So let’s get the headline out of the way. Donald Trump is not at all comfortable discussing God. That’s based on more than three hours of video covering more than 424 distinct segments spanning more than 200 events.
That’s why you probably clicked here. Now, you get a data science explainer before you get the data. We’re so bait-and-switch.
As part of a set of new features we’re deploying (see our Emotion Subtitles), we generated a huge amount of data from our new approach to Voice Stress Analysis. Each second of audio and video gets individually analyzed, as well as 10-second segments, sequential segments, and the entire speech, interview or press conference.
This compilation opened up an interesting opportunity for analysis. Since our data is extensively tagged and structured, we could document, statistically, exactly what makes him relax, and what makes him tense. So we thought: cool.
A Word about Voice Stress Analysis
You’ll read a lot about voice stress analysis. So let’s address one thing here: it’s not a lie detector test. This is hotly debated, and we prefer to stick with the known. It has not been proven definitively that increases in voice stress indicate lies. If a person believes a lie, they will be relaxed. If a person steps on a tack, stress will increase even if telling the truth.
What this does definitely detect is a level of comfort, stress and/or anxiety. The higher the frequency (due to muscles contracting, including muscles in the neck that affect the voice box, thus the frequency), the greater the indication of stress. By measuring patterns when this occurs, we can identify statements and topics where a person is not comfortable with what they are saying. Coupled with identification the underlying feelings and measuring factors such as word choice and rate of speech, among several dozen others (we gather 115 datapoints per word), it’s a powerful way to uncover how someone feels about what they are saying.
It doesn’t tell you WHY they’re stressed or anxious. They just are. When used in an individual conversation, you don’t have context. The person can just be having a bad day. Or a great day.
That’s why the next part is important: we have hundreds of hours of Trump documented, transcribed and keyworded. A bad day is possible. 200 bad days on the same topic? Unlikely. In fact, we did a basic statistical model and found the odds of having “a bad day” on 200 or more unique days exactly when a particular topic being discussed was… some big number. Excel showed one of those 1e12 things and we just moved on.
Back to Why You’re Here.
So we ran the data. The methodology is important, which we’ll explain in detail:
- Eliminate Bias. To remove bias, we selected only topics that Trump has discussed publicly 200 or more times, according to our database. Every one of those topics / subjects was checked and is reflected below.
- Find Midpoint. For each interview, speech, event, and so on, an individual middle (median) point was established for just Trump’s voice. So if he was having a relaxed day, we measured when topics moved the stress above or below that midpoint. If he was having a bad day, same thing.
- Phrase subjectivity. For phrases, we freely admit this was subjective. We checked our database for frequently used phrases and it found thousands. It’s a literal beast, so “I am going” appears in the list of three-word phrases. We punted and googled “Trump catch phrases” and selected about a dozen. We made a subjective choice to add “Make America Great Again” into the mix, as well as “Thanks”, “Thank You”, “God Bless You” and “God Bless America” into our checks, based on the findings in our topical analysis.
- “You’re Fired” We eliminated “You’re fired” since most of the references were short, pre-recorded clips from the television show vs. a real-world situation.
- Short Segments. We eliminated any segment less than four seconds long, as that can add anomalous spikes, and we want the phrase or topic in context.
- Sample size. This got us to 170.23 hours of video, spanning 30,899 unique segments (1- to 3-word sentences are a unit in our database based on size), from 1980 through this week, covering 1,634,208 words.
- <nerd>This then fed into our algorithm, which is an Adaptive Empirical Mode Decomposition (AEMD) process, to check for deviations outside of 8-12Hz. This is widely recognized as the normal frequency range to monitor. When it goes above 12Hz, it’s considered stress…</nerd>
- A reminder… but again, we use the midpoint from a particular event, to account for the fact that being President probably is stress in and of itself.
One note: you will see topics on the table below with less than 200 citations. Our check of topics included print interviews, his writings and tweets, indicating it is a topic he frequently discusses, but may be represented less than 200 times in the audio and video.
And from that data…
Back to the Lede
Trump is clearly, statistically, uncomfortable expressing gratitude. When he thanks people, based on 67 unique segments where thanking someone was the topic, and another 105 phrase references to thanking someone, he is consistently at an elevated stress level, indicating anxiety.
Similarly, when discussing God as a topic (424 unique segments), he is also uncomfortable, with his voice indicating stress and anxiety well above the midpoint established contextually in the conversation. Note this is specific to discussing God, vs any particular religion or religion itself.
Rounding out the top list of uncomfortable topics and phrases:
- “Make America Great Again” (32 segments)
- “Build the Wall” / “Build That Wall” (153 segments)
- The White House (as an institution – 323 segments)
- Veterans (402 segments)
- Law Enforcement (194 segments)
- The Wall (as a topic – 790 segments)
Okay, but what puts him at ease? On what topics is he comfortable?
The top of the list is what our system classified as “inner cities” but in looking at specific references, it’s discussions of urban planning, cities and infrastructure. He’s well below the stress midpoint when on this topic (145 references). A good number of these references were in interviews pre-dating his Presidency as well.
The Middle East is strongly represented on the list of topics where he is comfortable: Iraq (420 references), Iran (406 references), Syria (281 references) and the Middle East in general (305 references) are all points where he is clearly relaxed and not anxious when discussing.
Rounding out this list of topics where he is comfortable:
- War (248 references)
- The New York Times (89 references)
- Terrorism (366 references)
- “A lot of money” (275 references)
- “Many many” (126 references)
So was there anything else surprising?
Personally, for me, there were a few things, but the world doesn’t need another opinion right now, so take a look at the data below and decide for yourself. If you disagree with anything in the methodology, let us know. But be warned: we make available all our data on request, and will continue to do so. If you disagree with the points above, we’re happy to send you the algo and all the underlying data for you to verify the results for yourself, or to run through a different process. The world not needing another opinion doesn’t just apply to me :-). We’re all about data and verifiable facts at Factba.se, so you’re welcome to think we’re wrong, but be ready for us to challenge you to prove we’re wrong.
|Topic / Phrase||Deviation Score||# of Segments||Length of Segments [HH:MM:SS]|
|Phrase: “God Bless You”, “God Bless America”||1.4990||159||01:03:30|
|Phrase: “Thanks”, “Thank you”||1.3732||105||00:37:33|
|Phrase: “Make America Great Again”||1.1865||32||00:03:45|
|Thanks / Thanking Someone||0.9275||67||00:30:36|
|Phrase: “Build the Wall” / “Build that Wall”||0.6085||153||00:41:48|
|The White House||0.6006||323||02:09:06|
|Phrase: “Believe Me”||0.2056||834||04:38:59|
|Phrase: “Sad”, “So Sad”||0.1921||456||02:52:36|
|Phrase: “Many Many”||-0.1990||126||00:16:37|
|Phrase: “A Lot of Money”||-0.2124||275||01:16:13|
|The New York Times||-0.2714||89||00:36:41|
|Infrastructure / City Planning||-0.6432||145||01:09:01|