AMNA NAWAZ: The historic primary season, which has ousted 13 incumbents in Congress, is winding down.

And among the debates it's raised is one over polling.

Results in some high-profile races differed greatly from public opinion polls.

And, earlier this week, one supposed public polling organization announced its results had all been faked as part of a social experiment.

Our Lisa Desjardins has more.

LISA DESJARDINS: Some recent results at odds with polling include the Michigan Senate Democratic primary, where progressive Abdul El-Sayed had double-digit leads in polling, but won over moderate Haley Stevens by just one point, and the Wisconsin Democratic race for governor.

Democratic Socialist Francesca Hong was out front in polling, but ultimately lost the race to David Crowley.

To look at what this means and what it doesn't, as well as how to best use polls, I'm joined by Courtney Kennedy, senior vice president of U.S.

affairs at Ipsos.

So many questions, Courtney, but, first, can you help overwhelmed voters understand, what are primary polls good at and not good at?

COURTNEY KENNEDY, Senior Vice President of U.S.

Affairs, Ipsos: Oh, gosh.

Primary polling is incredibly, notoriously difficult.

I mean, at best, they can give you a sense of how the voters are thinking about the candidates and perhaps which issues are really driving people's motivations and their candidate preferences.

But, again, primary polls are notoriously challenging.

And we see that the amount of error in primary polls is actually like three times higher... LISA DESJARDINS: Really?

COURTNEY KENNEDY: ... than the error you see in national, rigorous polls that we tend to see in the news day to day.

So they're really of a different kind, both in who's sponsoring them and how they're conducted.

LISA DESJARDINS: I know part of the issue is that often primary polls are done by state organizations.

Journalistic organizations have less money to spend on it.

I do want to talk to some of these specific cases.

Let's start with Michigan.

You believe the polls there in that Senate primary overemphasized one kind of Democrat.

How did that happen?

COURTNEY KENNEDY: Yes.

Well, to your point, the organizations who are funding these polls really can only afford relatively inexpensive methods.

And, in Michigan, we saw a dominant way to do polling was what's known as text-to-Web, which is where the pollster sends out text messages with a link to go take a Web survey.

It's cheap because there's no human labor involved, but you get what to pay for.

That methodology, and we saw it in the results, was clearly more effective reaching engaged, more progressive, younger voters, and really struggled to reach moderates in the state of Michigan.

And Michigan was really instructive because there actually was one higher-quality poll that used more rigorous phone interviewing, and it did significantly better, reaching Stevens supporters, reaching more moderates, and representing more moderate support.

So how the polls were done in Michigan really told the tale, and in that state they paid the price for using those less expensive methods.

LISA DESJARDINS: We are separating these cases because they are different.

In Wisconsin, we had a very strange animal, in that right before the primary election, we had three candidates drop out, then one come back in to ultimately win that race.

My question for you about that is, how did polling handle all that?

And was that the issue here, or no?

COURTNEY KENNEDY: That was an incredibly tumultuous race, as you pointed out.

What jumped out to me about Wisconsin was the level of undecided voters.

Just weeks out from Election Day, you had polls picking up 30, 40 percent undecided.

I mean, on one hand, you have to wonder, like, did people in Wisconsin even know that they had an election coming up?

But, on the other hand, it was an incredibly chaotic race, and so perhaps it's understandable that they were unsure what to do.

But that level of undecided voters was really striking and unusual.

LISA DESJARDINS: Wisconsin is also one of those states where a fake organization calling itself Median Strategies put out false polling data, later posting this statement here saying it was a social experiment on how polling spreads.

Neither we nor any national news organizations use their data, but this has raised a point.

How should people figure out if a poll is legitimate, worth their time?

COURTNEY KENNEDY: Well, polls need to be vetted, just like any other source, to your point.

So if you're curious if a poll is real, have you heard of the pollster?

Do they have a track record?

That's a big tell.

And if you go investigate the polling organization, is there a named person with credentials who is putting the their face on that poll?

And details are another big tell.

The more rigorous polls tend to be more transparent about how they're done.

They provide details about the sampling and the weighting and exactly how the questions are asked.

So the more detail, the more documentation, the better.

And if you don't see any people or credible organizations associated with the poll, that's a serious red flag.

LISA DESJARDINS: People don't realize every poll is different, especially now with technology, many different challenges, an exciting time, but a challenging time for pollsters.

Courtney Kennedy, thank you so much for joining us.

COURTNEY KENNEDY: My pleasure.