Participation in Transition: Mobilizing Ugandans in Constitution Making

Survey research indicates that previous participation is often the strongest predictor of current participation.

Politicians, political parties, and interest groups usually target those individuals and groups who were already active in the past.

Initial turnout levels could be altered slowly over time by generational replacement. Individuals socialized after the initial honeymoon period of democratization may be less inclined to turn out than their elders. Nonetheless, if younger generations are socialized during periods of high turnout, they are more likely to become habitual voters themselves and so initial levels of participation can persist even across generations (Franklin 2004).

The survey, which I designed and managed, is based on a national probability sample to represent a cross-section of Ugandans aged 26 and older. The minimum age limit was designed to include only those individual who were of voting age during the constitution-making period. Nine districts (Gulu, Kitgum, Kotido, Moroto, Bundibugyo, Hoima, Kabalore, Kasese, and Kibaale) were excluded from the sampling frame because of instability and rebel attacks at the time when I conducted the sampling. Therefore, the resulting data is not representative of these troubled areas. To ensure that every adult Ugandan in the sample area had an equal chance of selection, I employed a clustered, stratified, multistage, area probability sampling design. After stratifying by urban/rural localities and region (northern, eastern, central, and western), a probability proportionate to population size (PPPS) method was used to randomly select districts, sub-counties, and parishes in successive stages. A single primary sampling unit (PSU) was randomly selected from each parish (population data did not exist at the PSU level). The randomly selected PSUs included 6 urban and 62 rural sites within 13 districts: Kampala, Jinja, Mpigi, Luwero, Nakasongola, Mayuge, Iganga, Apac, Lira, Sironko, Mbale, Mbarara, and Bushenyi. Working with the local council officials, my research teams compiled lists of all the households in each selected PSU and then selected a random sample of 16 households from each PSU list. A single individual was randomly selected from the list of household members through blind selection from a pack of numbered cards. If not available at first, two additional attempts were made to contact the selected individual before a new randomly selected household was substituted for the original household. The 92-item survey instrument was translated into the five languages using the technique of translation/back-translation to check for inaccuracies and to obtain congruence between the different languages. The survey was administered face-to-face by teams of trained native-speaking interviewers.

The one main exception is party membership, which appears to be higher in poor countries than in rich ones and higher in newer and semi-democracies than older democracies (Norris 2002: 121). However, these figures are based on World Value Surveys in 59 countries with a small number of poor, semi-democratic, and nondemocratic countries.

Scandinavia has the highest average turnout for any region (80.8%) and North America has the lowest (50.2%). Africa has the second lowest regional turnout rate (55.5%). These figures are based on the mean turnout in national-level parliamentary and presidential elections in the 1990s (Norris 2002: 45).

In round 1 Afrobarometer surveys, 47% of respondents reported attending a community meeting, 43% reported joining others to raise issues, 11% reported protesting, and 14% reported contacting an official (Afrobarometer 2002: 4).

United States Agency for International Development (2000: Annex A)

Verba and Nie (1972) use the name “SES model” to refer to demographics, resources, and engagement. In this article I use “SES” to refer to demographic and resource variables and “civic orientations” to refer to engagement and other attitude variables thought to affect participation.

In most advanced industrial democracies, the gender gap in voting has closed, and it may be narrowing for other modes as well (Burns et al. 2001; Inglehart and Norris 2003). In contrast, the education and income gaps seem to be widening (Kittilson 2005).

Some notable and enlightening exceptions are: the studies by Verba and his colleagues that use two-stage least squares procedures (Verba et al. 1997; Verba et al. 1995); analysis of panel date by Finkel (1985, 1987) and Gibson (1996); and path analyses by Clarke and Acock (1989)

The studies by Brady and Kaplan (2003) and Bratton (1999) may overestimate the effects of orientations on participation because their empirical models do not reflect the reciprocal nature of the relationship, even though they acknowledge it theoretically. Brady and Kaplan justify the choice of a simple unidirectional OLS model by noting that the results for the reciprocal two-stage least squares model and the OLS model did not differ much in an earlier examination of the American case. Yet the effects of participation on attitudes and knowledge are likely to be larger in new democracies than in established democracies. Thus, we can not conclude with any certainty whether orientations affect participation in new democracies without explicitly modeling the reciprocal relationship.

