The Backfire Effect: Can fact-checking make false beliefs stronger?

[Regina] (0:00 - 0:03)
This model pushes the data to the limit. Oh, I love that line, Kristin.

[Kristin] (0:03 - 0:41)
Yeah, isn't that great? That actually is the author's way of saying, yeah, we know this is super shaky, but we're doing it anyway.

Welcome to Normal Curves. This is a podcast for anyone who wants to learn about scientific studies and the statistics behind them. It's like a journal club, except we pick topics that are fun, relevant, and sometimes a little spicy.

We evaluate the evidence, and we also give you the tools that you need to evaluate scientific studies on your own. I'm Kristin Sainani. I'm a professor at Stanford University.

[Regina] (0:42 - 0:47)
And I'm Regina Nuzzo. I'm a professor at Gallaudet University and part-time lecturer at Stanford.

[Kristin] (0:48 - 0:53)
We are not medical doctors. We are PhDs, so nothing in this podcast should be construed as medical advice.

[Regina] (0:53 - 0:58)
Also, this podcast is separate from our day jobs at Stanford and Gallaudet University.

[Kristin] (0:59 - 2:00)
Regina, today I want to talk about the backfire effect. This is the idea that when you try to correct someone's false belief, you might actually make them believe it more strongly. It's gotten a lot of media attention.

Have you heard of it?

[Regina]
I have, actually. It is very counterintuitive.

[Kristin]
Yeah, I got sidetracked on this topic while researching vaccine hesitancy for our HPV vaccine episode. It turned out to be such an interesting story that I'm now dedicating a whole episode to it. It's been a big deal for fact-checkers and journalists because it suggests that correcting misinformation might reinforce the misinformation instead of dispelling it.

So that's the claim we're looking at today, that the backfire effect is real. Along the way, we're going to hit some fun statistical topics, interactions, detecting problems in randomization, replication studies, and the importance of understanding your data at a very basic level. We'll also encounter a ghost study and do some statistical sleuthing.

[Regina] (2:01 - 2:04)
Statistical sleuthing with you. I cannot wait.

[Kristin] (2:05 - 2:38)
All right, researchers talk about two main forms of the backfire effect, the familiarity backfire effect and the worldview backfire effect.

[Regina]
And of course, we've got jargony names for these.

[Kristin]
Of course, academics have to have jargon.

How else are they going to keep their jobs and sound important, right? But these two terms are pretty easy to remember. The familiarity backfire effect is the idea that when people repeatedly hear a claim, even if it's labeled as false, they might start to believe it just because it feels familiar.

[Regina] (2:38 - 2:48)
Ah, so if we tell people it's false that vaccines cause autism, they might just remember vaccines cause autism and forget the whole false part. Exactly.

[Kristin] (2:49 - 3:29)
Our brains often confuse familiarity with truth. So even if you originally heard, this is a myth or this is false, just hearing that myth repeated multiple times might make it feel credible later. Interesting.

[Regina]
And the worldview backfire?

[Kristin]
That's when a correction threatens someone's core beliefs, political, cultural, or religious. The theory is that when people feel their worldview is under attack, they start mentally counter-arguing.

They push back in their minds, defending their position so forcefully that they end up believing the false claim even more as a way to protect their identity.

[Regina] (3:30 - 3:54)
Oh, that makes sense. It's about reinforcing your view of the world. So like if someone is vegan and you tell them it's false that almond milk is better for the environment than dairy is, that might feel like a threat to all their vegan values.

And they become even more convinced that dairy is worse. But non-vegan, they're not going to have that doubling down effect. They might be more open-minded.

[Kristin] (3:55 - 4:58)
That is a great example, Regina. Though I have to say, I've never really understood almond milk. Now, not everyone agrees on what counts as a backfire effect.

So here's how we're defining it for this episode. A backfire effect occurs when correcting a false claim inadvertently increases belief in that claim compared to presenting the false claim alone or presenting nothing at all.

[Regina]
Got it. Okay, how are you going to start us off? Do you have a study?

[Kristin]
Yes, I want to start by talking about a 2010 paper that basically put the backfire effect on the map.

It made a huge splash because it claimed to provide evidence of the worldview backfire effect.

[Regina]
Worldview meaning identity threat. Exactly.

[Kristin]
The lead author of the paper was Brendan Nyhan. He's now a professor at Dartmouth, my alma mater, and he has an interesting history. Back in 2001, he co-founded a political fact-checking site called SpinSanity.

[Regina] (4:58 - 5:05)
Wow, a 2001 fact-checking website, and that was early internet era there.

[Kristin] (5:05 - 5:23)
Yeah, they actually predated sites like factcheck.org. It only ran until 2004, but Nyhan and his co-founders also wrote a book in 2004 that made the New York Times bestseller list. It's called All the President's Spin, George W. Bush, the Media and the Truth.

