Civic Education, Political Discussion, and the Social ......Civic education first affected the knowledge, values, and participatory inclinations of individuals directly exposed to

Civic Education, Political Discussion, and the SocialTransmission of Democratic Knowledge and Valuesin a New Democracy: Kenya 2002

Steven E. Finkel University of PittsburghAmy Erica Smith University of Pittsburgh

How does civic education affect the development of democratic political culture in new democracies? Using a unique three-wave panel data set from Kenya spanning the transitional democratic election of 2002, we posit a two-step process of thesocial transmission of democratic knowledge, norms, and values. Civic education first affected the knowledge, values, andparticipatory inclinations of individuals directly exposed to the Kenyan National Civic Education Programme (NCEP).These individuals became opinion leaders, communicating these new orientations to others within their social networks.Individuals who discussed others’ civic education experiences then showed significant growth in democratic knowledgeand values, in many instances more than individuals with direct exposure to the program. We find further evidence ofa “compensation effect,” such that the impact of civic education and post-civic education discussion was greater amongKenyans with less education and with lower levels of social integration.

Regime transitions to democracy bring with themtransitions of another sort: rapid changes in thepolitical behaviors, knowledge, and values ex-pected of citizens.1 Individuals quickly need to learn thebasic structure of a new set of institutions and norms, todevelop new political loyalties, and to figure out whereand how they fit in. How can citizens acquire these newcivic competencies and attitudes in such a short periodof time? Theorists initially posited that the acquisition ofdemocratic norms, values, and participatory orientationsamong individuals in new democracies would be a long-term process linked to social modernization and gen-erational replacement (Almond and Verba 1989; Lipset

Steven E. Finkel is Daniel H. Wallace Professor of Political Science, University of Pittsburgh, 230 S. Bouquet St., Pittsburgh, PA 15260([email protected]). Amy Erica Smith is a Ph.D. Candidate in Political Science, University of Pittsburgh, 230 S. Bouquet St., Pittsburgh, PA15260 ([email protected]).

Previous versions of this article were delivered at the 2008 annual meeting of the American Political Science Association, Boston, and atDartmouth College in 2009. We thank Aaron Abbarno, Scott Althaus, Michael X. Delli Carpini, Muge Finkel, Michael Herron, KosukeImai, Kristin Kanthak, George Krause, Dean Lacy, Reynaldo Rojo-Mendoza, William Mishler, the anonymous reviewers, and both thecurrent and previous editor of this journal for helpful comments. Thanks to David Leuthold and Paul Mbatia for their invaluable assistanceto the entire research project and to Melissa Baker, Michael Muindi, Joah Mboga, and Jacinta Mulwa from Research International, Kenyafor their efforts in the data collection activities. We also are grateful for the support of Roberta Warren and Lynn Carter at ManagementSystems International; Sheryl Stumbras, formerly of USAID/Kenya; Gary Hansen, former head of the Democracy and Governance Bureau,USAID/Washington; and Harry Blair of Yale University.

1Data and Stata files for replication are available at www.stevenfinkel.info/CivicEducationResearch/Finkel&SmithAJPS.html. A SupportingInformation document can also be accessed at that link. The study was supported by the United States Agency for International Development(USAID/Kenya), under AEP-I-00-00-00018, Task Order 806 to Management Systems International, Washington, DC.

1959). A spate of more recent research on third-wavedemocracies, however, has established a new truism: thatdemocratic orientations are malleable, in the medium andeven in the short term. Evidence abounds that democraticattitudes in transition societies respond to the regimes’political and economic performance (Mattes and Brat-ton 2007; Mishler and Rose 1997; Seligson 2002). Massattitudes in new democracies also have been found tomirror shifts in elite political culture, ideology, and at-titudes toward civil liberties (Gibson and Gouws 2003;Mishler and Rose 2007). And democratic values in somepostcommunist and authoritarian societies developedquickly due to “demonstration effects,” whereby Western

American Journal of Political Science, Vol. 55, No. 2, April 2011, Pp. 417–435

C©2011, Midwest Political Science Association DOI: 10.1111/j.1540-5907.2010.00493.x

417

418 STEVEN E. FINKEL AND AMY ERICA SMITH

democracies served as positive role models observedthrough the international mass media (Dalton 1994;Rohrschneider 1999). These findings all suggest thatdemocratic transitions can be accompanied by relativelyrapid changes in political orientations, as the new regimedemonstrates positive performance, and citizens learnfrom new leaders, institutions, and other democratic sys-tems.

But are there more direct ways to promote democraticnorms, values, and participatory orientations? Given theplethora of new democratic regimes that exhibit poorperformance or outright backsliding toward authoritari-anism, successful transmission of democratic norms andvalues in the short run would seem to require inter-ventions that are independent of potentially problem-atic institutions and elites. Perhaps the most promisingdirect means for promoting democratic orientations innew democracies is through civic education programs,which teach democratic citizenship to young people inclassroom settings or to adults in community workshops,lectures, or public fora (Finkel 2003a; Torney-Purta et al.2001). Over the past several decades, there has been anexplosion of such programs in the emerging democraciesof Eastern Europe, Africa, and Latin America, with thevast majority funded by the United States, other OECDdonors, or philanthropic organizations seeking to stimu-late more democratic political culture (Carothers 1999).These programs range from new primary and secondaryschool curricula on democracy, to local NGO programsproviding instruction about the social and political rightsof women, to voter education, to neighborhood problem-solving programs bringing individuals and local authori-ties together. It is difficult to estimate the precise numberof these programs in developing democracies, but UnitedStates Agency for International Development (USAID)data suggest that the United States alone spent between$30 million and $50 million a year on civic educationbetween 1990 and 2005.2

Despite the proliferation of civic education programsin new democracies, there has been relatively little re-search on their effectiveness in changing democratic ori-entations among students or adults. Given the amount

2These figures were obtained from official USAID activity data,available at http://www.pitt.edu/∼politics/democracy/democracy.html as part of the project “Deepening our Understanding ofthe Effects of US Foreign Assistance on Democracy Building,1990–2004,” principal investigators Steven E. Finkel, Anı́bal Pérez-Liñán, Mitchell A. Seligson, and C. Neal Tate. The data show that ex-penditures in the category “Civic Education, Civil Society and Lead-ership Training,” which includes all civil society activities fundedby USAID aside from activities related to labor unions and thesupport of independent mass media, totaled more than $2 billionbetween 1990 and 2005.

of money invested in civic education, as well as the the-oretical importance of supportive mass dispositions forthe consolidation of new democracies, clear and solid ev-idence regarding the effectiveness of these programs iscritical. A small number of published studies, however,has emerged in recent years. Most report that civic ed-ucation among secondary school students (Slomcyznskiand Shabad 1998; Torney-Purta et al. 2001) or adults incommunity-based workshops (Finkel 2002, 2003a) sig-nificantly impacts political knowledge and participation,as well as, under certain conditions, democratic valuesand norms such as tolerance, efficacy, and institutionaltrust. Thus, there is some optimism regarding the ben-efits of donor-sponsored civic education for democraticpolitical cultures in recipient countries.

In this article, we add to this emerging literature by as-sessing the impact of a major countrywide civic educationinitiative in Kenya during the period spanning the transi-tional democratic election of December 2002. The KenyanNational Civic Education Programme (NCEP) consistedof some 50,000 discrete workshops, lectures, plays, andpuppet shows, and community meetings conducted bynearly 80 Kenyan NGOs between late 2001 and December2002. These activities aimed to promote civic skills, demo-cratic values, and engagement in the democratic regimeamong ordinary Kenyan citizens—specifically to preparethem to exercise the vote, to provide input to an ongo-ing constitutional reform process, and “to contribute tothe consolidation of a mature political culture in Kenya”(NCEP Programme Document, cited in Finkel 2003b).We estimate that approximately 15% of all Kenyans ofvoting age were trained in the program, making it oneof the largest—if not the largest—coordinated civic ed-ucation program for adults yet conducted in developingdemocracies (Niemi and Finkel 2007). Our evaluationis similarly national in scope and can therefore speak toboth the program’s microlevel effects and its aggregate-level impacts on Kenyan democratic orientations.

Kenya, moreover, is an ideal context in which to carryout an assessment of civic education’s impact. It is a coun-try whose political culture has many characteristics com-mon to struggling democracies, including long-standingethnic rivalries and inequalities (which erupted into vi-olence most recently in 2008), high levels of intoleranceand distrust, and relatively low levels of citizen engage-ment with the political process (Barkan 2008; Gugertyand Kremer 2008). At the same time, after only fitfuland uncertain movement away from one-party rule inthe years leading up to the 2002 election, there was realhope that year that true democratic change in the coun-try was possible. The Kenya NCEP, then, took place ina difficult context, yet one where it had at least some

CIVIC EDUCATION AND POLITICAL DISCUSSION IN A NEW DEMOCRACY 419

chance of building successfully on recent democraticdevelopments.

