The Context

Descriptive statistics from Afrobarometer data show that in post-war Liberia, women’s participation rates lag behind the rates of their male counterparts on measures such as voting, community involvement, and associational life, as well as on such attitudinal indicators as interest in public affairs or political efficacy.Footnote ² For instance, when it comes to actual political behaviours, men outpaced women in participatory measures. A notable 78.3 per cent of men feel they can freely join political organizations, a freedom that only 69.2 per cent of women feel they have (See Figure SI 3). A similar pattern exists with respect to attitudinal measures. Men tended to engage in political conversations more frequently, with 18 per cent reporting regular discussions compared to 8 per cent of women (Figure SI 2). Interestingly, in the realm of political efficacy, women seem to exhibit a stronger resolve to question leadership, diverging from men’s tendency to show deference to authority. These gaps are notably wider in rural areas than in urban areas. Key barriers contributing to these gaps include low access to political information owing to lack of control over information media such as radio, and low usage of mobile phones and internet access. For instance, according to DHS data, the proportion of women owning information sources is low, with merely 36.6 per cent of women owning radios and 34.5 per cent owning cell phones (See Figures SI 9 and SI 10). Afrobarometer data also indicates similar findings: fewer than 53.6 per cent of women have ever used mobile phones, and more than 90 per cent of women lack access to the internet (See Figure SI 11 and SI 12), even for those with primary education or higher.

Over half the women surveyed by the United Nations (2014) reported either not having access to quality information or were uncertain whether they had access: radio was identified as the most frequent source of news and information about politics.Footnote ³ The Carter Center likewise concluded that Liberian ‘women do not access information as easily or as frequently as do men’.Footnote ⁴ Liberia typified postwar societies: a low information environment with poor media infrastructure and limited media choices, along with a lack of in-depth reporting, source transparency issues, and a tendency to align with government narratives (Fair Reference Fair2013; Spurk, Lopata and Keel Reference Spurk, Lopata and Keel2010). Liberian women have been targets of misinformation and disinformation, especially during electoral campaigns (Adams Reference Adams2016; Anderson, Diabah and Afrakoma hMensa Reference Anderson, Diabah and Afrakoma hMensa2011).

Increasing access to political information (qualitatively and quantitatively) has been integral to postwar peacebuilding efforts.Footnote ⁵ Whether these efforts have translated into narrowing gender gaps in media consumption and political participation is our still-open empirical question.

Intervention and hypotheses

Established in 2003 at the end of Liberia’s 14-year civil war, UNMIL Radio is a popular information source with the widest radio coverage in the country. It filled an important media void and was perceived as the most objective and trustworthy information source.Footnote ⁶ As one of the Radio’s first producers put it: ‘Coming out of a war in which the local media had been a part, the people of Liberia were desperate for a voice they could trust’.Footnote ⁷ While UNMIL Radio enjoyed widespread listenership and high credibility, listenership levels have been uneven, with women generally having less opportunity than men to access and engage with its programming (Mvukiyehe and Samii Reference Mvukiyehe and Samii2017, Reference Mvukiyehe and Samii2010). The ‘access to information’ pilot programme sought to overcome this informational barrier to women voters and enable equal participation in the political process. The intervention (i) provided women in selected communities with a radio and (ii) organized them to participate in a series of weekly group listening sessions to UNMIL Radio’s special programmes on elections.

The information component of the intervention focused on five special election programmes.Footnote ⁸ Some programmes, such as Dis Voting Ting, provided civic education about the importance of political participation, while others, such as Back to the Ballot Box, informed voters about the voting process (for example, how to correctly cast the vote or the importance of secret balloting). Political Happenings and Hot Seat brought voters information about political campaigns and served as a forum for political aspirants to debate their agendas. Finally, Around Town was an open mic programme allowing voters from different towns to express their opinions about the upcoming elections and their expectations from aspirant leaders.

Listening groups of women were set up in randomly allocated treatment communities to isolate programme effects. Participants met three times per week to listen to a specific electoral programme in real-time, followed by a discussion facilitated by trained staff from the implementing partner organization. For the fourth weekly meeting, groups were joined by staff from the implementing partner to listen to and discuss a pre-recorded programme. Implementing staff recorded a range of information, including the number of women and men attending the sessions, topics discussed, participants’ opinions and reactions, and level of enthusiasm for engagement with the topics. Each session lasted two hours on average. In this bundled treatment, combining radio provision, information provision, and to some extent facilitated group – each of which could have an independent effect – I focus on the information provision treatment, conceptually aligning this study with other civic and voter education interventions (for example, Hyde, Lamb and Samet (Reference Hyde, Lamb and Samet2023), Finkel and Lim (Reference Finkel and Lim2021), and Mvukiyehe and Samii (Reference Mvukiyehe and Samii2017)), where the treatment of interest is the political information being transmitted rather than the medium of transmission.Footnote ⁹

Importantly, while the programme was explicit in its goal of focusing on women, men were not explicitly excluded or discouraged. Though the discussions were largely driven by issues and concerns prioritized by the women, in most villages a relatively high proportion of men (20 per cent on average) also attended the sessions. Each intervention consisted of at least fifteen listening sessions over a period of five weeks. The partner organization monitored programme implementation weekly, in addition to their role in facilitation.

