Radio forums, both national and local, are powerful tools for discussing issues affecting people’s lives. Many listeners participate using SMS texting – a rich source of data which, if analysed with care and using sophisticated techniques, can provide invaluable insights into changes in the audience’s beliefs and opinions. These changes can be readily linked to socio-demographics (i.e. gender, age, location) collected by short SMS surveys.
Radio discussions can be thought of as large-scale focus groups – where diverse communities participate and different ideas flow. In these forums we can identify collective beliefs by comparing opinions across social boundaries. Adding a temporal dimension allows us to see how ideas change over time within and between social groups.
We understand that ideas are formed within a social milieu, shaped by social identifications and social norms. The core ideas we identify reflect shared worldviews. But there is also space for innovation; some voices are unique and can be powerful enough to contest widespread ideas. So we delve into the data to understand not only trends as well as outliers by bringing back unique voices at all stages of the analysis.
What are collective beliefs?
Groups in society adopt and form common-sense theories about their social and physical worlds. For example, stereotypes about other groups, beliefs about causes and treatment of diseases, or expectations about the roles of mothers and fathers in their children’s upbringing. Often these collective beliefs are acquired through interactions with others, through communication, or learned implicitly by social norms and rituals that can both persist and evolve across generations.
Shared mental models lie at the basis of beliefs, norms and practices, as people use them to make sense of the world by filtering information and interpreting situations (see World Development Report, 2015). Changing practices can be difficult to achieve only by passively transmitting knowledge and changing attitudes of a few individuals.
The power of collective beliefs
The digital and data revolution in sub-Saharan Africa presents new opportunities for development and governance actors to listen intelligently to communities and understand collective beliefs. Insights gathered can inform the creation of more effective, responsive and impactful public good and services. Africa’s Voices helps to create a channel – using media forums, mobile technology and digital platforms – through which citizens are reached and their world views can be understood.
For example if we’re working with an NGO who wants to identify, and then shape, the social beliefs toward polio vaccination, we can look at the ideas that distinguish people who decided to, and who decided not to, vaccinate their children (given equal access to vaccines). We might probe into whether certain beliefs about vaccination, such as misconceptions about side effects or distrust of ‘foreign’ medicine, hinder vaccination uptake. Insights into the language used by those who uptake vaccination provision can be used to shape new engagement strategies.
How do we analyse messages?
After we have collected messages from target audiences, we apply our unique analysis techniques. We start by identifying domain (or topic) relevant ‘seed words’ that are the most frequent in the data. Then we pair them with associated words that are also meaningful within the domain. Some of these words are subsequently paired with others through an iterative process that stops when all the associations are identified and represented in semantic networks. Native speakers from the target communities are key during this initial stage to select word associations, label ideas, pinpoint relevant expressions, and decode humour and sarcasm.
The insights of our analysis result from an interplay between the word clusters and the raw opinions expressed in the messages. Our analysis is:
- Iterative, as from initial seed words (e.g. most relevant words in data), other words are found based on association rules, and from these words others are found forming clusters of words;
- Selective, as word associations as well as coherent clusters of words, relevant within a certain domain, are selected;
- Supervised, as in all stages native speakers decode, interpret and select relevant inputs and outputs;
- Exploratory, as native speakers and researchers interact with raw messages and derive hypotheses linked to concrete examples;
- Emergent, as higher order findings (e.g. collective ideas) are abstracted from word associations whose meaning is linked to the conversational context.
What makes our analysis valid?
Africa’s Voices data is skewed, reflecting the reality of voices. Men, younger and more educated people are more likely to engage in radio discussions through mobile phones or to use social media platforms. But voices of those at the bottom of the pyramid can also be heard. Therefore we can identify these biases and compare groups.
The richness of voice is immensely valued in our analysis, at the expense of statistical generalisations. Because we analyse the full data from participants (often big, messy and unstructured data) representativeness is not the main criterion to assess the validity of our findings. Participants’ voices echo the reality of radio discussions, and they are influential to a larger group of people that listen to radio shows (i.e. 70-90% of the population).
Credible findings are then related to the robustness of our methodology coupled with meaningful data (gathered in a real context). What really matters is that people who participate in the discussions are diverse and heterogeneous, making it possible for us to identify collective ideas from different social groups, and how they change over time. Unlike surveys that “count heads” and frame questions upon researchers’ ideas of reality, our methods allow unexpected insights to emerge and interpretations to be grounded in alternative, and actual, social realities.
In our approach we “learn by doing” as new projects usually require tailored techniques and new tools. As human knowledge is central to our approach, we build resources for language interpretation (for example stop-word lists or idiomatic expressions for African languages) so new projects can benefit from previous ones. Using carefully designed methods, we learn what works and does not work, for example, when engaging audiences or working with new organisations.