We address two major deficiencies in previous stud-ies. One is theoretical, in that previous research has beenlimited by a restrictive view of the target population ofcivic education programs. We argue that, rather than fo-cusing solely on individuals trained in the program perse, evaluations of civic education must also examine itsindirect effects, whereby treated individuals go on to dis-cuss lessons and ideas in the workshops with “untreated”members of their social networks. To the extent that civiceducation stimulates such democratic discussions—andwe show below that they occurred in Kenya with greatfrequency—democracy education may exert even greaterimpact on democratic political culture than previouslyrecognized, as the “downstream” effects of political dis-cussions amplify the programs’ direct effects.

The other deficiency we address is methodological.Knotty problems of selection and control are endemic tothe evaluation of civic education in real-world settings.School-based civic education is often mandatory, mak-ing it difficult to compare trained students to appropriatecontrol groups, while the voluntary nature of adult civiceducation leads inexorably to problems of self-selectionbias. Overcoming these problems is difficult in observa-tional studies and nearly impossible with the single-shotcross-sectional designs utilized in the literature thus far.In this study, we make use of the first longitudinal de-sign to date in the evaluation of adult civic educationin developing democracies. The three-wave panel datahere allow us to disentangle the effect of civic educationfrom that of prior attitudes and to estimate the effects ofcivic education exposure while controlling for confound-ing variables that may be associated with both exposureto civics training and democratic knowledge, values, orbehaviors.

The results show strong support for a “two-step”model of the impact of civic education. Using a va-riety of longitudinal models to control for selectioneffects, we find first that the Kenyan National CivicEducation Programme (NCEP) affected the knowledge,attitudes, and participatory inclinations of those directlytrained in the program. These individuals then becameopinion leaders, communicating new democratic orien-tations to neighbors, family members, and friends withintheir social networks. We show that individuals with nopersonal exposure to the program who discussed others’civic education experienced significant growth in politi-cal knowledge, tolerance, and a sense of national versustribal self-identification, in many instances more than in-dividuals who were directly trained. Moreover, the resultssuggest that the effects of civic education depend criti-

cally on factors related to the nature of the individual’scivic education experience, as well as on factors relatedto the individual’s previous store of information and so-ciopolitical resources. Civic education had consistentlygreater impact when workshops were conducted with ac-tive, participatory methodologies; and the effects of civiceducation and post-civic education discussion were con-centrated among Kenyans with less education, and withlower levels of social integration. The study presents per-haps the strongest evidence to date of the full range ofcivic education’s effects and reinforces its promise forpromoting democratic political culture.

Civic Education, Political Discussion,and Democratic Learning:Theoretical Expectations

Several recent studies suggest that civic education in newdemocracies can have a significant impact on a variety ofdemocratic orientations, especially basic knowledge andpolitical participation. Finkel and Ernst (2005), for ex-ample, find that high school students who receive civicsinstruction on at least a weekly basis are far more likely toidentify correctly key South African political leaders andto possess basic knowledge of the South African constitu-tion than students who receive civics instruction less of-ten or not at all. Among adult populations, Finkel (2002)shows rather large effects of civic education on politi-cal participation, especially at the local level: adults innine programs in the Dominican Republic, Poland, andSouth Africa were nearly twice as likely as control groupmembers to attend municipal meetings or participate incommunity problem-solving activities. Evidently, con-ducting adult civic education through “advocacy NGOs”and other civil society organizations augments the nor-mal mobilization processes taking place within thesegroups.

The effects of civic education in these studies ondemocratic attitudes and values such as tolerance, effi-cacy, and institutional trust are typically more modest.3

These dispositions should be relatively resistant to change,as a long line of research on democratic values pointsto their origins in deeply rooted factors such as dog-matism, psychological security, and individual locus of

3In fact, the effects of civic education on political trust is some-times negative, as such training appears to raise awareness of thedeficiencies of less democratic regimes and the performance of po-litical elites (Finkel, Sabatini, and Bevis 2000). Moehler (2008) findsa similar pattern in an analysis of the impacts of civic participationin constitution-building activities in Uganda.


control (Sullivan and Transue 1999, 632). However, civiceducation may change even these “difficult” democraticorientations under certain conditions, specifically whenexposure to civics training is more frequent, and whentraining makes greater use of participatory teachingmethodologies such as role playing, group problem-solving activities, open discussions, and the like. Roleplaying and other active behaviors within the small-groupsetting allow individuals to practice or “try out” new ori-entations within a safe environment; and much researchin social psychology suggests that these kinds of exer-cises stimulate attitude change that is consistent with thebehaviors that are being acted out (Campbell 2008; Zim-bardo and Leippe 1991). Previous civic education researchconfirms these processes as well, as a host of studies amongschool-age children (Campbell 2008; Finkel and Ernst2005; Niemi and Junn 1998; Torney-Purta et al. 2001)and adults (Finkel 2002, 2003a) in both developed anddeveloping contexts shows that exposure to democracytraining that makes use of open discussion and partici-patory methodologies has significantly greater effects ondemocratic orientations than does lecture-based instruc-tion (Torney-Purta et al. 2001). This discussion leads totwo main expectations for the effect of exposure to civiceducation training: it should have somewhat greater ef-fects on knowledge and participation than on democraticvalues, and its effects will be greater among individualswho are trained more frequently with more participatorymethodologies.

All previous civic education research, however, hasbeen limited in theoretical scope, in that evaluations havefocused on its effects only among individuals directly ex-posed to democracy training . We argue instead that civiceducation involves a two-step process, whereby trainedindividuals go on to discuss the messages and ideas fromprograms with others in their social networks. The two-step process of message reception and subsequent dif-fusion through social networks has long been viewed asone of the primary ways that mass media affect politicalattitudes (Berelson, Lazarsfeld, and McPhee 1954); morerecent research has shown a similar effect for campaignmobilization, as individuals contacted by parties and get-out-the-vote efforts in turn mobilize others in their socialnetworks (McClurg 2004; Nickerson 2008). Despite thetheoretical importance of these kinds of social diffusionprocesses, however, they have not been taken into accountin any previous study of the effects of civic education.

There are several reasons to expect post-civic educa-tion political discussions to be relatively widespread. First,nearly all adult civic education in developing democraciesis conducted through NGOs or other secondary groupsthat organize these activities in their local communities.

Group-based discussion networks are thus already avail-able to most civic education participants. Second, as notedabove, civic education programs for adults in new democ-racies frequently consist of workshops that make extensiveuse of group-based political discussion as a pedagogicaltechnique. Thus, civic education participants are likely toengage in a good deal of political discussion as part oftheir training, and they are likely to have access to otherindividuals with whom they may continue discussionsafter formal training has ended.

When civic education leads to political discussionamong network members, it may then activate the wholerange of social network, discussion, and deliberation ef-fects on democratic learning and participation that havebeen found in previous research. A host of studies, forexample, shows, that political discussion promotes gen-eral political knowledge (Delli Carpini and Keeter 1996;Eveland et al. 2005), tolerance for and awareness of thereasons behind others’ views (Mutz 2006), support fordemocratic institutions and processes (Gibson 2001), andpolitical participation (Lake and Huckfeldt 1998; Mc-Clurg 2003, 2006). These effects should be particularlystrong in new democracies, where individuals are “activelytrying to learn the new rules of the political game,” andwhere everyday conversations may be “especially effec-tive in transmitting novel information about unfamiliarpolitical institutions” (Gibson 2001, 54). Indeed, Kenyansocial networks appear to play a major role in conveyinginformation and changing norms on issues such as birthcontrol and disease prevention (Behrman, Kohler, andWatkins 2002; Kremer and Miguel 2007).

But there are deeper links between civic education,political discussion, and democratic change, beyond thefact that civic education may simply lead to more politicaltalk and deliberation. Civic education trainees are likelyto be especially effective agents of democratic socializa-tion. For one thing, their participation in civic educationactivities is likely to make them de facto experts on demo-cratic processes within their social networks, in particu-lar networks in emerging democracies characterized byrelatively low levels of political information (Huckfeldt2001; Lake and Huckfeldt 1998; McClurg 2006). For an-other, trainees are likely to have relatively strong socialand emotional bonds with network members, makingthem especially effective at transmitting dissonant ideas(Huckfeldt and Sprague 1995; Walsh 2004). Given thatmany social networks in new democracies may be rela-tively closed, homogenous, and not particularly infusedwith democratic values (Gibson 2001), the affective bondsthat network members share may be valuable in persuad-ing those who are not exposed to civic education to acceptprodemocracy messages.


For all of these reasons, we hypothesize that civiceducation influences attitudes, values, and participatoryorientations in new democracies through a two-step dif-fusion process. In the first step, democracy training—especially training that makes use of open, participatoryteaching methodologies—influences the knowledge, val-ues, and behaviors of individuals exposed to civics in-struction. In the second step, there are indirect effects onthe individuals with whom trainees talk. To the extentthat previous research has failed to take these secondaryor indirect effects into account, civic education’s over-all impact on democratic change may have been greatlyunderestimated.