For instance, Bratton (1999) argues that the weakness of these recruitment institutions accounts for the low participation rates in Zambia.

In my survey, 59.9% of respondents reported voting for a CA delegate.

Parliamentary and presidential elections were also held just before independence and many of the political parties and attachments developed in this initial election persist to today. Local elections were also held before the CA elections.

The Cronbach’s alpha for the participation activities index is 0.73, indicating a single underlying concept, and the factors are almost equally weighted (.43, .61, .68, .65, .57, and .55). There does not appear to be a clear hierarchical ordering to the activities whereby those who participated in costly activities also participated in easier activities. For example, it is not the case that those individuals who attended a seminar where a member of the UCC was present (the least frequent and most demanding activity) also engaged in all other activities; two-thirds did not. Conversely, the most frequent activity was voting, but 21% of those who engaged in only one activity, did something other than voting. Nonetheless, I tested two different weighted index variables using the same questions as the participation activities index. First, I weighted activities more if they were more exacting in time (for both travel and attendance) and/or effort required. When the first weighted index is used, the main results remain the same as those presented in the rest of this paper except that poverty and exposure to newspapers are no longer statistically significant in the OLS model. The predictive power of the OLS model (adjusted R ² = 0.38) was only slightly less than with the unweighted participation activities index. Second, I weighted the activities by the proportion of the population that did not participate. This gives greater influence to those activities that drew fewer people. Again, when I used this second weighted index, there were no changes in the main results reported in this paper, except that poverty and distance to headquarters were no longer significant in the OLS model. The predictive power of the model (adjusted R ² = 0.40) was nearly the same as with the unweighted participation activities index. I use the unweighted simple sum of participation activities index in this paper for ease of interpretation.

If a respondent gave just one or two answers, the interviewer prompted, “any other ways?”

The correlation between the two participation measures is 0.57. It is noteworthy that both measures are self-reported participation. It is possible that some respondents forgot they participated. However, in open-ended interviews most respondents noted that these were very important activities given the nature of the topic (the constitution) and the timing of the activities (early in the transition). To assess the difference between actual and reported participation, I conducted interviews with subjects selected from lists of participants (memoranda, meeting attendance, and seminar attendance). I also matched survey respondents with names on the lists of participants and checked their answers. There were some discrepancies due to either memory loss or inaccuracies in the lists of participants, but by and large, reported participation accorded with recorded participation.

Most empirical research from advanced industrial democracies, and modernization theory, would lead us to expect that urbanites are more participatory than rural residents. Yet several studies of developing countries produce the opposite conclusion. In Verba et al.’s (1978) study of seven countries, rural residents were somewhat more participatory than urban residents and Nelson (1987) concluded that rural residents in developing countries are more participatory in community activities. Based on Afrobarometer survey results from Africa, Bratton (2006) found that rural residents were significantly more likely to vote, attend community meetings, and contact informal leaders (controlling for poverty, gender, age, employed, and education).

Including measures of other levels of education did nothing to improve the predictive power of the model.

The measure of poverty is based on the question: “In the last 12 months, how often have you or your family gone without: (a) enough food to eat, (b) medicine or medical treatment that you needed, (c) enough fuel, and a (d) cash income? Never, rarely, sometimes, or often?” Due to the difficulty of obtaining accurate income figures from respondents working primarily in the informal economy, I tested several different measures of wealth. Only the one that measured the lower range of wealth, or poverty was significant in predicting participation. A weighted scale of the number of durable consumer goods owned by the respondents’ households (owns consumer goods) was insignificant and the coefficient was virtually zero (standardized coefficient = 0.01, p = 0.79), nor did it improve the predictive power of the model. Another commonly used measure of SES, employment, was found not to influence participation in constitution-making (standardized coefficient = 0.03, p = 0.35). The correlation between the measure of poverty and the measure of consumer goods is 0.37, while the correlation between the measure of poverty and the measure of employment is only 0.09, reflecting the high level of self provisioning in Africa. Bratton (2006) conducted a more in-depth study of poverty and wealth in Africa based on the Afrobarometer surveys. He notes that in response to the open-ended question “In your opinion, what does it mean to be ‘poor’?” Forty-six percent mentioned lack of food and 36% mentioned lack of money. Bratton (2006: 8) constructs a similar measure of how often respondents report going without: food, clean water, medicines, electricity, fuel and cash and he concludes that “the Index of Lived Poverty is more highly correlated with both household income and the wellbeing ladder than either of these indicators is correlated with the other. We take this as evidence that this experiential indicator meets the goal of bridging objective and subjective perspectives on poverty. We expect that this item, which combines income with other, less tangible and ‘in kind’ manifestations of wellbeing (like food, education and health), is a more valid and reliable construct than either alternative.” Like Bratton, I find that the measure of poverty to be the most valid and reliable indication of the type of wealth that matters for political participation, and so I use this measure for the following analysis. Therefore, the results should be interpreted as distinguishing between those who are impoverished and those who more often have their basic needs met, and not for distinguishing between middle and upper class Ugandans. However, the other results in the model do not change if I include the alternative measures of wealth and employment in addition or instead of the measure of poverty (except for exposure to newspapers which loses significance when the other measures are used).

In contrast, however, Bratton (2006) found a significant negative relationship between education and voting in 2005 (though not in 2000–2001), and a significant positive association between education and attending community meetings and contacting informal leaders (controlling for urban/rural residence, gender, age, employed, and poverty). In addition, Bratton found that poverty in Africa was positively associated with voting in 2005 (though not in 2000–2001), attending community meetings, and contacting informal leaders (controlling for urban/rural residence, gender, age, employed, and education).

I encountered two obstacles to specifying a two-stage model. First, I am unable to predict enough of the variance in the variable interest in the first-stage equation. This was especially problematic because the uneven predictive power for the two endogenous variables would probably lead to biased results (the R ² for the first-stage equation predicting interest is 0.13, where as the R ² for the equation predicting participation is 0.40). Second, and even more problematic, I was unable to find adequate instruments for the model. I could not find a variable that was significantly related to interest but not to participation in the first-stage equations.

An alternative approach would be to exclude the measure of interest from the model. If I do this, then the coefficients on age and poverty are no longer statistically significant. All other findings remain the same, including the results about the relative influence and precision of resource variables versus mobilization variables (comparing the block of five resource variables and the five most influential mobilization variables).

Local councilors were not required to participate in the constitution-making activities.

Associational affiliation is an index variable that records the number and level of membership in associations based on the following question: “Now I am going to read out a list of voluntary organizations. For each one, could you tell me whether you are an official leader, an active member, an inactive member or not a member of that type of organization: (a) a religious organization like a church or a mosque; (b) art, music and drama clubs; c) farmers organization; (d) professional or business association; (e) credit and savings association; (f) women’s organization; (g) sports or drinking clubs; (h) parents–teachers association; (i) other.”

These three measures record how often respondents get news about local or national politics from meetings, radio and newspapers.

It is possible—though improbable—that these factors of mobilization are also subject to reverse causation. In some circumstances, participation might cause citizens to become involved in mobilizing institutions. However, most of my measures are unlikely to be the result, rather than the cause, of participation. Local council members typically are elected because they are thought capable of delivering resources and commanding respect. They tend to be individuals who occupy important positions in traditional hierarchies, have previous contacts to higher officials, and are relatively wealthy. Most of those involved assumed positions in the resistance councils or local councils prior to the constitution-making process. In addition, the question includes previous positions. Similarly, respondents most often reported that they personally know higher level government officials because they were relatives or village-mates, relationships that existed prior to the process. Furthermore, the organizations used to construct the associational affiliation measure (religious, art, farmers’, professional, credit, women’s, and sports associations) are not the types of political organizations that individuals would tend to join as a result of involvement in constitution-making activities. Ethnic identity and location of residence are also not factors participation affects. Participation in constitution-making activities is, however, somewhat likely to motivate greater exposure to news. As with interest, I am unable to adequately specify a two-stage model of exposure to news and participation. Readers should keep in mind that the coefficients on the variables of news exposure potentially include both news exposure’s influence on participation and participation’s influence on news exposure.

