Introducing FEE’s Media Analytics Platform (aka Jellyfish)

What is the FEE Media Analytics Platform?

The FEE Media Analytics Platform is a custom-built analytics tool that gathers data from Facebook, Facebook Ads, YouTube, and Google Ads and provides unified media performance insights. Using this platform, we’ve been able to tag videos with descriptive metadata, psycho-graphic factors and communications frameworks and generate predictive models to optimize FEE’s messaging. The platform also allows comparative persona analysis to understand our effectiveness in engaging different personas.

Who is it for?

We’ve started to present insights from the platform at national conferences such as SPN, and are seeking feedback and perspective on further research directions.

Insights from the platform are presented in a user-friendly web format, a self-serve Tableau dashboard, and as a python data science framework that consumes structured data generated by the platform.

Current Research Directions

We reviewed the following frameworks/theories:

• SCARF Model David Rock, 2008
• Moral Foundations Theory Jonathan Haidt, Jesse Graham, et al., 2004
• The Three Axes Model of Politics (The Three Languages of Politics) Arnold Kling, 2013
• The Five Factor Model (OCEAN), Lewis Goldberg et al., 1980
• FEE YEAR MarketLab Personas, 2016

We are currently focusing on testing two frameworks: the SCARF model and the Three Axes Model of Politics.

SCARF:

The SCARF model is “a brain-based model for collaborating with and influencing others.” The SCARF model identifies the factors that trigger a reward or threat response in social situations. David Rock identifies these five factors as Status, Certainty, Autonomy, Relatedness and Fairness. We want to apply the SCARF model to understand if emotional triggers improve the effectiveness of message performance. If so, which triggers are best for which audience, topic, etc?

The Three-Axes Model:

The three-axes model categorizes political messaging in terms of three axes:

• Oppressor/oppressed [naturally preferred by progressives]
• Civilization/barbarism [naturally preferred by conservatives]
• Freedom/coercion [naturally preferred by libertarians]

The purpose of the model is to frame issues in terms that resonate with progressives, conservatives, and libertarians respectively. It helps to categorize political language to predict the language people are likely to use and respond well to, as well as understand the negative consequences of incompatible political values. Unlike all the other models in this list, the model is a theoretical model based on introspection rather than the result of quantitative research or surveys.

The 3AM model is descriptive rather than prescriptive. Specifically Kling does not suggest using a specific framework to talk to different political groups. Furthermore, the most insightful model(s) to understand a particular issue depends on the issue. For example, the civil rights conflict of the 1960’s is best understood according to the oppressor axis, and not the civilization or freedom axes. This is a modern value judgment, which highlights the highly contextual nature of political messaging. Other issues (such as the minimum wage) can be fairly described using any of the three axes.

How We Test

We are analyzing messages in terms of which axis they are aligned with, if any, and correlate their effectiveness with specific groups on specific topics. We theorize that messages should be presented in terms of the axis preferred by the target group.

The Analytics Platform is used to test the hypotheses suggested above. Each factor was added as a binary tag to a set of videos. For example, the tags are in the form of SCARF_S1, to allow three options: positive (S1), negative (S0), or none (no tag). The platform is used to correlate video views and engagement (the dependent variables) and a list of normalized attribute tags based on the models (independent variables).

Technology Details

The platform is coded using the Microsoft .Net Framework, and runs on Ubuntu Linux. The data lake stores media performance data in a PostgreSQL database populated by pluggable services that pull data from social media and ad platforms. Raw data from multiple platforms is aggregated and transformed into a data warehouse. A React-based web interface provides reporting and tagging features.

What does it look like?

Dashboard:

Content Performance:

Audience Analysis:

Tagging videos with psychographic models:

Initial Research Insights

Initial research was shared the State Policy Network 2018 annual meeting. We will continue to share our insights in upcoming posts.