Finally, we can expect civic education and post-civiceducation discussion to have greater effects on some in-dividuals than others. In line with previous research onpolitical mobilization, Finkel (2002) found that the ef-fects of adult civic education programs on participationaccrue disproportionately to those with higher levels ofsocial integration and other political resources. This pat-tern may also hold for the participatory impact of post-workshop political discussions as well; U.S.-based work(Kwak, Shah, and Holbert 2004; McClurg 2003) finds thatthe highly educated and more socially integrated partici-pate more as a result of political discussion. Connectionsto civil society in Kenya have similarly been found to re-inforce participatory inequalities and privileges based aspatronage or clientelistic networks (Gugerty and Kremer2008; Orvis 2001a).

At the same time, theories of attitude change andpersuasion would support an alternative “compensation”effect, whereby civic education and post-civic educationdiscussion would have the greatest impact on the attitudesand values of individuals with fewer cognitive and socialresources (Campbell 2008). This pattern held, for exam-ple, in Valentino and Sears’s (1998) analysis of the effectsof new information conveyed during political campaigns,as this information tended to narrow the gap betweenadolescents and adults in political knowledge and atti-tude formation. The well-known McGuire-Zaller modelof attitude change (Zaller 1992) predicts that, conditionalon an individual’s attention to and comprehension ofthe content of a message, attitude change, or persua-sion is most likely to occur among individuals with lessprior information and among those whose prior atti-tudes and values are more ambivalent, less cohesive, andless crystallized. This would suggest that respondents withmore cognitive resources and who are more socially in-tegrated should be less likely to respond to democraticmessages conveyed through civic education and discus-sion. Not only are their prior attitudes likely to be morecrystallized, but in the Kenyan context they also will be

more likely to have encountered such messages elsewhere,whether through media to which access is socially strat-ified, through civic associations, or in the urban milieuitself (Orvis 2001b). We expect, then, that the less edu-cated and the less socially integrated will be more likelyto accept and to exhibit attitude change as a result of the“nonredundant” democratic messages conveyed throughcivic education workshops.4 Such a process would painta more benign picture of the impact of civic education,as its effects on attitudes and values would benefit thosewho, in important ways, “need it” the most.

Research Design and Data

We investigate these processes in the context of an evalua-tion of the Kenyan National Civic Education Programme(NCEP), a countrywide civic education initiative con-ducted during the run-up to the transitional nationalelections of December 27, 2002. This election produceda decisive victory for Mwai Kibaki and his NARC partycoalition over Uhuru Kenyatta, the candidate of the rul-ing KANU party, and the successor to the incumbentPresident Daniel Arap Moi. Moi had led the country incorrupt and authoritarian fashion since 1978, stifling op-position following an attempted coup in 1982 and es-tablishing Kenya as a one-party state thereafter. Inter-national pressure following the end of the Cold War inthe 1990s resulted in a gradual opening of the coun-try to multiparty elections in 1992 and 1997. Both elec-tions produced KANU victories, though both were widelyviewed as flawed, marred by intimidation of the opposi-tion and widespread violence. Nevertheless, some demo-cratic progress continued throughout the late 1990s. Moipromised to step down in 2002 in accordance with the10-year term limit set out in the Kenyan constitution,though until the date of the election was set (and evenduring the campaign), there was doubt about whether hewould follow through on this promise. In 2001, a Consti-tutional Review Commission was formed with the task ofwriting a new constitution in time for the 2002 election.Extensive public debate took place on a variety of propos-als, among them restricting presidential powers, creatinga prime ministerial position, and increasing the powers

4Provided, of course, that individuals receive and understand themessages conveyed. We expect that message reception among thepoorly informed in the case of the NCEP was likely to be quite high,as large portions of the curriculum and teaching materials weretailored specifically to socially marginalized communities in Kenyaand those lacking in basic literacy (see NCEP, Making InformedChoices).


of local and provincial governments. Though a new con-stitution was not agreed upon by late 2002, the electioncampaign proceeded, with the result being the first alter-ation of political power through the electoral process inKenyan political history. The election was widely viewedas a democratic breakthrough for the country (Barkan2004; Howard and Roessler 2006).

In this context, a consortium of so-called “like-minded donors” from the European Union funded an am-bitious civic education program to raise awareness aboutthe 2002 election and the associated constitutional re-form process. Some 80 Kenyan NGOs participated in theNational Civic Education Programme (NCEP), conduct-ing over 50,000 workshops and other organized activitieswith over 4.5 million individuals between late 2001 andDecember 2002. The program was national in scope, asthe NGOs developed and executed a detailed “roll-out”plan for covering different areas of the country. Further,the “treatment” was standardized to a significant degree,as the Programme administration spent nearly two yearsdeveloping a common curriculum, a handbook entitledMaking Informed Choices: A Handbook for Civic Educa-tion for the overall initiative, and a supplemental teachingmanual for conducting workshops in the field.

The United States Agency for International Develop-ment (USAID/Nairobi) commissioned Management Sys-tems International, a U.S.-based management consultingfirm, to conduct an evaluation of the project’s impacton participants in late 2001. The evaluation consisted ofpre- and post-civic education survey interviews with in-dividuals who attended NCEP workshops and with con-trol group respondents. We interviewed 2,601 individualsbetween February and April 2002, with half selected atrandom upon entering one of 181 selected NCEP work-shops sponsored by 26 different civil society organiza-tions throughout seven of Kenya’s eight provinces. Theother half consisted of individuals who did not attend thatworkshop, and who matched a given workshop attendeein terms of age, educational attainment, gender, and vil-lage or neighborhood of residence. This sampling strategyproduced nearly identical groups of initial workshop at-tendees and nonattendees in terms of the demographicfactors that we attempted to match. Two follow-up wavesof interviews were conducted. In October and November2002 (Wave 2), between six and nine months after theinitial NCEP workshops, we reinterviewed 1,787 individ-uals (901 initial workshop attendees and 886 from theinitial sample of matched nonattendees).5 Between late

5Response rates were extremely high. Of 2,601 respondents inter-viewed in the pre-civic education wave, we were able to obtainreinterviews in either the second or third wave with 2,301, or 88%.

March and early June 2003, we conducted a third waveof interviews with 401 respondents who had also beeninterviewed in Wave 2 of the study and with 514 “fresh”individuals who had previously been interviewed only inWave 1. The sample analyzed here thus consists of twowaves of panel data for 1,900 individuals (952 from theinitial workshop group and 948 from the sample of initialnonattendees) and three waves of data for an additional401 respondents (210 from the initial workshop groupand 191 from the sample of initial nonattendees). Table1 summarizes the sample information for the study; theanalysis is based on the 2,301 individuals interviewed inat least two waves.

Dependent Variables. The survey measured a num-ber of dimensions of political culture, including—mostimportantly for our purposes—political knowledge, par-ticipation, tolerance, and the individual’s sense of na-tional versus tribal identification. Knowledge, tolerance,and participation, of course, are core aspects of demo-cratic citizenship, and are of central concern to manycivic education programs in developed as well as devel-oping democracies. We focus on national versus tribalself-identification because the NCEP aimed specifically topromote a sense of Kenyan “nationhood,” given the oftentroubling relationships between ethnic groups that havehampered the country’s development (Ndegwa 1997;Orvis 2001b). In the wake of the explosion of ethnic polit-ical violence following the flawed elections of December2007, the importance of this variable, along with politicaltolerance, for Kenya’s democratic political culture becameclearer still (Barkan 2008).

The specific questions we used to measure each di-mension were similar to those used in recent Afrobarom-eter and other survey-based studies of public opinion,participation, and democratic values in Kenya and sub-Saharan Africa (e.g., Bratton, Mattes, and Gyimah-Boadi2004; Gibson and Gouws 2003). The questions were vet-ted with the Research International survey personnel aswell as with a Kenyan sociologist with extensive surveyexperience, Dr. Paul Mbatia of the University of Nairobi.The questionnaire was finalized after a thorough pretestand interviewer debriefing conducted in January 2002 byResearch International.

We measured Political knowledge through responsesto four questions regarding the names of the Vice

There was no difference in reinterview rates related to workshopattendance, region, income, education, age, religion, church atten-dance, or gender. Interviews were predominantly in Kiswahili andEnglish, with 14% conducted at least in part in one of nine local lan-guages. The interviews were conducted by Research International,a respected Nairobi survey research firm (http://www.research-int.com).


