data for politics #23: Analyzing How Americans Think About Free College

The polling Data for Progress shared on the New Progressive Agenda showed an exciting trend: support for progressive policies cuts across party, ideology, and 2016 presidential choices. Reading through the results, I wondered if there were differences in how these groups thought about the policies presented, given the sharp divisions in debate we see elsewhere. I dug into the open ended text responses from the survey to see if they shed light on why people respond to the policy the way they do.

First, I visualized what words respondents who support free college use versus those who oppose it in a wordcloud. The size of the words corresponds to how much respondents used these words overall, while the color-coding marks words that occurred disproportionately among “no” (red) or “yes” (blue) responses*. The visualization shows that respondents who support free college talk more about “education” while those who oppose it talk more about “college.”

To understand this potential difference in framing, I visualized top ngrams (phrases) from responses that included the word “college.” Those who opposed the policy (the left wordcloud) emphasized the fact that not everyone should go to college, and that college attendees have a responsibility to pay for it themselves. Responses were more diffuse among those who supported the policy (the right wordcloud), but “be able to go” to college was a recurring theme.

While not predictive, the open ends reveal a divergence in how voters think about college: a privilege to be earned by some students, or an economic necessity that should be accessible to all. This tracks against the changing role of college in the U.S., with 60% of Americans completing some college in 2016 compared to 46% in 1980 and 16% in 1950. Voters across party lines who face the present reality that most students feel they must enroll in college, even at the cost of debt, are more supportive of free college.

*Words were selected based a weighting factor equal to the inverse term frequency-document term frequency, a measure of how rare a word is, multiplied by the number of overall occurrences. I filtered to words with a weighted TF-IDF greater than the mean. If words occured in the “yes,” “no,” and/or “don’t know” response groups, I included only words with a range of weighted TF-IDF above the standard deviation of ranges.

Hannah Gorman (@SanFranHan) is an analyst at Adweek magazine. All opinions expressed are her own.