Chasing FDP reform in 2017 Part 3: Analytics, data and political science

I found curious Dave Trotter’s pieces over at the Political Hurricane last week about political scientists – I sometimes don’t agree with Mr. Trotter on matters related to the party but on this he is spot on. The reliance of the Democrats on political operatives who have inherent biases and often do not use scientific methods to target or analyze voting behavior has been a problem for the part.

On Tuesday, Trotter told me the following related to why we need to employ political scientists at the party level:

First and foremost, what is the most important is that political science should be based off of theory. The idea is to look at causation. Basically, A causes B. Just seeing if a relationship exists doesn’t mean that causation exists. So, someone needs to be convincing that A causes B, and know the political science literature to back up the reason behind their theory. Second, political scientists know the current literature done by political scientists. One thing that we always hear in politics is that “voters usually vote the way of their parents”. This is something that we hear all the time, but we never know where it came from. Well, it came from Campbell, Converse, Miller and Stokes (1961) in the groundbreaking book The American Voter. Having this knowledge allows political scientists to build strong theory, as it is usually building off of existing literature. Finally, political scientists seek to understand “why” something happens, not “what” happened.

Trotter’s article on the subject can be found here.

My personal feeling is that Trotter is most certainly on to something. Political scientists make a genuine effort to understand why something happens and that is critical to predicting future voter behavior. Whether or not Democrats across the state feel academics should be empowered to help make data-driven decisions (which is after all part of the reason the part exists), one cannot argue that the results doing it the current way have yielded fruit. Many targeting decisions are made based on subjective factors  – a lobbyist or special interest groups wants to take our a certain Republican legislator or give another a pass and other factors that mirror these types of considerations.

Analytics and understanding them have transformed so much of what we do on a daily basis. They’ve also transformed business, the internet and sports. In soccer, an industry I am familiar with English clubs now have full-time analytics staff that are employed simply to track and rate players at other clubs for potential transfers in the future. This process is largely data-driven and in addition to the raw data focuses on various causation factors that are then thrown into a formula to rank potential transfers.

Data doesn’t drive the decision making process in the Democratic Party as much as it should or as much as we are led to believe it does. Part of this is due to decision makers being in a Tallahassee-based bubble where those who  make the final calls on the allocation of Democratic resources are often influenced even subconsciously by the buzz created by Republican-leaning or incumbent-protecting driven lobbyists and consultant. This prevents a truly objective process and also prevents an emphasis on understand voter behavior and causation.

The next Democratic Party Chair needs to employ data and political science in a more meaningful and substantive way.


  1. Give the people a candidate that can articulate a vision for the future. Who is honest and speaks from the heart. Who can debate issues and fearlessly speaks truth to power and I’ll show you a winner. It’s not science, it’s common sense.


  2. Hank Porter · ·

    Specifically, which decision-making staffers within the party do not utilize data-driven analysis to drive their decisions? Specifically, which vendors of the FDP do not use data-driven analysis to drive their work product? Additionally, which current FDP vendors/operatives have inherent biases beyond those that ANY replacement-level operative would not?

    In addition to listing the names of these staffers and vendors, please provide the evidence for the basis of your claim.


    1. I’m not going to name manes but biases are pretty clear and natural. We’re all human and most if not all of us would likely make the same decisions if in the same fishbowl. Lobbyist X think Republican legislator Y is unbeatable so let’s not target there. Special Interest group G thinks Republican Z is pretty terrible so let’s target her. The bottom line is we keep losing doing the same things – let’s incorporate some academic thinking of people who have little bias or agenda-driven or faction-driven considerations to help us do better.


    2. There is a difference between having data and actually knowing what to do with the data. I have never seen any theory-based analysis by the Florida Democratic Party, which then used data to test if certain hypotheses are true or false. Instead, the FDP just gets a poll, looks at numbers on that poll, and make face-value assumptions based on contingency charts, and say “okay”. There is absolutely no statistical inference whatsoever made by random sample data received by the party. I have also seen no inference based on aggregate-level data either.

      Still, I have seen the data and the “number crunching” that the FDP has done either in house or commissioned by someone else, and it is extremely elementary. Basically, anyone with basic knowledge of political maps and Excel can come up with conclusions. Yes, maps are pretty but don’t explain anything. And since many of the people in the Florida Democratic Party have no experience in political analysis, they cannot challenge any of the elementary data presented.

      This is the reason that I have advocated in my article that there should be at least two trained political scientists working for or independently contracted with the Florida Democratic Party. Having one political scientist can lead to a party blindly listening to the conclusions of a political scientist. But as I mentioned in my article, political scientists love to go after each other. Thus having multiple political scientists would help the party because they would learn about what options are available.

      Basically, I don’t think that you know what a political scientist does. Just looking at numbers and maps does not make you a political scientist. As I mentioned in the quote, science is based off of showing causality and testing hypotheses. Simply “having data” doesn’t make you a political scientist…just like owning a car doesn’t make you a Formula 1 driver.


  3. Science is great. I believe in science and use it daily in my profession. In addition to science, we need to have current data from actual voters. That means contacting those voters and asking them about how they feel, what they think, and what they want. And the only people who can do that in sufficient numbers are the DECs.


