American Political Ideology, A Twitter Bot Approach

American Political Ideology, A Twitter Bot Approach

Note: This article was written in conjunction with Alexander Agadjanian, a fellow political science undergraduate student at Dartmouth College. Follow him on Twitter @a_agadjanian.


Voters are often overanalyzed. They’re put into buckets based on their party or ideology. They are often assumed to have consistent policy preferences — either liberal or conservative — across various issue domains.

But this is hardly the case. One of the more longstanding and well-established findings from political science research on political beliefs has been that Americans aren’t particularly ideologically coherent. Phillip Converse most notably revealed this idea about Americans’ political character in 1964. More recent work by Donald Kinder and Nathan Kalmoe has confirmed that such a description applies to the current day as well. Americans who self-identify as Republicans sometimes hold liberal positions on certain issues, and Democrats skew more conservative on some other policies.

This isn’t necessarily intuitive. Research has shown that the last few decades has ushered in a development of partisan sorting, wherein partisanship and ideology have become increasingly correlated. A simple scan of the current political climate would imply that Americans are sharply divided along partisan and ideological lines. Simply looking at headlines published from national opinion polls will reveal the same. But when you drill down to actual policy preferences that Americans tell pollsters and researchers, these sharp divisions are not as apparent. Sometimes that’s hard to realize, and so to address that, we thought of a project that easily shows how cross-pressured — influenced by some liberal and some conservative positions — Americans really are.

We wanted a way to view these micro-level divisions and powerful, telling inconsistencies in an accessible way. So we did just that.

We created a Twitter bot and dubbed it “@American__Voter” (a nod to another seminal and related study) that uses data from the 2016 Cooperative Congressional Election Study, a nationally representative survey of 64,600 American adults, to fuel the bot. Every hour, a computer program running the account (1) randomly selects an individual who took the survey, (2) selects 3 various issue positions held by the individual, (3) and spits out the individual’s party, self-described ideology, and those 3 policy preferences in one convenient tweet.

Sometimes, the voters profiled by American__Voter will make “logical” sense: for example, a voter will support both gun control, the affordable care act, and oppose immigrant deportation, all of which represent liberal attitudes. But that’s not what makes it interesting. Rather, we are interested in those less “constrained” Americans. Take the following two voters as an example of this:

The first is an example of a possibly liberal voter with liberal positions, but who identifies as Republican. The second, a voters with a logical social identity (conservatism and Republicanism are very much compatible) but inconsistent policy preferences on social issues that fall on the liberal side of the spectrum.

We can learn a lot about the current state of US politics by simply analyzing the oft-unpredictable attitudes of American voters. Our bot makes that clear, every hour on the hour until it runs out of people who took the survey. It will reach that expiration date in 2031. Think you’ll be on Twitter around until then? Follow the bot here!


For more details on what went into this process, here are the steps we took to generate everything:

1) Select variables from the first wave of the 2016 CCES survey: party (converted to a 3-point scale the groups leaners with the parties they lean toward), ideology (5-point scale), and 7 issue position variables, numbered A through G below.

A. “On the issue of gun regulation, do you support or oppose each of the following proposals?” Make it easier for people to obtain concealed-carry permit (support or oppose) B. “What do you think the U.S. government should do about immigration? Select all that apply.” Identify and deport illegal immigrants (support or oppose) C. “Do you support or oppose each of the following proposals?” Always allow a woman to obtain an abortion as a matter of choice (support or oppose) D. “Do you support or oppose each of the following proposals?” Strengthen enforcement of the Clean Air Act and Clean Water Act even if it costs US jobs (support or oppose) E. “Do you support or oppose each of the following proposals?” Eliminate mandatory minimum sentences for non-violent drug offenders (support or oppose) F. “Congress considers many issues. If you were in Congress would you vote FOR or AGAINST each of the following?” Would repeal the Affordable Care Act of 2009 (also known as Obamacare). (for or against) G. “Congress considers many issues. If you were in Congress would you vote FOR or AGAINST each of the following?” Raises the federal minimum wage to $12 an hour by 2020. (for or against)

For the tweets, many of these issue position texts are condensed to fit the 140 character limit.

2) Remove all individuals that answered “not sure” to any of these questions, skipped the questions, or were not asked any of the questions

3) Randomly select an individual survey-taker from the CCES survey and randomly choose 3 of the 7 available issue positions (A-G above) that they express

4) Combine party, ideology, and the three issue positions into a tweet

5) Tweet it, and that’s it! Five easy steps to an insightful and powerfully communicative Twitter bot.

We hope you will follow along (and if you’re reading this far down, chances are you will).