Avoiding distortions: inequality and polarization
Critics of deliberation worry that the apparent commitment to equal consideration of everyone’s views on their merits will in fact mask domination of the process by the most privileged. The problem is that any microcosmic deliberation taking place in a modern developed society will be in one in which there are significant social and economic inequalities in the conduct of ordinary life in the broader society. It seems difficult or impossible to ‘bracket’ these inequalities - for participants to behave ‘as if’ they do not exist (see Fraser 1993: 1-32, especially 10-11).
How might the apparent equality of deliberative processes mask domination by the more privileged? Iris Marion Young distinguishes ‘external’ and ‘internal’ forms of exclusion. The external forms are the most obvious. Not letting someone be part of the participating group, either because their participation is barred or they are not effectively recruited, has been the focus of electoral/political reform for decades. And even in survey work, efforts to get at non-response, to reach the more difficult to reach, those with difficult schedules, those without phones, internet access or even fixed addresses7 can rightly justify significant expenditures of time and effort.
But Young’s point is that there are more subtle forms of exclusion that turn on manners of speaking and listening. Some people, even if formally included, may not have their voices, if they speak at all, taken seriously. They may give off cues that indicate they are not well informed or not worth listening to. Those who are accustomed to every advantage in the conduct of their everyday lives may be more assertive in pressing their views on others and less open to listening to those without similar advantages (Young 2000). They may also be more accustomed to orderly forms of reason giving argument that weigh with other participants. Or so the argument goes.
The empirical question for our research programme is whether or not those advantaged in actual life use the opportunity for shared deliberation to dominate the process. If all or most of the opinion changes move sharply in the direction of the more advantaged viewpoints, then that might be an indication that the advantaged are dominating. The issue is complex in that the advantaged may be more informed, at least on some issues, and if the idea of deliberation is that participants move in the direction of their more informed opinions, then participants may move in that direction because of information effects, not because of distortions, by a factor extraneous to the merits of the issues, that is, the social positions of some of the participants.
Those coming to this debate from a feminist perspective will be concerned about men dominating. Those focused on socio-economic inequalities will be concerned about domination by the rich and the more educated. The metrics for evaluating this claim could turn on the movement of policy attitudes toward or away from the supposedly dominating group, as well as on the distribution of speaking time among the participants.
In some of the Deliberative Polls, all the small group discussions were recorded, permitting study of the distribution of words used and the kinds of reasons offered. In a study of five American Deliberative Polls, Alice Siu looked at these two questions. About the distribution of talk, she concluded ‘that no particular gender, race, or demographic dominates deliberations’. In one of the Deliberative Polls, on health care and education in the US, by far the most talking (measured in number of words used) was done by non-white, less educated females with non-white higher educated females close behind. The least number of words were expressed by the white higher educated males. In a separate Deliberative Poll on the 2004 primary campaign, the pattern was largely reversed. All of the Deliberative Polls used representative samples of the US adult population (Siu 2009).
Even if talk is evenly distributed, it is possible that influence is not. If we look at issue indices for each topic of discussion and analyze the movement at the small group level, the five Deliberative Polls had 354 small group issue combinations (the number of issue indices in each Deliberative Poll times the number of groups calculated for all five Deliberative Polls). Siu found no significant pattern of movement in the direction of the initial positions of the whites, the males, the high income participants, or the more educated. In each case, movements in the direction of the more advantaged groups occurred only about half the time. In other words, about half the time a group ended up moving toward the initial position of the more advantaged (the more educated, the whites, the rich or the male) but about half the time it moved away. And when the magnitudes of the movement are examined, rather than the numbers of movements in each direction, the amounts are small considered as a percentage of the total range of possible movement (Siu 2009).8
Regardless of socio-economic and gender inequalities, there are long-standing concerns that the process of group discussion itself may bring distortions. Cass Sunstein (2003), building on earlier work on the so-called ‘risky shift’, has claimed there is an inevitable ‘law of group polarization’. If there is a dimension for which there is a midpoint, and the mean of a small group starts out on one side of that midpoint, he hypothesizes that it will move farther out from the midpoint in the same direction. If it starts out to the right it will move further right. If it starts out to the left it will move further left. The argument is not dependent on liberal and conservative accounts of left and right - just on there being a dimension with a midpoint. The effect is meant to apply generally to any issue dimension that is the topic of discussion.
The idea is that this distortion will occur because of two dynamics. First, if the group starts out on one side of the midpoint, then there is likely to be an imbalance in the argument pool on that side of the issue. More arguments will be offered to motivate further movement in that direction. The second dynamic is a social comparison effect. People will compare their own positions to that of the others in the small group and feel social pressure to conform to the direction of the new consensus.
The argument is a challenge to the legitimacy of deliberative democracy because if there is a reliable pattern of group psychology that predicts the movement of opinion, then it is hard to hold that the movement is based on the merits. Regardless of the merits in a particular case, the group will supposedly move in a stated direction. The issue is an empirical one and may well vary with the precise institutional design of a deliberative process. Sunstein (2003), however, holds that it will apply generally to group discussion processes.9
The polarization argument put forward vigorously by Cass Sunstein, represents a different kind of distortion. Sunstein argues that there is a predictable pattern: group discussions lead to extremes. If there is an issue for which there is a midpoint, his ‘law of group polarization’ asserts that, if the mean position of the group begins on one side of the midpoint, it would move farther out from the middle in the same direction. If the mean position begins on the other side of the midpoint it will move farther out from the middle in that direction (Sunstein 2006; Schkade et al. 2007).10
However, we looked at the degree of polarization in fifteen Deliberative Polls with 1,848 group/issue combinations (the number of issue indices in a given Deliberative Poll times the number of small groups in that Deliberative Poll for all fifteen Deliberative Polls). The proportion of small groups moving away from the midpoint turns out to be 50 per cent. In other words, the other 50 per cent of the time, the movement was toward the midpoint and so there was no tendency at all towards polarization in Sunstein’s sense. These Deliberative Polls took place in various countries, including the US (six cases), Britain (five cases), Bulgaria, China, Greece, and Australia. All employed scientific random samples and face to face discussion (Luskin et al. 2007a).