674M response paper 2

13 minute read

As part of a course on the internet and political campaigns at HKS, I wrote a response paper to Daniel Kreiss’s Prototype Politics.


Media coverage, as the saying goes, is the first draft of history – and the recent history of political campaigns, as these public drafts describe, has been heavily focused on data and analytics. Voter files, databases, predictive models, and similar artifacts have become objects of public debate and professional practice in a way that they weren’t even a few years ago. How and why this shift occurred - how analytic techniques in politics have risen to prominence - is the story of political campaigns in recent years. The campaigns themselves, the party-based networks of political actors who influence and staff them, and the media who cover it all have brought analytics forward in a mutually reinforcing cycle.

Daniel Kreiss describes this process in Prototype Politics, his 2016 book on the development of political data. Ironically, given the scientific reputation the analytical techniques have, Kreiss takes a social constructivist approach. He examines how these techniques spread, or failed to spread, from pioneering campaigns (the “prototypes” of the title) to the rest of both parties. How a campaign comes to be viewed as a prototype, a pioneer of a useful new way of doing things, is a socially situated process that’s not as simple as just winning an election. It involves the “extended party network” (Kreiss 38) coming to a socially constructed agreement about whether some new way of campaigning is useful, and then mediating the diffusion of the new techniques throughout the party.

This emphasis on the social processes driving the “uptake of technology, digital media, data, and analytics” (2) is a useful contribution. They’re easily overlooked, especially in political contexts, because of the difficulty of observing them: Kreiss, after all, needed extensive interviews and a dataset of political staffers’ job changes to do the analysis in his book. His analysis, however, is incomplete. After extensive discussion of these social processes, he places too little emphasis on their implications (for actually winning elections, for who holds power within the parties, etc) and how related feedback loops may drive technology uptake.

Parties as networks

The central actor in Kreiss’s story of technology diffusion is the “extended party network.” It includes not just the formal party organizations like the DNC and RNC, but also a broader universe of “technology, digital, data and analytics consultancies and knowledge-producing organizations” (35) and the pool of staffers who work at such organizations. These networks are flow networks, with knowledge of new campaign innovations spreading (or, in some cases, failing to spread) through them to corners of both parties far away from where they were first developed.

Beyond their role in spreading knowledge, Kreiss emphasizes the ways in which these networks not just assist campaigns, but constitute them. He traces the different paths taken by the Democratic and Republican networks after the 2004 election, and concludes that the way the 2012 campaign played out was path-dependent: “the Obama campaign in 2012 was in part the outcome of the work of the extended Democratic Party network in between cycles as actors crafted a new data architecture, new analytics practices, and technologies.” (14) If the party actors who spent the years from 2008 to 2012 developing that infrastructure had decided to spend their time on nonpolitical employment or other party-aligned work, the 2012 campaign would have gone very differently. (The same is true, perhaps even more so, for the Republican Party actors who didn’t spend those years developing analytics infrastructure.) It’s a useful perspective, and it captures the informality and constant job hopping that characterize campaign work. Driven ultimately by the two-year election cycle, few staffers work at the same place for long1. Stints of two years, one year or even less are common. But even with such frequent job changes, or perhaps because of them, it’s been my experience that political staffers have a keen sense of themselves as a team. Kreiss’s extended party network identifies and functions as a community, keeping its members in touch across their many campaign job changes and bouts of nonpolitical work.2

More subjectively, but no less importantly for that, it’s common knowledge which side of the aisle any given firm or person is on. The parties’ networks don’t overlap much, and the occasional exception proves the rule that they try to regulate aisle-crossing. Kreiss, for example, mentions the “third-party vendor” NationBuilder, a provider of “voter data” and “field tools.” (32 - 33) While it may not work exclusively with liberal or conservative clients, NationBuilder is not a third-party vendor in the same way that Google or Microsoft would be. Its founders have roots in the extended Democratic network, and I can attest that there’s considerable ill will and informal social sanction over their decision to take Republican clients.

The party organizations

For all that they’re diffuse and broad-based, these party networks still center on the formal organizations of their respective parties. The DNC and the RNC are the the most central organizations in the parties, and provide a natural place to host most kinds of data work. The advantages they bring are significant: they’re “permanent entit[ies]” (22, quoting Azarias Reda) that won’t be allowed to go out of business, they’re centrally located in their networks, and (though Kreiss doesn’t discuss it much) they have powerful brand names. The importance of a well-known brand for attracting or even getting the attention of “field-crossers” (130 - 131) from the technology or other industries shouldn’t be understated. Without a high-profile presidential campaign for most of every four years, the parties may be the next best thing.