TABLE 1 Sample Design for Longitudinal NCEP Evaluation

Wave 3 April–June 2003 Total RespondentsWave 2 October–

Wave 1 November 2002: Follow-up Follow-up All Two- orFebruary– Follow-up from from Wave 1 from Waves 1 Three-Wave Two-Wave

April 2002 (A) Wave 1 (B) Only (C) and 2 (D) (B+C) Only (B+C-D)Initial Workshop Attendees 1308 901 261 210 1162 952Initial Nonattendees 1303 886 253 191 1139 948Total 2601 1787 514 401 2301 1900

President and the Provincial Commissioner, the lengthof the President’s term in office, and the provisions foramending the Kenyan constitution. We summed respon-dents’ correct answers to create a general knowledge scaleranging from 0 to 4. The reliability of the scale was .50in both the pre- and postworkshop interviews. We mea-sured the respondent’s level of democratic Political par-ticipation by asking whether the respondent had done anyof the following in the past year: worked for a politicalparty or candidate; participated in an organized effort tosolve a neighborhood or community problem; attendeda meeting of the local council or with other governmentofficials; contacted a local official; contacted a nationalelected official; taken part in a protest, march, or demon-stration on some national or local issue; and contacted alocal chief or traditional leader about a problem. Waves 2and 3 questions explicitly called for respondents to reportwhether they had taken part in any of these activities sincetheir previous interview. We sum these responses to createan overall scale ranging from 0 to 7. The reliability of thescale is .74 in the postworkshop interviews, and .70 in thepreworkshop interviews.

We measured Political tolerance with four standardquestions regarding whether atheists and individuals whowant to abolish elections in favor of military rule shouldbe allowed “to speak publicly in your locality” and “toorganize a peaceful demonstration to express their pointof view.” Answers are on a 4-point agree/disagree scale,and we create a tolerance index by averaging responsesto the four questions, yielding a scale that runs from 1 to4. Its reliability was .90 for the postworkshop interviewsand .84 for the preworkshop interviews. To measure Na-tional Versus Tribal Identity, we asked individuals “howimportant is being Kenyan to the way you think of your-self,” with responses coded as 3 for “very important,” 2for “somewhat important,” and 1 for “not important.”The same question was asked for the individual’s “tribeor ethnic group.” We created a composite variable by sub-tracting the individual’s ethnic identity score from his orher national identity score, resulting in a scale ranging

from −2 (greater ethnic identity) to +2 (greater tribalidentity).6

Independent Variables. Exposure to civic education.The primary variable for civic education is based on aseries of questions administered in Waves 2 and 3. As de-scribed in the previous section, we conducted interviewsin Wave 1 with both workshop attendees and with nonat-tendees who were matched to the workshop group on sex,age, educational status, and place of residence. Individu-als in the nonattendee group, of course, could not be pre-vented from seeking out subsequent NCEP workshops ontheir own, and many did so. Some 30% (349 individuals)in this group received civic education at some point dur-ing the study. Moreover, many from the initial workshopsample sought out additional civic education workshopsas well: in Wave 2, 38% of them reported additional civiceducation, while in Wave 3, 30% of them did. We thereforeasked a series of questions designed to measure the extentto which interviewees had been exposed to civic educationsince their previous interview. We asked members of theinitial workshop sample whether they had attended othercivic education workshops since their previous interview,and whether they had attended organized teachings aboutdemocracy and the constitution at their church, mosque,or place of worship. Members of the initial nonattendeesample received the same questions without reference toan initial workshop. For the workshop sample in Wave2, we code Civic education exposure as 1 + the numberof additional workshops or organized teachings. For theinitial nonattendee sample and for individuals who wereinterviewed a third time, Civic education exposure is sim-ply the number of workshops plus organized teachingssince the last interview. Because of the small number ofrespondents attending more than four workshops, and toreduce the right skew of the variable, we truncated theexposure count to produce a variable coded from 0 to 4.

6The four dependent variables are empirically, as well as theoreti-cally, distinct: the maximum intercorrelation between the indica-tors is a very low .18.


Our use of self-reported civic education exposureraises legitimate concerns about the reliability of theseverbal reports. We cannot absolutely validate these self-reports (any more than can most other survey-based stud-ies of political or social participation), but other evidencegathered as part of the larger research project suggeststhat overreporting of attendance at civic education eventswas not common in Kenya in 2002. As noted above, theNCEP’s Technical Assistance Team (TAT) kept detailedrecords of each program activity, including the numberof individuals present. According to these records, the to-tal number of trained individuals was approximately 4.6million, or 23% of the Kenyan voting age population. Thisfigure, of course, does not represent the number of uniqueindividuals the program reached because some portion ofthe population attended more than one event. To estimatethe program’s actual reach, we take advantage of a repre-sentative national survey collected after the program con-cluded in December 2002, in which 16% of the populationreported attending at least one event (Finkel 2003b). Ofthe attendees, 37% reported participating in one event,31% in two events, and 31% in three or more events. If weapply these proportions to the TAT’s estimate of 4.6 mil-lion individuals trained, we arrive at a program reach of2.8 million unique individuals, or 14.4% of the country’svoting age population. The two estimates—16% from thenational sample versus 14.4% from the discounted TATrecords—are strikingly similar. This gives us confidencein the self-reported exposure measures we use here; socialdesirability or other pressures on individuals to overre-port their attendance at civic education events seem tohave been minimal in the Kenyan context in 2002.7

Discussion of civic education. We measured post-civiceducation political discussion by asking all individuals inboth the treatment and control groups: “Setting aside anyevents or workshops that you may have attended person-ally, has anyone you know talked to you about events orworkshops about democracy and the Constitution thatthey have attended this past year?” If respondents said

7We conducted several additional tests of the impact of the self-reported exposure measures on the statistical results we reportbelow. First, the impact of civic education exposure was statis-tically significant for all four dependent variables, regardless ofwhether the “treatment” group consisted of all individuals whoreported some civic education exposure, or consisted only of thoseindividuals in the original workshop sample. Second, a dichoto-mous “treatment” indicator for whether the individual attendedthe single workshop for which we can completely verify attendancewas significant in models predicting political knowledge, tolerance,and national versus tribal identification models, and, as we showsubsequently, significant in the political participation models forindividuals trained in workshops that made use of participatoryteaching methodologies. The results of these tests can be found inthe article’s Supporting Information document (see footnote 1).

“yes,” we asked whether the number of such individu-als was “one or two people” (coded as 1), “three to fivepeople” (coded as 2), or “more than five people” (coded3). Respondents who reported no discussion with civiceducation workshop attendees were coded as 0 on thevariable Discuss others.

Control variables. We also include in the analysis sev-eral variables related to political awareness and engage-ment. We measured Political interest on a scale from 1to 3 as the mean of responses to two questions regardingwhether the respondent has “a great deal,” “some,” or“very little interest” in local community affairs and na-tional politics and affairs. The correlation between the twovariables in both the pre- and postworkshop interviewswas moderate at .43. We measured Media consumption ona scale from 1 to 4 based on the mean of responses to twoquestions regarding how often the respondent pays atten-tion to news about politics on the radio and in newspa-pers. Response categories ranged from “never” to “aboutevery day,” and they were moderately correlated at .47.We measured Group memberships as a count (from 0 to11) of the number of types of formal organizations inwhich the respondent is a member, including churches,burial societies, sports and women’s groups, and businessassociations. Each of these variables was then recoded toa 0 to 1 scale, with the index showing a reliability of .69.We measured General political discussion by responses toa question: “How often have you discussed political issueswith friends, family, or co-workers in the past year [sinceyour previous interview]?” with codes of 0 for “have notdone this,” 1 for “once,” and 2 “for several times.” De-mographic variables such as education (1–9 scale from“no primary schooling” to “completed university”) andan indicator for urban/rural location were measured aswell.

Statistical Methods

Observational, or nonexperimental, evaluations of civiceducation face a number of important obstacles tosuccessful causal inference. Let Y 1i be an individual’sdemocratic orientation (say, “tolerance”) measured if shereceives civic education training, and Y 0i be the same in-dividual’s tolerance measured at the same point in time ifshe does not receive civic education training. The causaleffect of civic education exposure is:

�Yi = Y 1i − Y 0i , (1)or the difference between the tolerance of an individualwho receives civic education training and her tolerance if


she is not trained. The “fundamental problem of causalinference” (Holland 1986) is that, for any given individuali, we observe only one of the two quantities on the right-hand side of (1); that is, we observe only Y 1i for individualswho received civic education, and we observe only Y 0i forindividuals who did not. Thus, estimating the causal ef-fect by comparing tolerance levels for the observed “civiceducation” and the “no civic education” groups is unwar-ranted without invoking additional assumptions.

In cross-sectional studies, the most common assump-tion is that, conditioned on a series of observed X vari-ables, the expected “no treatment” outcome for those whoreceive and who do not receive treatment is the same. Let-ting D be an indicator for civic education exposure, thisassumption means that:

E(Y 0i | X, D = 1

) = E (Y 0i | X, D = 0). (2)

If this assumption holds, then controlling for X, theoutcome for the group which did not receive civic educa-tion would be exactly what the civic education treatmentgroup would have looked like in the absence of treatment.In that case, comparison of the mean outcomes for thetreatment and control groups (adjusted for X via regres-sion or some matching procedure) would yield the causaleffect of civic education for those who received it or the“average impact of the treatment on the treated” (ATT).8

The problem, however, is that the assumption in (2)is likely to be untenable, given the self-selected natureof civic education exposure. There are likely to be myr-iad factors that lead individuals to attend such programsthat also would lead to higher levels of democratic orien-tations in the absence of any civic education treatment.Some, such as education, age, and membership in sec-ondary associations, are likely to be included in X, thevector of observed control variables, but some relevantfactors—for example, democratic or authoritarian per-sonality, motivation, or other idiosyncratic attributes—are likely not to be known to the researcher or included inthe data set.9 The result will be a correlation between thetreatment variable and the unobserved error term of theequations predicting democratic orientations, and henceendogeneity in the outcome equations and potential biasin the estimation of causal effects.

