A fired up Chris Hipkins may have had "swagger" and scored "hits" in Wednesday night's leaders' debate, but the winner could have been a single word repeated an astonishing 76 times by National's Christopher Luxon.
Or perhaps it was the moderator, Newshub's Paddy Gower, who got more words in than either of the leaders who were there to debate.
Who talked the most?
After last week's first head-to-head debate, hosted by TVNZ, an analysis by RNZ found Luxon spoke 43 percent of the words and Hipkins 31 percent. The moderator, Jessica Mutch McKay took 24 percent.
In Wednesday night's Newshub debate, Luxon again got more words in than Hipkins - 36 percent for the National leader to 27 percent for Labour's man. The big mover was the moderator - Gower in this debate - who had 37 percent of the words.
Gower's verbosity was no surprise, however. In the 2020 leaders' debate he hosted, Gower took the same proportion of words ahead of Judith Collins (35 percent) and Jacinda Ardern (28 percent), the two leaders at the time.
Hipkins was widely judged to have won last night's debate - or he was at least credited with being much more feisty than in the opener last week.
That sentiment is reflected on another measure in our analysis - the longest answers by number of words.
In a turn-around from last week's performance, Hipkins gave seven of the 10 longest answers in the Newshub debate, compared with only three from the previous TVNZ debate.
What was said
If you've watched both debates, you might have noticed there's one word Luxon uses over and over again. In the TVNZ debate last week, he said "actually" 69 times. In last night's slightly shorter format, he said it 76 times.
Hipkins' top word last night was "people", followed by "going". Last week, his top word was "think" (31 times), followed by "health" (26 times).
But the leaders may be getting the hint about the tedium of repetition for other terms which have been frequently used during the campaign. "Coalition of chaos", "bread and butter" and "squeezed middle" weren't mentioned by either leader last night.
Terms still in use included "gone backwards", which Luxon said three times, and "six years" which he said four times. On one occasion he used both terms in a single sentence: "Everyone knows in this room that we have gone backwards over the last six years."
The term "wasteful spending", which Luxon has used heavily in past interviews, was picked up by Hipkins, who used it five times to counter Luxon's attack on debt: "$200 million of it goes on health. Are you saying that's wasteful spending?"
RNZ counted how many times each leader used certain words which relate to key topics. The most-mentioned keyword was police, which was a topic covered early in the debate when Gower asked about crime and gangs.
Dodging the hard questions
A common tactic to avoid answering a question is to use what's referred to as a 'block and bridge'. The politician acknowledges the question using a 'block' and then uses a 'bridge' phrase to move the topic to something they are more comfortable with. Common bridging terms are "I would say", "what I can tell you", "the reality is". The bridge allows them to transition the conversation to their own talking points.
The standout in last night's debate was "what I say", which Luxon used 12 times, up from the eight times he used in the TVNZ debate.
What was the vibe?
According to a language sentiment data dictionary, which rates words as either positive, neutral or negative, Luxon used a slightly higher percentage of positive words and lower percentage of negative words than Hipkins during the debate.
This was similar to the previous debate where Luxon said the most positive words at 4.5 percent. For negative words, Hipkins edged Luxon, with 2.8 percent of his words classed as negative compared to 2.2 percent in TVNZ's debate.
Luxon made light of Hipkins' more negative approach in last night's debate, even offering some analysis of his own, with one of his 76 "actually's" thrown in: "I gotta say this, this negativity from Chris isn't good for him, right? Actually, I mean, I feel like I should give him a hug or something. I mean, it's not working for him, right?"
How we analysed the data: The debate was transcribed by Otter.ai and then manually cleaned to improve accuracy. Cross talk - where debate participants spoke over each other was removed where words were unintelligible. Interjections when intelligible were included. Text was analysed using R and the Quanteda package.