The fact that most political data and analytics work nevertheless happens at other organizations seems puzzling. Kreiss’s explanation, which agrees with the one that I took away from working at the DNC, is that it’s fundamentally about money: neither party’s committees are “able to license these tools, charge for them, and use that money to reinvest” (13, quoting Bryan Whitaker) as a matter of election law. The DNC, for example, isn’t allowed to own a fee-for-service data subsidiary, which causes funding for its data work to boom and bust with the election cycle. Wild swings in an organization’s budget are obviously not conducive to long-term infrastructure investments, and so most analytics and data work has migrated away from the party committees over time.3

Rather than being inevitable, in other words, the existence of Kreiss’s “extended networks” of data and analytics organizations is a contingent historical fact. The early impulse was to centralize their work in the party organizations; had the law allowed a sustainable funding model, it likely would have stayed there.

Incentive structures

This discussion of funding gets to a broader point: the importance of the incentive structures facing the party networks and the actors within them. Kreiss is careful to distinguish his discussion of these incentives from “rational choice perspectives on campaign strategy,” which are “overly presentist and narrowly focused on structural constraints.” (208) In keeping with the sociological bent of his book overall, Kreiss stresses the social construction of these incentives and the ways that party networks determine how best to achieve their interests.

This construction, in his telling, is something that happens within each party, and is prompted though not determined by election outcomes. Losing an election touches off “a collective process of meaning-making”, in which “particular campaigns are transformed [] into ‘prototypes’” that ground future campaign strategy. (15) It is only “particular” campaigns that receive this treatment, however - the example of John McCain’s loss in 2008 reveals how some losses are written off as predetermined and don’t prompt changes. (15)

Moreover, even if the choice and uses of prototypes are socially constructed, Kreiss acknowledges that incentives once formed can have predictable effects. One of his “core arguments” is that the “near universal buy-in” to the current Democratic Party voter file system has created something that the “party’s technology ecosystem convenes around.” (211) The strong incentives for campaigns and consultants to use the party’s file, given how rich a data source it now is, keep the system stable against campaigns’ desire to keep their data private and consultants’ natural desire for vendor lock-in. Kreiss provides convincing evidence that the Republicans’ less centralized system allows both of those desires to be indulged, with overall negative consequences for the party’s campaigns. (212 - 213)

Kreiss pays too little attention, however, to this social construction’s dependence or lack thereof on external signals. Rational choice perspectives do, after all, have an appealing simplicity: campaigns and their constituent actors should do things that help them win elections. And yet, somehow, the reality of how these actors make decisions is more complicated. Part of the explanation must be the weakness of the feedback election results provide about methods of campaigning.

It’s obvious that this feedback is limited: elections for most offices happen at most every two years, and only indicate whether a candidate won or lost. It’s less obvious that the feedback may not even be closely connected to campaign methods at all: Kreiss, citing Sides and Vavreck and ultimately former campaign staffers, reports that the staffers themselves “stated the campaign was worth [at most] two points” of vote share, or were unsure. (164) Campaign organizations ultimately matter less in winning elections than other factors: candidate quality, the state of the economy, which party is in power, any active wars, etc. Good campaigns can lose elections because of a bad candidate or a poorly timed recession, and a poorly run campaign can win with a great candidate or a favorable economy.

To use an evolutionary metaphor, the selective pressure on campaign techniques is weak, but they can still undergo large changes over time due to genetic drift. The degree of social construction Kreiss describes isn’t a given: it’s the natural consequence of very limited electoral feedback provided to a campaign industry desperate to make meaning and gain a competitive edge.

Indeed, the greatest impact of campaign innovations on big-picture questions like public policy, how the government is run, and the structure of the party system isn’t likely to occur through actually winning elections. The greatest changes may be in the distribution of power within the “extended party networks” and even outside them. For example, however much new analytic techniques help win elections, we know they can help enormously in raising small-dollar donations. Both Obama campaigns raised a majority of their funding, hundreds of millions of dollars in each case, from small online gifts. Knowing that this sort of funding is available lessens party actors’ dependence on the billionaires they might otherwise rely on for funds. The greater uptake of analytically informed fundraising techniques by Democrats may be an underappreciated contributor to current policy differences between the parties.4

Where we go from here

Kreiss concludes, as books in this vein frequently do, by trying to synthesize his analysis into a unifying theoretical conception of how campaigning is changing. He describes the new way of doing things as “networked ward politics”: a “data-driven, personalized and socially embedded form of campaigning” that is “the high-tech form” of politics as formerly practiced by machines like Tammany Hall. (217) Kreiss elaborates that he views campaigning as newly “personalized” in a threefold sense: first, in terms of campaigns leveraging “people as media” in the course of field canvassing and social media appeals; second, in terms of campaigns meeting citizen expectations to determine their own levels of often highly temporal political engagement; and third, in terms of campaigns increasingly appealing to citizens as individuals and not as representatives of demographic groups. (217)

His analysis is unfortunately at its weakest here. The evidence he provides certainly supports these contentions, and they have an intuitive plausibility. But calling this mode of political organization “ward politics,” and indeed even describing it as “socially embedded,” is misleading. The differences are clear enough:

  • Real ward politics involves building strong social institutions. Tammany Hall may have been corrupt, but it was vital and locally rooted. Politics as conducted on social media, by contrast, involves large numbers of people who may not know each other in any other context, are not geographically near each other, and usually lack enough in common to form any offline organization.5 This atomizing effect is not conducive to building enduring machines like Tammany, and may be related to the current era of “strong partisanship and weak parties.” (see, e.g., Azari 2016 for an overview of that concept, summarizing her related academic work.) Even door-to-door canvassing is a fleeting interaction: canvassers speak with their targets for at most a few minutes, rarely see them again, and work alone or in pairs. They often come together in a group only at the beginning of their shifts to receive walk packets.
  • Offline, relationship-driven machine politics happens in a stable social context: those involved, from the boss down to the voters whose votes are being purchased, know each other and have a long time horizon for their political relationships. People would vote for Tammany because they expected to continue interacting with it and knew that Tammany could do something for them: patronage, help with a government service, or other benefits. Social media, by contrast, is a hall of mirrors: it’s frequently not even clear if the participants in a discussion are all human. Nor is it clear, assuming they aren’t bots, that they are who they say they are, rather than being (for example) Russian operatives. When “the returns of networked ward politics come in less tangible forms of identity and expression” (220) than the old model of patronage, it’s difficult to identify what if anything holds this supposed new machine together.

Kreiss acknowledges that “there is no enduring social, on-the-ground knowledge [for campaigns] that comes through physical presence” in the new way of doing things, and that “it is a technology-intensive practice.” (218) He, like most everyone else, failed to anticipate how this more mediated and less socially rooted state of affairs would be abused in the 2016 campaign (by Russian intelligence, profit-seeking purveyors of disinformation, and the ad-optimizing platforms themselves), and how easy and efficient that abuse would be.


Kreiss, Daniel. Prototype Politics: Technology-Intensive Campaigning and the Data of Democracy. New York: Oxford University Press, 2016. Kindle edition.

Azari, Julia. “Weak parties and strong partisanship are a bad combination.” Vox Media. https://www.vox.com/mischiefs-of-faction/2016/11/3/13512362/weak-parties-strong-partisanship -bad-combination. (published Nov 3, 2016; accessed February 26, 2018).

(Citations with only a page number are to Kreiss 2016.)

  1. This is obvious to anyone who’s worked on political campaigns. Kreiss, for example, mentions that “a number of staffers [in his dataset] worked on multiple presidential bids.” 

  2. Why the parties are organized this way is an important but separate question from the ones Kreiss considers. That the well-known weakness and loose-jointed organization of American political parties is reproduced in their networks of data practitioners is his more original contribution. 

  3. Civis Analytics is an instructive example. The same group of people that currently make up that company’s leadership were mostly on staff at the DNC in 2010, before moving to the better-funded Obama campaign’s payroll in 2012 and then incorporating after the election. 

  4. Consider a hypothetical: had the extended Democratic Party network not developed expertise in online fundraising, and specifically if Revolution Messaging hadn’t been around to raise money for Bernie Sanders, would his campaign have done as well and been able to pull the party as far left as it did? 

  5. They may certainly be members of offline political organizations, and even coordinate action through them, but very few enduring movements have started entirely online between people with no offline ties.