8The ATT provides the most direct answer to the question ofwhether an intervention affected those individuals who were ex-posed to it. See Heckman and Vytlacil (2001) for a discussion ofother impact evaluation parameters.

9We assess the degree to which workshop attendees and nonat-tendees differ on observed covariates in the article’s SupportingInformation document (see footnote 1). There we also show theresults of models that combine the panel methods described be-low with propensity score matching methods to control further forselection effects due to Wave 1 observables (see also footnote 13).

Similar problems of selection biases exist in the anal-ysis of the effects of political discussion on democraticattitudes. The amount of political discussion in which aperson engages is likely to be a function of a number ofusually unobserved personal traits—extroversion, moti-vation, and the like—that may also relate to democraticattitudinal and behavioral orientations. Individuals mayalso seek out congenial discussion networks based onthese same unobserved traits or their own prior politicalattitudes (Conover, Searing, and Crewe 2002; Mutz 2002).Thus, greater tolerance observed among those with higherlevels of post-civic education discussion may be due to thedifferences in their (counterfactual) “no discussion” levelof tolerance, compared to a control group of nondiscus-sants.

The longitudinal design implemented in the KenyaNCEP evaluation allows us to overcome these difficul-ties to a significant extent. With panel data, we observetolerance or some other democratic orientation (Yit) fora given individual i at several t points in time, and wealso observe the individual’s exposure to civic education(D1) and subsequent political discussion (D2) that takesplace between any two given time periods of observation.Following Allison (1994), the task is to estimate the effectof the two civic education–related “events” that may havetaken place between time points t−1 and t on the indi-vidual’s democratic orientations at time t . For two-wavedata, the basic model may be written as:

(a) Yi1 = �1 + �1 Xi1 + (Ui + εi1)(b) Yi2 = �2 + �1 D1i2 + �2 D2i2 + �1 Xi2 + (Ui + εi2)

(3)

where the Xit represent a set of control variables that areobserved in the given data, and the unobserved error termat each point in time is comprised of two components:the idiosyncratic component εit, which varies across indi-viduals and over time, and the Ui term, which representsall unobserved individual-level factors that are related toY and that vary across individuals but do not vary overtime. The absence of D1i1 and D2i1 in equation (3a) re-flects the fact that they are zero for all individuals at time1, which represents the “pretest” before any civic educa-tion or post-civic education discussion took place. If weassume that the idiosyncratic errors εit are unrelated tothe Xit, to Ui, or to one another, we can estimate �1 and�2 by first subtracting equation (3a) from (3b):

(Yi2 − Yi1) = (�2 − �1) + �1 D1i2 + �2 D2i2+�3(Xi2 − Xi1) + (εi2 − εi1) (4)

and then regressing the change score in Y against the Dsand the change score in X. The intercept in this estima-tion will yield the average change in Y for individuals who


never experienced D, i.e., the pure control group, with � 1and � 2 representing the additional average change in Yfor every unit increase in civic education exposure andpost-civic education discussion. Equation (4) is equiva-lent in the two-wave case to the “fixed effects” model,which subtracts an individual time-averaged value of Yand X from equation (3) and regresses “mean-deviatedY” against “mean-deviated X” and time dummy variablesto yield identical estimates (Allison 1994). These differ-ence or “within” estimators provide the causal effects � 1and � 2 while controlling for the selection biases that mayhave been produced by the unobserved Ui, i.e., by stablepersonality or other idiosyncratic factors.

As in the cross-sectional case, however, the unbiasedestimation of causal effects with panel data depends on acritical assumption. We can define civic education’s causaleffect on democratic orientations analogously to (1), thistime in terms of gains:

�Yit =(Y 1i t − Y 1i t−1

) − (Y 0i t − Y 0i t−1). (5)

The causal effect is thus the difference between thegain for individual i at time t and time t−1 if she experi-ences civic education and the gain for the same individualover time if she does not experience civic education train-ing. Again, for any given individual, only one group ofterms in the parentheses in (5) is observed: (Y 1i t − Y 1i t−1)for the treatment group, and (Y 0i t − Y 0ot−1) for the controlgroup. In the longitudinal case, the key assumption foridentifying the causal effect is that, controlling for bothobserved X and unobserved stable U variables that influ-ence both treatment and outcomes, the gain in tolerancethat the treatment group would have experienced in theabsence of treatment is the same as the gain in tolerancethat the control group did experience in the absence oftreatment:

E(Y 0i t − Y 0i t−1

∣∣ X, U, D = 1)

= E (Y 0i t − Y 0i t−1∣∣X, U, D = 0) (6)

The assumption in (6) is weaker than in (2), requiringonly that the counterfactual growth rate, not level, of Yfor the treatment group in the absence of treatment bethe same as the control group.

It may be the case, however, that even this weakerassumption is not tenable. It is plausible that the sameunobserved factors that lead individuals to seek out civiceducation are those that determine subsequent gains intolerance over time, and these gains may have been re-alized in the absence of treatment as well. Disentanglingthe “true” causal effect of treatment from such a differen-tial trend is impossible with two-wave data, but the thirdwave of data gathered here enables us to make progress.Following Morgan and Winship (2007, 269), we may es-timate a model with three-wave data that includes both a

separate time trend for the treatment and control groupsand time-specific treatment indicators:

Yit = � + �1 Dit + �1 Xit + �2 D∗i + �3T + �4(D∗i T) + εit(7)

where D∗i is an indicator for whether the individual wasever exposed to civic education over the course of thestudy, T is a variable measuring the survey wave or time,and all other variables are defined as before. The modelallows the estimation of the causal effect of the time-specific level of civic education exposure (Dit ), control-ling for initial differences between the “ever-treated” andcontrol groups (�2) and controlling for differences inthe trend in democratic orientations between the “ever-treated” and the control group as well (�4). This analysisprovides an additional strong test for the causal effects ofcivic education and post-training political discussion onthe democratic orientations examined here.10

ResultsThe Frequency of Civic Education Exposure:

Direct Effects

In the columns marked (a) in Table 2 we present the resultsof two-wave fixed effects estimation of the direct impactof civic education exposure on our four key dependentvariables: Political knowledge, Political participation, Po-litical tolerance, and National versus tribal identification.The columns marked (b) in that table report estimatesof the same causal effects from three-wave regressionmodels that control for potentially different over-timetrajectories in each democratic orientation for the civiceducation treatment and control groups. As can be seenin the two-wave models, civic education significantly af-fects each dependent variable. The models control forthe individual’s level of political interest, media exposure,group memberships, and general political discussion; forover-time changes in each orientation that are commonto all individuals; and most importantly, for stable factorsthat could have led individuals to select into civic edu-cation and that were also potentially related to the fourdemocratic outcomes. This is strong evidence that adultcivic education programs are able to affect orientationsrelevant to democratic political culture.

Contrary to the findings of previous research, how-ever, the effects of exposure in the two-wave models were

10Models including separate time trend and time-specific treatmentindicators are more robust with longer wave panels; for this reason(and the fact that we have three-wave data on only 401 respon-dents), we report the results of both the two-wave difference andthree-wave time trend models in the analyses to follow.


TABLE 2 Two-Wave Fixed Effect and Three-Wave Differential Trend Models: Civic Education’s Effecton Democratic Orientations

Political Knowledge Participation ToleranceNational vs. Tribal

Identification

Two-Wave Three-Wave Two-Wave Three-Wave Two-Wave Three-Wave Two-Wave Three-WaveFixed Differential Fixed Differential Fixed Differential Fixed Differential

Effects (a) Trends (b) Effects (a) Trends (b) Effects (a) Trends (b) Effects (a) Trends (b)

Civic education 0.120∗∗ 0.117∗∗ 0.106∗∗ 0.100∗∗ 0.129∗∗ .033# 0.071∗∗ .082∗∗

exposure (0.02) (0.01) (0.03) (0.03) (0.02) (0.02) (0.02) (0.01)Media consumption 0.520∗∗ 1.342∗∗ 0.447∗∗ 0.561∗∗ 0.231∗ .455∗∗ 0.090 0.161∗∗

(0.08) (0.06) (0.15) (0.11) (0.09) (0.07) (0.06) (0.05)Political interest 0.048 0.088 0.505∗∗ 0.627∗∗ 0.036 −0.061 −0.067 −0.092∗

(0.08) (0.06) (0.14) (0.10) (0.09) (0.07) (0.06) (0.04)Group memberships 0.441∗∗ 0.591∗∗ 2.972∗∗ 3.235∗∗ 0.103 0.054 −0.181∗ −0.325∗∗