UNMIL Radio was by far the most trusted news source compared to other local media outlets, deemed the most reliable and detailed in its coverage of political happenings and news; community radio was a distant second choice, attracting one listener for every five UNMIL radio listeners.Footnote ¹⁰ UNMIL Radio’s trustworthiness matters as a key driving factor behind its popularity with the Liberian public – both women and men – especially on electoral issues (and general political issues), making it the most appropriate medium for this treatment.

Identification strategy

The study targeted forty rural communities, with an average village size of around 1,500 households. I used a matched pairs design as a basis for selection. First, the local partner organization selected eighty communities from two counties (Bomi and Rural Montserrado) deemed potential beneficiaries of the programme. The research team then carried out a community assessment survey in each of these to gather demographic and socioeconomic statistics based on the 2008 census by the Liberia Institute of Statistics and Geo-Information Services (LISGIS), a government agency. This information was used along with the staff’s own knowledge of these areas to create community pairs that were similar in terms of geographic and socioeconomic characteristics such as population size, number of adult women, quality of dwellings, access to main roads, infrastructure quality, ethnicity, religion, and economic activity. The radio programme was assigned to one community in each pair by rolling a die. Figure 1 depicts the forty study towns by treatment status.

Figure 1.

Study Locations.

Outcomes measurement and data

Following the empirical literature, I define political participation as activities by citizens that are aimed at influencing the selection of governmental personnel and/or government policy (Mvukiyehe Reference Mvukiyehe2018; Verba, Schlozman and Brady Reference Verba, Schlozman and Brady1995). This definition is broad enough to encompass both formal/institutional forms of participation such as voting or involvement in electoral processes and less formal/institutional forms of participation such as participation in community meetings or contacting traditional leaders (Gottlieb’s Reference Gottlieb2016; Coffé and Bolzendahl Reference Coffé and Bolzendahl2010). Modes of political participation used in the survey were organized around dimensions established in the empirical literature (Mvukiyehe and Samii Reference Mvukiyehe and Samii2017; Mvukiyehe Reference Mvukiyehe2018; Bratton Reference Bratton1999):

• The voting mode, encompassing voting in the 2011 election;

• The campaigning mode, including contributions in time and/or money to a political campaign and political rally attendance;

• The contacting mode, including contact with one’s representative in the national assembly or senate, or contact with a government official or agency;

• The communing mode, encompassing community meeting attendance, speech and the contacting of community leaders;

• The protesting mode, encompassing a wide range of variables such as political demonstrations or calling into radio stations to complain about public officials.

The study estimates programme effects on each of these modes of political participation (as well as their constitutive indicators) and on the overall measure of political participation that combines some or all these dimensions.

Sample and data collection

Data collection targeted all forty villages in the sample (twenty each of the treatment and control villages). Given that the unit of treatment in this study is the village, my sampling frame encompasses all eligible female and male voters in each village. In each village, I conducted interviews with a sample of fifteen households – ten with female respondents and five with male respondents – which I estimated to be adequate for the analysis.Footnote ¹¹ The total target sample comprises 400 females and 200 males. The response rate was an astonishing 97 per cent – a feat achieved thanks to flexible scheduling designed around women’s availability – providing a strongly suggestive technique for avoiding bias in who gets treated and who gets surveyed.

Household surveys served as my main source of outcome data.Footnote ¹² Additionally, observable (behavioural) measures of many outcomes were included as a cleaner indicator of individuals’ propensity to participate (for example, sending a message to political leaders or signing a petition calling on presidential candidates to abstain from election violence), which are designed to send a costly signal (Collier and Vicente Reference Collier and Vicente2014). I developed the survey instrument in collaboration with LISGIS, the implementing agency, and refined it through several rounds of pilot testing and field testing. I employed the same survey instruments in both the female and male samples. Survey interviews were conducted by experienced, professional enumerators from LISGIS trained by the Research Team and for the most part, were delivered in Liberian English, which most respondents understood and spoke.