These results are robust to different specifications of the model. As I mentioned before, if the variables for secondary education, possession of consumer goods, employment, and measures for the different ethnic groups are included these variables are not significant. In addition, they add virtually nothing to the predictive power of the model, and the effects of the other variables remain stable if they are included (with the exception of poverty and road conditions which lose some statistical significance). This is also the case if a measure of support for the NRM is included. Citizens who did not support the government in power were no less likely to participate than citizens who did support the NRM government. Including this measure of party support has no influence on the predictive power of the model.

These substantive rankings are based on the standardized coefficients.

As noted previously, the measure of poverty may not capture wealth’s full effect on participation. It is possible that a more detailed measure of wealth that distinguishes between individuals along the full spectrum from extreme poverty to great wealth might have produced a larger estimated effect. However, the complete lack of estimated influence of my measure of ownership of consumer goods suggests that wealth probably does not have a sizeable influence on participation.

The predictive power of the variables is measured by the drop in the predictive power from the full model. The drop is 0.41–0.33 = 0.08 for the individual-level resource and orientation variables and 0.41–0.30 = 0.11 for the mobilization variables.

Mujaju, who conducted a study of the CA campaigns and election in Fort Portal Municipality, Kabarole, supports this conclusion. Mujaju (1996: 51) writes, “NRM managed to create an environment where usually apathetic people turned up in large numbers to vote.”

BIC approximations can be interpreted as measures of overall model fit and “two models can be compared by taking the difference of their BIC values, with the model having the smaller (i.e., the more negative) BIC value being preferred” (Raftery 1995: 134). Table 1 provides “very strong support” for the claim that the full model (A) fits the data better than either of the reduced models (B and C), indicating that both individual and mobilizations variables help predict participation. Raftery (1995: 140) asserts that BIC differences greater than 10 correspond to posterior odds greater than 150:1 and should be interpreted as “very strong” evidence in favor of the model with the lower BIC. The difference between models A and B is 55 and the difference between models A and C is 87. In addition, the reduced model with the mobilization variables (B) fits the data better than the reduced model with individual traits (C). The difference of 32 between the BIC values provides “very strong support” for the hypothesis that mobilization provides a better explanation of who participated than individual demographics, SES and orientations.

For a detailed description of these models see Moehler (2007).

Three-stage least squares estimates are not used due to the potential “spread” of specification error (Cragg 1983).

The systems are identified. Most of these instruments (variables excluded from one equation in the system but included in the other) were deliberately incorporated into the survey questionnaire for this purpose. The appendix reveals that the instruments are orthogonal to the alternative dependent variable in the multivariate context. They are also uncorrelated in bivariate relationships. The first equation in each system is the same as that used in Table 1. The second equations for the four systems are as follows:

The reliability of the index is high (Cronbach’s alpha = 0.82) indicating that the different questions measure a single concept.

The reliability measure is low (Cronbach’s alpha = 0.33) but this is not surprising since I am measuring adherence to different dimensions of democracy and have no reason to assume that the different dimensions will cohere. A democratic person would be high on all dimensions, but being high on one does not necessarily cause one to be high on the others. The reliability analysis shows that the factors are weighed equally.

Cronbach’s alpha = 0.73. The loadings on a single factor are police = 0.59; courts = 0.63; local council = 0.44; and electoral commission = 0.46. Each individual question was coded as yes = 1, no = 0, or it depends = 0.5 (comments on “it depends” answers were written down and checked for accuracy of meaning).

The first-stage results can be found in the appendix.

Participation in Ugandan constitution-making did raise political and constitutional knowledge and heighten democratic attitudes. It had no effect on civic capabilities, and surprisingly, it lowered institutional trust (Moehler 2007).

I would like to thank an anonymous reviewer for this insight.

For a summary of the debated about autonomous versus mobilized participation see Nelson (1987).