(0.11) (0.07) (0.19) (0.12) (0.11) (0.08) (0.08) (0.05)General political .089∗∗ .209∗∗ 0.181∗∗ 0.263∗∗ 0.027 .063∗∗ .054∗ 0.058

discussion (.028) (0.02) (0.04) (0.03) (0.03) (0.02) (0.02) (0.01)November reinterview 0.318∗∗ −0.690∗∗ −0.232∗∗ 0.122∗∗

(0.04) (0.07) (0.04) (0.03)March-June 0.013 −0.528∗∗ −0.230∗∗ 0.120∗∗

reinterview (0.06) (0.10) (0.06) (0.04)Treatment group −0.045 0.143∗ −0.035 0.050#

(0.04) (0.06) (0.04) (0.26)Time trend 0.024 −0.305∗∗ −0.122∗∗ .079∗∗

(0.03) (0.05) (0.03) (0.02)Trend × treatment 0.092∗∗ −0.039 0.086∗ 0.017

group (0.03) (0.06) (0.04) (0.03)Constant 1.582∗∗ 0.916 0.808∗∗ 0.213∗ 1.730∗∗ 1.646∗∗ 0.174∗∗ 0.175∗∗

(0.08) (0.06) (0.14) (0.09) (0.09) (0.06) (0.06) (0.04)No. of observations 4593 4993 4593 4993 4586 4983 4583 4983R-squared 0.181 0.214 0.166 0.222 0.021 0.018 0.069 0.056

Note: Robust standard errors in parentheses are clustered on 2,301 individuals in all models. Coefficients are significant at #p < .10; ∗p <.05; ∗∗p < .01. R-squared within is presented for two-wave fixed-effect models.

not significantly weaker in the case of difficult demo-cratic values such as tolerance and identity compared withknowledge and participation. The effects of exposure onknowledge, tolerance, and identity are all generally mod-erate in substantive terms, and often rival the impor-tance of the other variables in the model. For example,the effect of each civic education workshop on politicalknowledge is estimated to be approximately .12, compa-rable to the effect of attending to political information inthe mass media for one-and-a-half extra days per weekor moving one additional scale category on the 3-pointGeneral political discussion variable. Only in the politi-cal participation model do more traditional mobilizingfactors such as group memberships and political interestfar outweigh the substantive impact of civic education.Interestingly, group memberships actually work to inten-sify identification with one’s ethnic group, an effect that

is likely due to most groups’ ethnic homogeneity and tothe politicization of ethnicity in postindependence Kenya(Barkan 2008; Ndegwa 1997; Orvis 2001b; Posner 2007).Civic education thus proves to be useful in developingeven difficult democratic values such as tolerance, and incounterbalancing the in-group identifications promotedthrough some other forms of civil society engagement.At the same time, the program appears to have been rel-atively less successful in promoting general political par-ticipation; we attribute this finding to the lack of directmobilization appeals in the NCEP curriculum aside fromencouraging individuals to vote in the 2002 election.11

11We do find significant effects of NCEP exposure on voter turnoutin a separate, two-wave fixed effects model among individualswhose post-test interview was conducted after the December 27,2002 balloting. The results indicate that the turnout rate among


TABLE 3 The Number of People Who DiscussedTheir Own Workshop withRespondent (% of Respondents)

More ThanNo One 1–2 People 3–5 People 5 People

Treated respondentsWave 2 31.79% 12.43% 23.74% 32.05%Wave 3 26.62% 16.17% 27.36% 29.85%

Untreated respondentsWave 2 51.13% 24.03% 17.58% 7.26%Wave 3 57.89% 14.81% 14.04% 13.26%

The results from the three-wave differential trendmodels, shown in column (b) of Table 2 for each vari-able, are remarkably similar to the two-wave estimates,with one exception. For knowledge, participation, andnational versus tribal identification, the magnitude andsignificance of the effects found for civic education arenearly identical to the two-wave estimates. This providesstrong evidence for the robustness of the civic educationeffects, as they hold even controlling for the general ten-dency of those attending workshops to show differentialchange over time on the outcomes, change that may havebeen already ongoing or have been unrelated to the in-dividual’s civic education exposure at any given time. Inthe tolerance model, however, the effect of civic educa-tion drops dramatically after including the differentialtime trend variables, with a (barely) significant effect of.033. This is due, as can be seen, to the pattern of generaldecline over the course of the campaign for the controlgroup and a significantly less pronounced decline overthe course of the campaign for those who were exposedto some civic education training. We conclude that theimpact of direct civic education exposure is robust inthe case of knowledge, participation, and national versustribal identification, and significant, though somewhatmore equivocal, in the case of political tolerance.

The Extent and Effects of Post-CivicEducation Discussion

We turn next to the indirect effects of civic education viapolitical discussions within individual social networks. InTable 3, it can been seen that the potential for indirecteffects from the Kenyan National Civic Education Pro-gramme is substantial, given the number of individuals

treated respondents was 6 percentage points higher than amongnontreated individuals.

who discussed workshops that others attended. Amongindividuals in the treatment group in Wave 2, that is, in-dividuals who attended at least one NCEP workshop be-tween February and November, over half also discussedthe workshop experiences of at least three other individ-uals, and only one-third engaged in no discussion aboutother individuals’ workshop experiences. The figures inWave 3 are even more striking, indicating that treatedindividuals were also very likely to discuss the workshopexperiences of other treated individuals within their so-cial networks. The typical NCEP workshop participantthus discussed the workshop experiences of a good manyother people as well.

But the diffusion of NCEP messages also reached asubstantial number of individuals who themselves didnot attend any of the program’s workshops. Among in-dividuals in Wave 2 with no exposure to civic educationworkshops of their own, approximately half neverthelesshad some discussion with others in their networks whodid attend NCEP activities. In fact, about one-quarterof these ostensibly “control” individuals discussed theworkshop experiences of three or more other “treated”individuals. The figures for Wave 3 are again compara-ble, with over 40% of the “control” group neverthelessdiscussing civic education issues with at least one NCEPparticipant, and 27% discussing the workshop experi-ences of three or more other individuals. This indicatesclearly that the NCEP reached many more individuals viasubsequent political discussions than via formal train-ing. Among this sample of individuals interviewed uponentering NCEP workshops between February and March2002 and a corresponding sample of nonattendees, ap-proximately 80% either attended their own workshop ordiscussed the NCEP workshop experiences of at least oneother individual. There is thus at least significant poten-tial to observe secondary effects of the civic educationtraining.

In Table 4, we show the results of two-wave fixedeffects models and three-wave differential trend modelspredicting the four democratic orientations with bothCivic education exposure and Discuss others. The lattervariable measures the extent to which individuals dis-cussed the workshop experiences of others exposed toNCEP messages. Direct exposure to civic education itselfremains statistically significant in all models in column(a), with these effects supplemented by significant sec-ondary effects of political discussion in all models asidefrom political participation. The more individuals en-gage in discussion of civic education workshops that oth-ers attended, the greater the change in knowledge, toler-ance, and national versus tribal identity over time. Theresults in column (b) strongly support these findings with


TABLE 4 Two-Wave Fixed Effect and Three-Wave Trend Models: The Effects of Civic Education andDiscussion on Democratic Orientations

Political Knowledge Participation ToleranceNational vs. Tribal

Identification

Two-Wave Three-Wave Two-Wave Three-Wave Two-Wave Three-Wave Two-Wave Three-WaveFixed Differential Fixed Differential Fixed Differential Fixed Differential

Effects (a) Trends (b) Effects (a) Trends (b) Effects (a) Trends (b) Effects (a) Trends (b)

Civic education 0.093∗∗ 0.063∗∗ 0.086∗ 0.081∗ 0.101∗∗ −0.013 0.054∗∗ 0.051∗∗exposure (0.022) (0.016) (0.037) (0.032) (0.026) (0.021) (0.018) (0.02)

Discuss others 0.066∗∗ 0.113∗∗ 0.050 0.039 0.069∗∗ 0.095∗∗ 0.042∗ 0.065∗∗

(0.023) (0.014) (0.040) (0.028) (0.026) (0.019) (0.018) (0.14)R-squared 0.185 0.222 0.167 0.222 0.024 0.024 0.071 0.061

Number of 4593 4993 4593 4993 4586 4983 4583 4983observations

Note: Coefficients for media consumption, political interest, group memberships, general political discussion, time of interview, and timetrend are omitted here; see the article’s Supporting Information document for full results (see footnote 1). Robust standard errors inparentheses are clustered on 2,301 individuals. Coefficients are significant at #p < .10; ∗p < .05; ∗∗p < .01. R-squared within is presentedfor two-wave fixed-effect models.

three-wave differential trend analyses. As in Table 2, theimpact of direct civic education exposure on tolerancedrops essentially to zero, and the insignificant effect ofDiscuss others on participation in the two-wave case re-mains negligible in the three-wave model. All other civiceducation and discussion variables remain significant andof roughly equal (and sometimes greater) magnitudecompared to the two-wave case.12 The pattern of effectsin Table 4 confirms the notion that both civic educationexposure and post-civic education discussion have robustand independent effects on important variables related toKenyan democratic political culture, effects that obtainin the context of strenuous controls for the selection bi-ases that confound causal inference in nonexperimentalresearch.13

12We note that these effects obtain in models that control for GeneralPolitical Discussion, which itself has significant effects on three ofthe four dependent variables. To this extent, the results for DiscussOthers indicate the impact of workshop discussions per se, andnot size or intensity of the individual’s normal political discussionnetwork.

13We conducted one final robustness check on the results by es-timating the effects of both Civic education exposure and Discussothers via the propensity score matching methods for multipletreatments outlined in Imai and van Dyk (2004) and Yanovitsky etal. (2005). We first estimated an ordinal logistic regression modelpredicting levels of Civic education exposure and Discuss others withgroup memberships, political interest, media exposure, general po-litical discussion, education, age, gender, rural/urban location, andfrequency of church attendance, all measured at Wave 1. Aftereliminating the relatively few cases that fell outside of the regionof common support (i.e., “control” individuals whose propensitieswere lower than the lowest “treatment” individuals and “treat-

Examination of the relative sizes of the Civic edu-cation exposure and Discuss others effects, as well as thedistributions of these variables in our sample, reveals sev-eral additional points of interest. First, the direct effectsof civic education exposure outweigh the secondary ef-fects. Attending, for example, three civic education work-shops leads to higher levels of knowledge, participation,tolerance, and national versus tribal identification thandoes discussing three other individuals’ workshop expe-riences, other things being equal. But at the same time,discussing the workshop experiences of three or moreother people has a greater effect on knowledge, tolerance,and national identification than does attending only oneworkshop and not engaging in any discussion of others’experiences. Thus, political discussion about civic educa-tion workshops among individuals who were not treatedoften produces more democratic change than formal civiceducation exposure itself.

ment” individuals whose propensities were higher than the highest“control” individuals), we divided the sample into four equal strataon the two propensity scores and verified that, within each strata,the observed covariates were balanced with few minor exceptions.We then ran separate models estimating the effect of civic educa-tion exposure, and then Discuss others, and aggregated the resultsacross strata. In every test, the magnitude of the estimated co-efficient remained nearly identical to those obtained in Tables 2and 4 above, though in one case, Civic education exposure’s effecton political participation, the standard error increased enough toyield an insignificant effect (likely due to the smaller sample sizeswithin strata). The results of these tests and further discussion canbe found in the article’s Supporting Information document (seefootnote 1).


Second, the combined effect of both direct and in-direct civic education effects can be quite substantial.For example, attending four or more workshops and dis-cussing the workshop experiences of an additional fiveindividuals leads to an approximately .70 increase in theindividual’s average amount of Political knowledge, andincreases of about two-thirds of a point on the 4-pointPolitical tolerance and 5-point National versus tribal iden-tification scales, respectively. Taken together, high “doses”of both formal civic education treatment and subsequentdiscussion about the treatment of others can be especiallyeffective in producing democratic change.

Third, it is also the case that many more respondentsare located toward the high end of Discuss others thantoward the high end of Civic education exposure; that is,in this sample more individuals discuss the workshopsof other individuals than attend a high number of work-shops themselves. In Wave 2, for example, only 29% ofrespondents attended more than one NCEP event, whilea full 45% discussed the workshop experiences of three ormore other people. In Wave 3, only 12% of respondentsattended more than one workshop, while more than one-third talked about the workshop experiences of at leastthree other people. This means that the average respon-dent receives in general a greater boost in democratic ori-entations from discussing others’ workshops than fromattending his or her own.

The Conditional Effects of Civic Education:The Role of Participatory Methodologies

According to our theoretical discussion above, the effectsof civic education should depend in important ways onfactors related to the nature of the individual’s civic ed-ucation experience, as well as factors related to the indi-vidual’s previous store of political resources. That is, weexpect to find greater effects when individuals are taughtthrough more intensive, engaging participatory method-ologies. We also expect greater effects on democratic atti-tudes and values among those individuals who have lessprior information and who are less socially integrated.

We measured the extent to which participatory teach-ing methodologies were employed in the first workshopof the initial group of attendees, taking advantage of thefact that interviews were conducted with approximatelyeight different attendees for each workshop. Participantswere asked whether the following took place in that par-ticular workshop: breaking into small groups to discussmaterial; staging plays or dramatizations; playing games;solving problems and developing proposals; role-playingexercises; or mock elections. All six items loaded stronglyon a single factor, with a reliability coefficient (alpha) of

.78. We counted the number of these activities from 0 to 6and averaged these values across all participants in a givenworkshop to arrive at a measure of the kinds of teachingmethodologies employed.14 We then divided workshopsinto those that did not make use of participatory methods(average value of less than 1, 25.4% of all workshops) andthose that did use open, participatory teaching techniques(average value of greater than 1, 74.6% of all workshops),and estimated the impact of attending each kind of work-shop on changes in the study’s four dependent variablesin two-wave fixed effects models.

The results show clear support for the impact of ex-posure to civic education that uses participatory teachingmethods. As can be seen in Figure 1, attending a work-shop that made use of at least one active, participatorymethodology was associated with a significant change ineach dependent variable. Attending a single participatoryworkshop, for example, raised knowledge by a third of apoint on the 4-point scale and increased tolerance by overtwo-tenths of a point on its 4-point scale as well. On theother hand, attending a purely lecture-based workshopthat made use of none of the six participatory methodolo-gies had effects that were statistically distinguishable fromzero only in the case of imparting factual political knowl-edge. For the three other democratic orientations, it wasonly when workshops made use of active methods that anysignificant effects obtained. These results confirm the cen-tral importance of teaching methodologies in generatingattitude and behavioral change through civic education,as individuals develop participatory orientations and sup-portive democratic values through the role playing andother active learning processes that may take place withinworkshops. Moreover, as nearly all of the “participatory”teaching methods involve political discussion—activitiessuch as group problem solving, small-group exercises, andthe like, the findings reinforce the importance of socialtransmission processes in democratic learning as well.

The Conditional Effects of Civic Educationand Discussion: Cognitive and Social

Resources

The effects we have shown thus far indicate much opti-mism for the potential of civic education and post-civic

14Campbell (2008) uses a similar procedure to measure the “open-ness” of U.S. student civics classrooms, but excludes the respon-dent’s own estimation from the calculations in order to guardagainst endogenous self-reports of classroom characteristics basedon the respondent’s own levels of knowledge, tolerance, and thelike. Following this procedure yields a measure that is correlatedat .99 with the measure that includes the respondent’s own as-sessments of participatory methodologies and a virtually identicalpattern of results in models that estimate this variable’s impact.


FIGURE 1 The Effect of NCEP Participatory PedagogicalMethodologies on Democratic Orientations

TABLE 5 Two-Wave Fixed Effects Models: The Interaction of Civic Education and Discussion withEducation and Social Integration

Civic Education Interactions Discussion Interactions

Knowledge Participation Tolerance Identity Knowledge Participation Tolerance Identity

Civic education 0.186∗∗ 0.131 0.053 0.061 0.099∗∗ 0.085∗ 0.102∗∗ 0.063∗∗

exposure (0.049) (0.088) (0.061) (0.042) (0.022) (0.037) (0.026) (0.018)Discuss others 0.070∗∗ 0.049 0.073∗∗ 0.047∗∗ 0.169∗∗ 0.055 0.006 0.106∗∗

(0.023) (0.040) (0.026) (0.018) (0.049) (0.089) (0.058) (0.038)Interactions with:

Group memberships −0.070 0.085 −0.010 −0.138∗ −0.131# 0.049 0.068 −0.223∗∗(0.071) (0.139) (0.096) (0.060) (0.069) (0.136) (0.088) (0.059)

Education −0.225∗∗ −0.182 −0.135 −0.009 −0.212∗∗ −0.131 −0.140# −0.020(0.074) (0.132) (0.089) (0.062) (0.072) (0.132) (0.082) (0.058)

Rural 0.024 −0.024 0.152∗∗ 0.081∗∗ 0.039 0.033 0.125∗∗ 0.048∗(0.030) (0.055) (0.037) (0.026) (0.029) (0.054) (0.034) (0.024)

R-squared within 0.189 0.168 0.034 0.078 0.191 0.168 0.032 0.079Number of 4593 4593 4586 4583 4593 4593 4586 4583

observations

Note: Coefficients for media consumption, political interest, group memberships, general political discussion, and time of interview areomitted here; see the article’s Supporting Information document for full results (see footnote 1). Robust standard errors in parentheses areclustered on 2,301 individuals. Coefficients are significant at #p < .10; ∗p < .05; ∗∗p < .01.

education political discussions to instill democraticknowledge, values, and participatory orientations. How-ever, it remains to be seen whether those effects obtain forall individuals, or whether NCEP training and the subse-quent political discussion within Kenyan social networkshad greater impact on some individuals than others. Inparticular, if these effects depend on prior levels of politi-

cal or participatory resources, the benefits of civic educa-tion and social network discussions may be concentratedamong those who “need” them the least.

We test these possibilities by creating interactionterms between Civic education exposure and Discussothers and three variables measuring cognitive resourcesand social integration: Rural (versus urban) status and


Wave 1 levels of Education and Group memberships. Wethen ran separate two-wave fixed effects models (simi-lar to model (a) in Table 1, including all time-varyingcontrol variables) for Civic education exposure and Dis-cuss others, along with each interaction variable to testfor conditional effects. The results of these “bivariate”interaction models are suggestive: to the extent that dif-ferential effects exist, civic education and especially sub-sequent discussion appear to benefit those less integratedinto secondary groups, those with lower levels of edu-cation, and rural respondents. Out of the 12 estimatedinteraction effects (three cognitive and social resourceson four dependent variables), five are significant for Civiceducation exposure and seven for Discuss others, with all ofthese effects showing greater impact on those with greater“need.” Only for participation do we find that neither theimpact of civic education nor subsequent discussion iscontingent on respondent characteristics.

In Table 5, we show the multivariate estimation ofthese conditional effects models. Of course, relatively highlevels of collinearity between all of the interaction termsmake these models somewhat fragile. Nevertheless, whensignificant interactions exist, they all indicate that Civiceducation exposure and Discuss others are more effectivein producing democratic change among individuals withlower levels of education and secondary group member-ships, and among rural residents. The estimated effectof civic education workshops on political knowledge isstrongly concentrated among those with lower levels ofeducation, and its effect on tolerance and national versustribal identification is more than twice as large in rural ar-eas as in urban ones. The conditional effects of postwork-shop discussion are at least as strongly clustered amongthose with greater “need” as well. Discussion has signifi-cantly greater effects on knowledge and tolerance amongthose with less education, significantly greater effects onknowledge and national identification among those lessintegrated into secondary associations, and significantlygreater effects on tolerance and identification among ru-ral respondents. The results, then, offer qualified supportfor the “compensation” hypothesis: civic education andpost-civic education discussion disproportionately bene-fit those who are disadvantaged in education and socialresources, enabling them to acquire knowledge and val-ues their more advantaged fellow citizens may acquire inother ways.

Conclusions

The analysis of the impact of the 2002 National KenyanCivic Education Programme yields several important

findings with significant implications for theories of civiceducation, democratic learning, and the effects of polit-ical discussion and social networks. First, we have pro-vided perhaps the most conclusive evidence to date thatexposure to adult civic education training can matter forthe development of democratic knowledge, values, andparticipatory orientations. Unlike previous civic educa-tion evaluations, the design implemented here includeda pretest, comparison of individuals who attended civiceducation training and matched individuals not attend-ing the given training activities, and statistical controlsvia fixed effects and differential trend models for possibleselection effects due to (stable) unobservable variables.The results indicate that adults trained in NCEP activitiesshowed significant increase in political knowledge andparticipation, and in such critical democratic values asthe sense of Kenyan versus tribal identification and polit-ical tolerance. Civic education, especially when it is con-ducted with active, participatory teaching methodologies,can work, and it can have direct and immediate effects onstrengthening democratic attitudes and heightening po-litical awareness among adults in new democracies. On atheoretical level, these findings reinforce the message thatdemocratic norms, values, and behaviors can change overfairly short time periods. On a policy level, they shouldencourage donors seeking to promote positive change indemocratic culture in recipient countries.

Second, the analysis has shown for the first time com-pelling evidence of widespread indirect or secondary ef-fects of civic education training through postworkshoppolitical discussions. Individuals who were trained inNCEP workshops went on to talk about their experi-ences with many others in their social networks, leadingto large numbers of Kenyans—even those not directlytrained in NCEP activities—being exposed to civic edu-cation messages via discussions with network partners.Moreover, these discussions had a significant and sub-stantial positive impact on all of the democratic orien-tations we examined aside from political participation.Again, these effects obtained from statistical models thatcontrol for possible endogeneity due to stable individualcharacteristics related both to the likelihood of engagingin postworkshop political discussions and to democraticattitudes, knowledge, and participation. The results thusprovide strong support for a two-step process of socialdiffusion of democratic messages through civic educa-tion.

Taken together, the findings suggest that the potentialimpact of adult civic education on strengthening demo-cratic political culture in transition societies is far be-yond what has previously been estimated. The NationalCivic Education Programme in Kenya was an extensive,countrywide effort consisting of over 50,000 activities


coordinated by dozens of implementing NGOs over a14-month period in 2001–2002. As noted above, we es-timate that these workshops trained between 14% and16% of the Kenyan population. But the reach of the pro-gram from secondary effects was far greater. Our datasuggest that some 40% of “control” individuals never-theless discussed the workshop experiences of others, aslightly higher figure than was reported in the 2003 na-tional survey. This means that somewhere between 40%and 50% of all Kenyans were exposed in some way to civiceducation messages during the run-up to the 2002 elec-tion, the vast majority consisting of discussions betweenfriends, family members, and others in the trainees’ so-cial networks. Given the significant effects estimated herefrom both direct and indirect exposure, it is clear that theNCEP had a nontrivial impact on the aggregate level ofKenyan democratic political culture. Such secondary ef-fects need to be taken into account in future assessmentsof the impact of civic education and other short-termdemocracy assistance interventions undertaken by inter-national donors.15 And given that other emerging democ-racies face many of the same difficult conditions as arefound in Kenya—high economic inequality, poverty, eth-nic tensions, and often poor political performance—thefindings suggest that large-scale, coordinated civic edu-cation programs such as the NCEP, along with the socialdiffusion processes that they are likely to generate, may bea generally promising short-term means of strengtheningdemocratic attitudes and heightening political awarenessamong adults.16

The results also have a number of implications forthe burgeoning literature on political discussion and de-liberation effects in new democracies. Both workshopexposure and post-civic education training have “com-pensation effects” on knowledge and democratic values,with greater impacts being observed among those whoare less socially integrated, either through civil society or

15The possibility that social interventions may have impacts onindividuals or communities beyond those directly targeted in spe-cific programs is being given increased attention in other fields ofdevelopment assistance as well. See, e.g., Angelucci and De Giorgi(2007) and Kremer and Miguel (2007).

16It is possible that the positive effects seen here were conditionalon the political climate at the time in Kenya, which, as noted above,was guardedly optimistic in anticipation of the 2002 elections anda possible victory of the democratic forces opposing President Moiand his KANU administration. Whether and how the immediatepolitical context conditions the receptivity of citizens to democraticmessages should be important topics for future research; prelimi-nary results from a follow-up study of the second NCEP programin Kenya in 2007 (Finkel and Horowitz 2010) suggest, however, thatorganized and well-implemented civic education programs can stillhave some impact on civic attitudes in even more difficult politicalconditions.

through the urban milieu, as well as those with lowereducational levels. This is a promising result, temperedonly by the finding that the amount of post-civic edu-cation discussion is greater among those belonging tomore secondary associations and possessing other politi-cally relevant resources. Still, prodemocratic messages aremuch more likely to reach such individuals through po-litical discussion than through direct exposure to civiceducation activities. The pattern of results thus suggeststhat discussion and political talk can be particularly im-portant mechanisms for the development of democraticvalues among the socially and cognitively disadvantagedin new democracies.

The results have clear implications for the implemen-tation of future civic education programs in transitionsocieties. Most obviously, there is strong support for theenterprise of adult civic education itself, given the reachand impact of organized and coordinated efforts such asthe Kenyan NCEP. It is also clear that such efforts mustbe conducted with active, participatory methodologies,as our results suggest that lecture-based civic educationhas virtually no impact on democratic attitudes, learning,or participatory orientations. And the fact that so muchof the overall impact of civic education stems from post-training discussions also implies that programs shouldencourage participants to talk about their experiences af-ter the fact. This may be done in many ways, rangingfrom role-playing exercises during the training to pro-viding trainees with pamphlets, books, or other materialsthat they can share with family, friends or others in theirsocial networks. Further, the more that individuals areencouraged to speak to others outside of their immediatesocial networks, in particular to those with lower levels ofpolitical or organizational involvement, the more likelyit is that programs will extend their overall reach andexert secondary effects on individuals who themselvesdid not participate in any training. Given that discussioneffects are stronger among individuals with fewer con-nections to existing civil society organizations and withlower educational levels, the gains from doing so in termsof democratic change can be high.

Finally, the positive findings here regarding civic ed-ucation must be tempered with the recognition that thereare limits to what even an extensive civic education effortlike the 2002 NCEP can achieve in developing demo-cratic societies. Successful democratic transitions dependon a good many other factors aside from a supportivemass political culture, most importantly elite behaviorand the crafting of institutions that can ameliorate ethnicand other potentially destabilizing social cleavages (Dia-mond 1999; Posner 2007). And, as we have stressed, massopinion in many developing contexts such as Kenya is


characterized by high levels of political intolerance andintense ethnic identifications, creating relatively difficultbaseline conditions in which programs such as the NCEPmust operate. The combination of reckless elite behaviorand mass intolerance came

Documents

Civic Education, Political Discussion, and the Social ......Civic education first affected the knowledge, values, and participatory inclinations of individuals directly exposed to