Project Background

Many funders and storytellers work to shift the cultural narratives that condition public understanding and constrain institutional action on issues like racism, immigration, and climate. Oftentimes, however, this work lacks a shared understanding of what narrative is, how to identify and characterize it reliably, and how it connects to people with the power to make change.

In this project, Harmony Labs aims to solve some of the technical, definitional, and practical challenges that bedevil narrative change research and measurement. The project incorporates the work of many sectors and organizations. And it is already producing tools—like this website—to support ongoing narrative change efforts, as well as industry-grade data infrastructure to identify, measure, and track narratives over long time scales, and connect them to people. We call this data infrastructure the Narrative Observatory, obi, for short.

So far, with initial funding from Bill & Melinda Gates Foundation, we are focused on the narratives of poverty and economic mobility in the U.S. We are looking to expand to additional issues and geographies.

Email us to start a conversation.

What’s a Narrative?

People make and reflect culture by telling stories. Patterns exist in the stories people tell, even when those stories pertain to different social issues or topics. For example, stories on the topic of healthcare, food, or homelessness may all carry the same basic view that “people in poverty are damaged.” These story patterns we call narratives.

Narratives can be reliably identified, tracked and measured across media types over long time scales using a combination of computational text analysis and human reading and annotation. They can also be connected to the people or audiences who consume, create, and amplify them in culture. Well, at least that’s the challenge we’re working through now. :)

Once narratives and audiences have been identified or mapped, then we can start working to understand how to produce change.

Method

As with all the tools we create, this website strives to present as simply as possible only what is useful. Keeping our findings short and sweet is really hard for us, because we’re data nerds. Behind everything you see here is some fancy data dancing, which we’re eager to share.

Audiences

Audience segmentation is pretty standard stuff. We’ve spiced it up with some behavioral cultural consumption data that allows us to build audience profiles based primarily on values, secondarily on poverty attitudes, and enriched with media engagement habits.

We started by reanalyzing some existing audiences created by GOOD. This gave us baseline survey questions and working audiences.

Next, using demography and geography, we fuzzy-matched the working audiences with behavioral data capturing online and television habits of 50,000+ Americans and read media for each audience related to poverty, COVID, and BLM. To validate what we found, we duplicated this process on a second media consumption data set. These readings informed additional survey questions, connected to core values and attitudes for race, gender, place, and class. By combining values—It’s more important to me to be helpful than successful—with poverty beliefs—Poor people manage their money well—we were striving to capture in our audiences both poverty perspectives and information to predict how each audience is likely to respond to COVID, BLM, or whatever other crises 2020 brings.

Then we conducted a voter-file matched survey with ~2,900 respondents, and analyzed survey results to update our audiences and generate an 8-question audience classifier.

Finally, we re-read the behavioral data regarding online and television consumption to enrich each audience profile and associate them with poverty narratives in news and Twitter. So all the sample Pinterest pins, streaming videos, tweets, news articles, and other media are actually derived from data.

Narratives

We started with text from online news, Twitter, music, and open-ended responses to multiple surveys, between January and October 2020. This gave us millions of documents to process and organize in a way that yielded readily to human and machine analysis.

The first layer of analysis separated out poverty-relevant content, using keywords and human annotation to build supervised relevance models for each media type. We considered a document poverty-relevant, if it told us what people experiencing poverty or financial wellbeing are like (hard-working, virtuous, impure); or what it is or feels like to experience poverty or financial wellbeing; or how people go from poverty to financial wellbeing or vice versa.

Then, our human analysts read a randomly generated poverty-relevant sample of each media type, in order to extract key dimensions for story pattern variations. At the same time, we used machines and natural language processing to capture and cluster naturally occurring story patterns. The outputs of both human and machine analyses were used to generate a preliminary narrative structure for news and Twitter. This structure was used by a team of annotators to code another sample of poverty-relevant news articles and tweets. We built supervised narrative models from these annotations. We use these models to predict which narratives each article, tweet, or song is associated with.

In the charts and graphs used throughout this site, numbers may not add up to 100% because of rounding.

Funders, Partners, Friends

Many thanks to everyone who makes this work possible. You’re the best!

Funders

Partners

Friends

About Harmony Labs

Harmony Labs has been doing narrative analysis for more than a decade. One of the first papers we co-authored looked at fracking narratives in documentary film. Since then, we’ve worked on narratives for climate, gun violence, political corruption, artificial intelligence, and other issues. With the Narrative Observatory project, for the first time ever, we’re harnessing powerful industry relationships and an academic research network to develop data infrastructure purpose-built for narrative identification and tracking over long time scales, across media types. This website is one of the first Narrative Observatory outputs.

Harmony Labs builds communities and tools to reform and transform media systems. Our mission is to create a world where media systems support healthy, democratic culture and healthy, happy people. Learn more.

Email us to start a conversation. Follow us @harmonylabs on Twitter and LinkedIn.