saotomean authorities chose to organize proportionately more forum meetings in the less-populated districts, to give equal access to all citizens. Our sampling method follows this approach and should therefore be seen as representative of the population as it was targeted by the forum meetings, rather than of the unweighted national population. There were two districts for which we could not sample respondents before the forum meetings, since the forum was already starting in those locations at the time the survey interviews were launched. The remaining districts, which we did sample, account for 86 percent of the voting population in the country. Th e forum meetings in the two missing districts (Caué and Cantagaló), moreover, were demographically similar to the forum-going population nationally.

³⁴ For more details about the survey and a more detailed analysis of the effect of forum participation on privately reported individual preferences, see Martin E. Sandbu, “Does Deliberation Engender Public-Spiritedness? A Study of Deliberative Democracy in Sao Tome and Principe's National Forum” (Manuscript, Wharton Business School, 2006).

³⁵ These two are the first session (the pilot) held in Santa Margarida and an extra final session (the makeup) held in Trindade.

³⁶ To illustrate the logic, consider a binary setting and assume that in the absence of manipulation by leaders a given group would select option 1 with probability q. Assume that share α of leaders in fact support option 1. Finally, assume that if a leader does not like the outcome that a group would otherwise select, he has the ability to exercise influence and change the outcome with probability β. This bias parameter, β, is not directly observable. Nonetheless we can show that the R ² statistic—that is, the share of variance in outcomes explained by the characteristics of leaders—provides rich information on β. In this case the expected R ² is given by

We can see from this expression that conditional upon some level of heterogeneity across leaders (α€(0, 1)), R²=0 if and only if β=0, and R²=1 if and only if β=1. In addition, for all values of q, α β, we can show that R ²≤β. Hence if R ² of the variation can be explained by leader characteristics, then at least share R ² of the time, leaders can enforce their will on the groups that they lead.

³⁷ Our “intention to treat” variable was determined by replacing our leader indicator (in cases where the leader for a meeting did not correspond to one of the assigned leaders for a site) with a leader randomly selected from the pool of leaders that ought to have gone (but did not go) to a particular meeting. In some cases there were no such individuals. This can arise, for example, if due to large turnout more meetings are held at a given venue than originally planned. In such cases in which a leader is moved from one venue to another to lead an unanticipated meeting, the leader is not in fact replacing any other leader. Rather, there is an unanticipated observation in the data.

³⁸ If we add a fixed effect for each location, we find that using an F test on the leader dummies we continue to be able to reject the null of no leader effect in nine of twelve cases at the 90 percent level and in seven cases at the 95 percent level. By construction, location variables are independent of leader dummies. Moderador dummies are not, however; the introduction of these moderador dummies is discussed below.

³⁹ The weights placed on leaders' preferences to generate these averages are determined by their relative frequency in leading discussions.

⁴⁰ The tests on the effects of age and gender effectively assume that the observations we examine are all independent. However insofar as the responses of meetings run by a given discussion leader are correlated, this assumption is inappropriate. It is not possible to include fixed effects in our examination of gender and age effects, since these effects are themselvesfixed.In order then to isolate these effects, we also implemented a test that focuses narrowly on the effects of gender and age characteristics independent of individual fixed effects by taking the average response to a question over the set of meetings led by a given leader and regressing average responses on the gender and age of the leader. This reduces the number of observations to between thirty and forty for each regression and so provides a difficult environment to identify these effects. We find that for two issues gender effects remain strong and significant at conventional levels and for another two age effects remain strong, accounting for between 10 and 20 percent of the observed variation. These results are available on request from the authors.

⁴¹ Chattopadhyay, Raghabendra and Duflo, Esther, “Women as Policy Makers: Evidence from an India-Wide Randomized Policy Experiment,” Econometrica 72 (September 2004)CrossRefGoogle Scholar.

⁴² Although, we note, there is no relationship between the number of older people attending the meetings and responses on this issue.

⁴³ A counterargument is the following: plausibly individuals are more likely to change their opinion on some issues than on other issues. For issue-areas in which opinions are relatively fixed, we should expect that none of the variance in responses is explained byfixedeffects and that there will be no correlation between leader preferences and group preferences', in those areas in which opinions are more easily swayed, it is conceivable that individuals in a group are influenced by some feature not necessarily related to a leader's attitudes (for example, they may be affected simply by a leader's age or gender) and subsequently adopt a position that the leader in turn adopts, swayed by the position of the group (itself a function of some feature of the leader other than her prior preferences). This argument relies on the notion that leaders are different from participants in the sense of being very open to suasion, whereas participants, though not open to suasion based on the attitudes of leaders, are very ready to change their views based on other characteristics of leaders. Convoluted as this argument is, we are not yet able to discount it.

⁴⁴ For an experimental study on cue taking, see Kuklinsky, James H. and Hurley, Norman L., “On Hearing and Interpreting of Political Messages: A Cautionary Tale of Citizen Cue-Taking,” journal of Politics 56, no. 3 (1994)Google Scholar.

⁴⁵ In a given instance it may not be possible to determine whether a particular conclusion provided by a leader that does not correspond to the opinions of individuals is a result of deliberate misrepresentation or self-censorship; both may be in operation. Persuasion and information provision may in practice be inseparable. And there is much scope for interactions: it is possible that that self-censorship by one section of a discussion group will lead to a preponderance of arguments by a second section that leads to a third section being persuaded by the arguments of the second section.

⁴⁶ It is of course possible for leader influence to alter the preferences of participants during the course of discussions but for this influence to be very short lived with a near-immediate reversion to prior attitudes subsequent to the discussions.

⁴⁷ Note that leader suasion (direct or indirect) is a sufficient but not necessary condition for changes in opinions.

⁴⁸ Although not reported here, no significant changes in any given direction were observed for any of the questions, with one exception: tax rates. Average preferred tax rates declined from 16 percent to 11 percent. Median rates for this issue-area were constant over time at 10 percent.

⁴⁹ We note, however, a caveat on the interpretation of our results. Our interpretation assumes that individuals do not systematically misrepresent their views when responding to surveys at the individual level. If instead respondents misrepresented their views either because of a consistency bias —with respondents wishing to provide the same answer to the question that they had previously provided, even though their positions may have changed, in order to appear to be consistent—or another audience effect on the part of respondents—with respondents not willing to provide a given answer in private even though they feel comfortable expressing the attitude in a public setting—then we would fail to reject the null for these reasons. While neither appears a priori plausible for the questions at hand, we recognize that we cannot rule out these possibilities.

⁵⁰ In further experimental work, however, such an effect could be more satisfactorily identified by distinguishing between the moderating and reporting functions of discussion leaders.

⁵¹ Note again that for situations like that described in fn. 36, the R ² is simply a lower bound on leader influence.

⁵² Chattopadhyay and Duflo (fn. 41).

⁵³ Ackerman and Fishkin (fn. 1).