  4. Jim Callahan · ·

    Regardless of age you need someone whose head is in the 21st century; there have been a lot of changes in politics, political science and in data science and a 20th century only mindset won’t cut it.
    1. Politics changed with the 2000 and 2008 elections (no longer 50%+1, but 50% + recount margin + 1). It remains to be seen whether the neighborhood team leader (NTL) / snowflake model endures.
    2. Political Science changed with Gerber and Green’s revival of field experiments, with their “Research Note” in September 2000 issue of the _American Political Science Review_. “Gerber and Green were surprised to learn that the use of field experiments had begun and effectively ended with the publication of Gosnell’s getting out the vote in 1927.” Victory Lab, page 74, “Despite the enthusiasm that greeted Gosnell’s method for studying campaigns, no one tried to copy him, replicate his study, of build upon it. After printing Gosnell’s article, the _American Political Science Review_ did not publish another finding from a randomized field experiment for half a century” Victory Lab, page 27.
    3. Data Science combines computer intensive statistics, Bayesian statistics (including simulation methods such as MCMC) and machine learning techniques from computer science. Advances in machine learning have enabled “self driving cars” (“autonomous vehicles” in DARPA speak), Netflix recommendation systems and Google Translate; these are things not dreamed of in old fashioned statistics.Although 20th century statistics students might have learned “Logistic Regression” they would learned it as a regression technique for when you have a categorical variable on the left hand side of the equals sign and not as a classification technique in the same category as “K-Nearest Neighbors (K-NN) or tree based methods such as “Classification and Regression Trees (CART)” which evolved into the Kaggle Competition winning random forests and boosting methods. Finally data science uses visualizations including displaying data on maps from Geographic Information Systems (GIS). The political reputation of data science has risen and fallen with Nate Silver’s predictions.
    Yes, there are unicorns, I have met a few; Dave Trotter is one. Other unicorns would include Travis Brooks, Matt Isbell (haven’t met but read his posts at, Doug DeClue and myself. Matt Isbell and Dave Trotter have political science degrees, Travis Brooks has a computer science degree, Doug DeClue has a degree in Aerospace Engineering and I have an Economics/MBA degree with work experience in economic forecasting (at Chase Econometrics) before managing my first political campaign. Unicorns exist, but they are rare and not necessarily Political Science majors.


    1. Wow, Gosnell!! I love kicking it old school. In the 1920s, the University of Chicago was center stage for voting behavior research (with Charles Merriam there as well). In the 40s and 50s it changed to Columbia, with the Michigan model cementing UM’s status as the premier place for behavioralist in 1960 after The American Voter.

      I am not going to disagree, because I think you are correct, but I am going to bring up a concern in the political science community. The biggest concern in the academic field is that there are too many mathematicians and statisticians entering the field of political science because their respective fields are becoming more competitive. Since methodology courses are becoming part of the undergraduate requirements for BA and BS in political science, universities look for those who have a strong understanding of math, statistics and methodology. However, many of these people lack an understanding of basic political science, and their knowledge of previous academic scholarship on the subject is non-existent. Yes, they might mention The American Voter because it is part of their literature review, but there seems to be a lack of understanding what it is about.

      To give you an example, I was at APSA in 2014 and two political scientists from Stanford, with their impressive Monte Carlo simulation, came to the conclusion….wait for it…..that black voters in the inner city are more likely to vote Democratic. It was apparent that what they examined was common knowledge in the political science field, as well as partial observers of politics on CNN! A number of political scientists in attendance asked the famous “so what” question.

      This is starting to become a huge problem in the academic field of political science, and it is because the universities are looking for methodology specialists, not Americanists or those in other disciplines. Basically, they use political science to show how well they know methodology, when it should be the other way around. We are about to embark on an era in political science where data will drive theory. Instead, theory needs to come from an understanding of political science, and only then should you work on a data set to test a theory.


    2. You can see why my first sentence in the quote is that political science needs to be based on theory 🙂


  5. Jim Callahan · ·

    Dave, I ran into a similar problem with a women consultant to the Rick Singh(D) for Orange County Property Appraiser campaign. She did a regression relating Rick’s performance to various precinct level properties. She showed that Rick Singh does well in “low-turnout” precincts. I laughed and said yes, Rick Sing does better in precincts which have high Democratic registration which tend to have lower voter turnout, but that Democrats still win, despite lower turnout because of the registration advantage in Orange County, Florida. Republicans (if they want to win) have to have higher turnout (via efficiency or voter suppression of Democratic votes). Given the Democratic registration advantage in Orange County, Democrats (in Presidential election years) can afford to be sloppy and not as efficient at turnout as the Republicans. So, yes, just relying on mathematical models without insight (even supported by data) can lead one astray. That’s why we need to look at data science teams as described in “Doing Data Science” by Cathy O’Neill and Rachel Schutt.
    Political Science and political experience are the “domain expertise”. Doug DeClue, Travis Brooks and I all benefited from working with Orange County State Committeeman (and former DEC Chair) Doug Head and I also benefited from working with former DEC Chair Patti Sharp. Matt Isbell might have (I speculate) have benefited from working with Jon Ausman.


  6. Jim Callahan · ·

    Of course, data, theory and models are all good; but as consultants we need to derive actionable steps that will lead to competitive advantage for our client, be it our party, candidate or issue. Then we need to pile on enough competitive advantages that we achieve victory.


%d bloggers like this: