I ran an extensive focus group practice in my previous life in an ad agency. What Luntz does is treat the debate performance much the same way a snack food company will test potato chips. Everyone sample the product and a skilled moderator leads a discussion. On the other side of the one-way glass you have some other folks, if you're smart one of them is a research psychologist, who are not only listening but feeding questions to the moderator via an earpiece. You're also watching group dynamics. Is someone checking out of the discussion? Is someone dominating? In the end you may or may not get something useful.
The technique Luntz uses has been used by political consultants for years, the audience have a device, like a dial, that lets them rate what they are hearing in real time and then the aggregated reactions are overlaid with the video.
There are a couple of key points to keep in mind. First, it is impossible to create a group model in a focus group. On one project I ran for National Highway Traffic Safety Administration, the client wanted "two Hispanics, an elderly black woman, a three-toed red haired gypsy, etc." (just joking about the last one). What they were trying to do with at twelve person focus group was create an demographic sample. You can't do that. No matter how you choose them, you've just got a dozen plus people. Second, focus groups are only useful if you are using them in conjunction with a larger poll. In the potato chip example, the focus test isn't very useful unless we know potato chips are popular. Your group screening criteria would only include people who regularly ate potato chips.
Here is the entire focus group experience:
If you want to avoid the brain lesion, this is the bottom line
When asked for a show of hands, the majority of members agreed on Kasich, who they said was the “only adult in the room” and was “mature” in his discourse.
The voters revealed very negative feelings about the debate, saying it was an “embarrassment,” “shameful” and “despicable,” voicing disapproval over the tone of the discussion and the personal attacks that ensued. One even described the evening as a “schoolyard brawl.”
In fact, no one in the group felt the debate helped the Republican Party. “They are giving soundbite fuel for the Democrats,” one Michigan voter told Luntz.
The lowest point of the night, according to voter reaction, came when Republican presidential contender Marco Rubio slammed Trump University, which resulted in the billionaire businessman saying the Florida senator “couldn’t get elected dog catcher.”
“We’ve never tested something so low,” Luntz said, adding that Kasich, a governor, scored very high ratings when it came to foreign policy.
Fine. But none of these questions or answers are all that useful. Socially we are hardwired to avoid confrontation. So if I call Mitch McConnell a liar, you might say "gee, that was pretty harsh and unnecessary." It doesn't mean that you disagree it just means you don't like how I said it. It's pretty much like when your wife asks you, "do these jeans make my ass look big?" The other thing that happens in this environment is that your audience ends up expressing opinions that they think several million television viewers will find acceptable. This is a well known issue in telephone polling, if you have a human interviewer polling on a sensitive subject, i.e. race, sexual behavior, etc., the respondent is likely to give even an anonymous telephone interviewer the answer that they perceive is more socially acceptable. As an aside, I've seen focus group moderators destroy a research product by their own body language, word choice, inflection, etc. So what are the chances of a lot of people who know they are very visible agreeing that any of the personal attacks or insults were a good thing? Zero. Not going to happen.
What you witness with Luntz's focus groups is a reality show. No matter how skilfully it is conducted, the participants know they and their opinions are going to be exposed to the public and no one, other than Donald Trump, want's to beclown themselves before millions of people.