Fieldwork for both rollout and data gathering took place from August–December 2011.Footnote ¹³

Estimation strategy

The programme was successfully rolled out across all twenty targeted villages. Attendance levels were relatively high throughout the fifteen facilitated sessions available over a five-week intervention period.Footnote ¹⁴

The main analysis focuses on the intervention’s Intent-To-Treat (ITT) effects, thereby considering all adult individuals in treated communities as programme beneficiaries, regardless of whether or the extent to which they participated in the programme (to wit, weekly listenership sessions). Having achieved balance (see SI 2E), I fit regression models of outcomes on an indicator variable for the intervention. I estimate programme effects on both individual outcome indicators and on the ‘overall’ aggregate measure of political participation. For the latter, an index measure was calculated by averaging across standardized means of individual outcome indicators. I estimate the following linear regression:

(1) $${Y_{iv}} = {\beta _0} + {\beta _1}{\rm{UNMIL\ Radi}}{{\rm{o}}_v} + {X_v}\delta + {\varepsilon _{iv}}$$

Where:

• ${Y_{iv}}:$ is the overall standardized participation score for individual i in village v. This measure has two additional variations – one that captures overall political participation through institutionalized forms (for example, voting, attending political rallies, etc.) and less institutionalized forms (for example, attending village meetings or contacting traditional leaders.)

• $UNMILRadio:$ is a dummy variable indicating whether the village received the intervention or not.

• $X:$ represents matched pair fixed effects. I control for this to allow for the possibility that there are differences between pairs that influence political participation.

• ${\varepsilon _{iv}}$ is an error term assumed to be clustered at the village level.

To avoid over-rejection of the null hypothesis due to multiple comparisons, I also estimate Average Standardized Treatment Effects (ASTEs) following the approach pioneered by Kling et al. (Reference Kling, Liebman and Katz2007).Footnote ¹⁵

For marginal effects on individual outcome indicators, since most of these are dichotomous, I estimate a probit model with a simple regression equation of the following general form:

(2) $$Pr({Y_{it}}) = {\beta _0} + {\beta _1}{\rm{ UNMIL\ Radi}}{{\rm{o}}_t} + {X_t}\delta + {\varepsilon _{it}}$$

Where: $Pr({Y_{it}} = 1):$ is the probability of person $i$ in town $t$ engaging in the relevant political activity. Each of the outcome indicators of political participation is binary and takes on the value of 1 if the person is engaged in it and 0 otherwise. All other terms are defined as in equation 1.

Results

The primary focus in this analysis is on the UNMIL Radio programme’s effects on women’s political participation; this will be the locus of the discussion, while the effects of the programme on gender gaps are ascertained indirectly by way of comparing outcomes for the two genders. Figure 2 presents the mean effects of UNMIL Radio on women’s (and men’s) political participation estimated with OLS regressions. Figure 3 presents marginal effects on individual outcome indicators from probit estimations.Footnote ¹⁶ I also check the robustness of my findings by controlling for unbalanced (pre-treatment) covariates.Footnote ¹⁷

Figure 2.

Effects on political participation indices.

Figure 3.

Effects on political participation outcomes.

As can be seen in Figure 2, the intervention has positive effects on the overall political participation of women, with an estimated mean effect of 0.21 standard deviations, statistically significant at the 1 per cent level. The effects are virtually similar for more institutional/formal and less institutional/informal forms of political participation. These effect sizes represent a relatively large difference in participation rates between communities that did and did not receive the intervention. Further, from Figure 3, it appears that these effects are mostly driven by voting, participation in political rallies, protests, attendance, and speaking at local community meetings unrelated to the intervention.

I estimate a 5 percentage point increase in the attendance of community meetings and a 10 percentage point increase in the probability of speaking out in such gatherings, the measures that displayed the largest gender gaps initially. These results are significant at the 5 per cent levels or better.

By contrast, the results on the men’s sample show no evidence of the programme’s positive effects on overall levels or individual outcome measures of participation. If anything, results on a few individual level outcomes – in particular political rallies attendance, contacting government officials and traditional leaders, and attending and speaking at community meetings – show relatively large negative and statistically significant effects on men but, again, the overall mean effects are null. The results on men are not unexpected because they were not specifically targeted by the programme and suggest that, contrary to findings in other studies (for example, Gottlieb Reference Gottlieb2016), the presence of men in listenership sessions did not have perverse effects on women’s political participation. Taken together, these results suggest that even short-term and logistically less demanding broadcast interventions enhance women’s political participation without necessarily increasing or taking away from men, thereby contributing to the narrowing of gender gaps in political participation.

The study also looked at the heterogeneous effects on several dimensions, including marital status, age, schooling, employment and income, and by information sources.Footnote ¹⁸ Overall, the results, while tentative, do suggest the programme effects were concentrated among women with low information at the baseline (for example, those with no formal schooling, unpaid work, below median income, etc.) Finally, I also investigate potential geographic spillovers. I do not expect significant spillovers because the villages are relatively distant from one another (the minimum and median distance between two pairs of treatment and control villages is around three and twenty-two km respectively), and accessibility is difficult.Footnote ¹⁹ Even if spillovers were present, this would suggest my estimates underestimate the true effects. Out of thoroughness, I performed a formal analysis for geographical spillovers (See SI 9 for detailed method and estimates): the programme effect is robust to the inclusion of nearby treated or control households in the female sample.