Deception Detection In Non Verbals, Linguistics And Data.

Turnbull, Shorten + 100 Years Of Election Speeches

How does Malcolm Turnbull's election speech and Bill Shorten's Budget response compare to the last 100 years of election speeches in Australia?

To find out, I downloaded election speeches from 1903 from:

I cleaned up the data, removing things like [APPLAUSE], and used a psychological text analysis tool called LIWC (Linguistic Inquiry And Word Count) developed by James Pennebaker at the University of Texas and which has been verified and used in over 6000 articles and studies on Google Scholar.

The program looks at about four and a half thousand words in eighty categories. These categories tapped emotional and cognitive dimensions of the speaker, revealing things like state of mind, the tone and how analytic someone was, as well as keeping track of the almost "invisible" function words used in speech.

Custom MatLab code was used to find highly significant groups of words that was able to separate the winners from the losers over all the elections.

This in itself was amazing to me, I wasn't sure whether there would be a clean separation between what the winners of the last 100 years of elections said, and what the losers said.

It turns out election winners tend to use certain language, as do the losers. Comparing Malcolm Turnbull's election speech to this "model" placed him in the winner group, while placing Bill Shorten in the losers group by a long margin, hence the prediction of Turnbull winning by about 85-90%.

Some of these categories are positive (which winners have more of) and some are negative (which typically losers have more of). So for instance, the text analysis program has groups of words relating to power-awareness - this captures the degree to which people use words such as command, boss, victim and defeat. This measure to the degree the awareness of the power they have. In Australian politics, this equates to less being better, it tended to be higher in the losers group.

Turnbull comes across higher on the Authentic algorithm (capturing cognitive complexity and relatively low rates of negative emotion) and also his Tone is higher, both being positive.

On the downside for Turnbull has greater use of the word They (from the last post) which is negative.

Shorten has a problem with far too much equivocation, this is hedging language which reduces commitment and allows one to minimise what has been said if it turns out to be wrong at a future date. Words like I believe, I think, I thought, suppose and so on. He also uses too many negations such as couldn't, should't and wouldn't. (which is an indicator in deception or spin if it's not in response to a specific question).

This analysis was Turnbull against ALL the election winners, then against all the election losers; the same was done with Shorten. It compared them to the model - how well did they fit in the previous election winners group, how well did they fit in the losers group.

It did not compare Turnbull directly to Shorten with their speech, I will be doing that next post.

No comments

Post a Comment

© ElasticTruth

This site uses cookies from Google to deliver its services - Click here for information.

Professional Blog Designs by pipdig