Text Analysis Explains Why Obama Wins Debates

Tom H. C. Anderson
October 23rd, 2012

Scientific Text Analysis of 2012 Debates Explains why Obama’s Style May have Out-performed Romney’s

Following up on this weekend’s analysis of the first two presidential debates we added yesterday’s final debate to the analysis. General consensus among viewers and pundits alike seems to be that Romney performed better than expected in the first debate, and Obama performed better in the final two debates. Here’s why this may be:

Obama’s Style More Persuasive

Obama’s success may in no small part be due to his speech style which combines words emphasizing certainty, thinking, and causation, compared to Romney’s directness and more perceptual style. In other words, Obama is far more likely than Romney to use words related to thinking (cognition) such as “I know” or “I understand” rather than Romney’s more perceptual words “I feel” or “I see”. Perceptive words are often considered less scientific and more subjective than thinking words. Obama’s speech is also more likely to contain causation, in other words he is more likely to back statements up as indicated by increased use of terms like “effect”, “hence”, “because” or “results”.

Additionally, Obama’s higher certainty (terms like “never”, “always”, “sure” and “definitely”), can provide stronger and more convincing arguments and can help paint the speaker as an expert.

In regard to sentiment, as discussed in yesterday’s post, throughout the debates while both candidates seemed to use equal amounts of positive emotion in their speech, understandably perhaps, Romney was more likely to also use negative emotion (especially sadness in the first debate). This can serve as an implicit way of attacking your opponent.

However, Obama’s insistence on the use of more collectivist language (i.e. use of “we” rather than individualist “I”), seemed to serve him well in countering these attacks. While it was not always clear whether Obama’s “we” referred to himself, his team or the nation as a whole, speaking in this ‘voice of the nation’ style undoubtedly served him well and together with the more sophisticated (longer words) and thinking words in his speech, may have helped paint him as a responsible, more certain and expert steward of our nation.

Read more on content analysis of the debates here.


[NOTE: Text analytics can be used to investigate some of the reasons why Obama seemed to perform better than Romney. Explaining the success of Obama or Romney in debates they are considered to have won is difficult because neither candidate changed style much across the debates. In a few instances they switched style slightly from debate to debate, but more often their style categories were consistent across all three debates (including those lost). This emphasizes the fact that language use is something inherent to speakers and is quite subconscious. Rather than being a strategic choice, differences in style between debates may be a result from their reaction to debate-specific circumstances (questions from the moderator, behavior of the other candidate, etc.). Therefore this analysis focuses on Obama’s and Romney’s styles across all three debates.

The above speech characteristics are not the only indicative attributes enabling a candidate to win. However, since most likely a win is a combination of several language characteristics appearing in some or all debates (plus other non-linguistic issues including facial expressions, candidate’s general appeal to the public, their political affiliation and issue position) we can assume that some of these categories played a role in contributing to the success in the debates.]

11 thoughts on “Text Analysis Explains Why Obama Wins Debates”

  1. I think there are two overarching issues in this interesting presentation. The first is that I’m not sure that words are the best unit of analysis that people (viewers apprehend). Perhaps themes are more applicable to voter perceptions. Second, even if words are the right unit of analysis, they don’t seem to be related to some underpinning concept like leadership. In the end, all sorts of data show that perceptions of leadership remains the decisive variable for a candidate to portray. But of course as a communication guy I like the approach in general. Something to think about.

  2. Cvetko, this does represent analysis of both words and themes (literary/psychological constructs).

    Unlike our typical work involving more ‘unstructured’ analysis of comments from hundreds of thousands of consumers, where we look at more detailed ‘themes’ or issues, we did not find that approach as useful or interesting here as there are only two actors speaking about the specific issues presented to them in the questions.

  3. Interesting study. I did notice that Romney used more qualifiers than Obama. However, as Cvetko stated, what is not factored in is viewer perception, including speaker demeanor, perceived straightforwardness, leadership ability and manner of speaking. Based on the input and post debate inclinations of a couple of network focus groups, President Obama appears to have generally lost ground due to the debates.

  4. @Josh It’s 2012, of course there’s no human coding. 😉 Quite a common technique in content analysis. As noted in the initial analysis POS and emotional constructs are counted and chi-square is performed based on total words in the corpus.

  5. @Dean unless otherwise indicated (such as when we mentioned “sadness” in first debate), Positive and Negative emotion in this analysis would be referring to use of words such as “happy” “good” etc. VS “worthless”, “awful” etc.

  6. Tom,
    It would be interesting to see the three debates broken out since they differed so much (Obama certainly didn’t “win” the first one).

  7. While at OdinText we typically analyze thousands or even millions of comments from surveys or call center logs, content analysis can be used to analyze even small unstructured data sets like debate transcripts.
    The point of this fun non-partisan analysis of the debates was not to predict the winner of the election or even the debates, rather previous research in the area of content analysis proves that people prefer speakers who are similar to themselves. Further analysis could compare how Democrat, Republican or Independent voters speak to better understand which candidate’s style appeals to them.

  8. If the intention was not to predict the winner, then the title “Text Analytics EXPLAINS why Obama wins debates” is both pretentious and misleading (which is often the case with many text analytics vendors). One would sill have to demonstrate the connection between those linguistic styles and people’s perceptions. Was there any analysis of previous debates (starting from the Nixon-Kennedy debate) to see whether some of those characteristics are constantly associated with the winner or with people’s perception about who won the debate?

    It also neglect all other factors like the non verbal behaviors (which was important in the first debate in explaining the poor performance by Obama) as well as the impact of catch phrases, memorable replies (“horses and bayonets”), etc.

    Text analytics describes, after the fact, how Obama’s speeches differed from Romney’s speeches. Until we get more evidence of the causal effect of those styles and of their effect sizes, I would try to avoid words like “explains”.

Comments are closed.