Data-driven, data-represented

Is the disconnect between representatives and represented being driven by our increasingly data-driven campaigns?

Many of the essays here begin as questions, and some never get far beyond that initial curiosity. This one is definitely in the latter category of provocation — and may warrant some deeper exploration.

There is an old aphorism about democracy from Thomas Jefferson that “the government you elect is the government you deserve.” This statement has been used in myriad contexts to talk about the value of participation, about the costs of non-participation, about who has power, about who that power is for. If we reverse the focus, it also hints at something about how our representatives see us and how they understand who they represent.

The “voter file” has a long history that begins all the way back in 1896 with William Jennings Bryan’s card catalog. But since 2004, as our civic engagement and especially our electoral campaigns have gotten more and more data-driven, our identities as voters have become represented by increasingly complex datasets. And over those same nearly two decades, our elected leaders have come to rely more and more on those representations to understand us. Those data generally begin with publicly available lists of all registered voters maintained by each state’s Secretary of State (with a couple exceptions) and are then augmented from there. Polling, modeling, marketing automation data, social media and digital engagement data, the results of direct voter engagement from field operations, third-party datasets of all sorts, and various analytics tell particular stories about who we are, what we want, what we are willing to do. Both the sources of those data and the algorithms that drive the engagement that produce much of that data have everything to do with what stories they tell and how we appear in them.

Perhaps our elected officials, who so often seem so deeply, despairingly out of touch or seem to be responding to a reality of their own or driven by some cartoon version of America — perhaps they are just seeing us through a fun-house mirror made up of modern datasets, analytics, and modeled scores?

These representations of us are inherently incomplete, and the data and the algorithms that generate them are never neutral. They are imbued with bias and assumptions and are generated based on a particular set of incentives that may have nothing to do with creating a complete or comprehensive view of who we are. Those data representations are shaped by how we interact with the algorithmic systems that feed them. Because those algorithms are not trying to represent us accurately or completely but simply trying to maximize what drives engagement, perhaps we are increasingly being represented based on an incomplete view of reality. That view of reality is shaped by incentives that end up skewing how our representatives see us and how they think they should engage us and how their campaigns think they will resonate and activate us most effectively. But that view is based on what drives engagement and is not optimized for truth or based on who we really are or how we really think or what we really need — more data, but less understanding. What drives engagement might correlate with truth, but is unlikely to correlate with completeness or the nuance required to understand complex modern life. If the model is off and the incentives that create the data the models are based on are off then their ability to represent us and what we want accurately will shear away from reality too.

If this initial provocation has any weight, how do we start to diagnosis what’s wrong with how we are represented in data? How might a clearer understanding lead us to “deserving more from our government”? What does a more accurate view look like? How can we create and maintain that more accurate understanding of each other? Can we expect more from our leaders without being willing to reveal more of ourselves, not less? Where do we reveal ourselves and does that need to be a conscious, open opt-in conversation? Much has been suggested about making our data and the models we use to define increasingly fine-grained communities coarser on purpose in order to discourage the polarization and sorting that comes from hyperpersonalization. But would coarser data without different intentions about how we use them only create less accurate engines of polarization and extremism?

More questions than answers this morning, but the fun-house mirror view of us created by the technology platforms we rely on for information and connection everyday and the tools we use (and how and when we choose use them) to connect and engage voters combined with the perverse incentives of partisan primary and campaign finance systems that encourage polarization and extremism are all shaping how we are seen and heard by our leaders (and who they see at all). Our leaders, too, are often deeply flawed but perhaps “a better mirror” might help guide them back to us and make us feel more seen and heard more often more accurately as a part of a path back to feeling more represented by our representative democracy.