[Regina] (5:24 - 5:35)
Great book title, and now I'm nostalgic for the early 2000s. So this means he has been interested in correcting misinformation for a long time.

[Kristin] (5:35 - 5:49)
Oh, yeah. After that, he went on to Duke for a PhD in political science, and his famous 2010 paper actually came out of research that he did for his dissertation, and this paper has had a huge ripple effect.

[Regina] (5:50 - 6:05)
Were fact-checkers and journalists like completely freaking out at this point? Because if the backfire effect is real, then them debunking myths could actually do more harm than good.

[Kristin] (6:06 - 6:20)
They were freaking out, yes, exactly.

The paper has been cited over 4,000 times, it's been referenced hundreds of times in the media, and it's had a long shelf life. It got a lot of attention in 2010, when it first came out, but then there was a resurgence of interest around 2015.

[Regina] (6:20 - 6:24)
Hmm, what happened in 2015? I am guessing Trump?

[Kristin] (6:25 - 7:08)
Exactly. That's when we entered the alternative facts era. Misinformation was everywhere, and people were really interested in understanding misinformation, and this paper became a go-to citation.

Fact-checkers were even warned to not repeat misinformation when correcting it. We are going to dissect that 2010 paper in just a minute here, but before we do that, Regina, I want to talk about an earlier paper from 2007. It's sort of a ghost or phantom paper.

[Regina]
A ghost paper? What is a ghost paper?

[Kristin]
It's this unpublished study that everyone cites as evidence of the familiarity backfire effect, but no one ever seems to have actually got their hands on this study.

It's a ghost.

[Regina] (7:09 - 7:16)
But wait a minute. So they are citing an unpublished study with no peer review?

[Kristin] (7:16 - 7:38)
That's right. It was never published. Sometimes people cite the unpublished version directly as unpublished paper.

Other times they cite a review article that contains a short summary of the ghost paper, and that review article was co-authored by Norbert Schwarz, and he was actually one of the co-authors of the ghost paper, so presumably he should know what was in the ghost paper.

[Regina] (7:38 - 7:43)
You would think. Um, what was the ghost paper actually about?

[Kristin] (7:43 - 8:07)
It was an experiment where people were randomly assigned to see a CDC, Centers for Disease Control, flyer about the flu vaccine, or to see nothing. The control group just saw nothing. The flyer contained myths and facts about the flu vaccine.

For example, it contains the statement, the side effects are worse than the flu, but that's labeled very prominently as false.

[Regina] (8:07 - 8:15)
Ah, and this is where people might just remember the statement itself and somehow forget that it was labeled as a myth.

[Kristin] (8:16 - 9:09)
That's the worry. After they saw the flyer, they waited 30 minutes to let people's memories settle because the familiarity backfire effect has to do with faulty memory. And then they gave both groups a questionnaire about their attitudes toward the flu vaccine.

And when people cite this paper, they claim that it shows two key backfire effects. Supposedly, the group that saw the facts and myths flyer rated the vaccine as less important and said they were less likely to get the vaccine compared with the control group. In fact, that's exactly how Schwarz himself summarizes the results in that review article that he coauthored.

He writes, the facts and myths flyer backfired after a delay. These participants reported less favorable attitudes toward flu vaccination and lower behavioral intentions than control participants who read no flyer at all.

[Regina] (9:10 - 9:33)
Lower behavioral intentions. Does that mean just lower intention to actually get the vaccine?

[Kristin]
Yes, exactly.

[Regina]
Okay, Schwarz was saying the backfire effect was real. And since, as you pointed out, he was one of the coauthors on the ghost study, you would think that he got the results right. You would think, but of course no one could check it because no one seems to have a copy of this phantom paper.

Kristin, did you go looking for it?

[Kristin] (9:33 - 9:53)
Of course I did. I took this as a challenge to find it.

And through the magic of the Wayback Machine, I found a copy. The paper is titled, Myths and Facts About the Flu. Health Education Campaigns Can Reduce Vaccination Intentions.

[Regina] (9:53 - 9:57)
And we can put a copy in the show notes of that PDF that you found.

[Kristin] (9:57 - 10:15)
We will definitely put a copy in the show notes so everybody can read it. It appears to be a draft that's ready or near ready for journal submission. So it doesn't appear to be a draft that's unfinished or in its very early stages.

In fact, the title of the PDF, it's got the word JAMA in it.

[Regina] (10:16 - 10:24)
So they may have submitted this paper to JAMA, the Journal of the American Medical Association, but then gotten rejected.

[Kristin] (10:25 - 10:31)
I think so because the paper was never published. So presumably anywhere it was submitted, it got rejected.

[Regina] (10:32 - 10:34)
Which means it may have had some problems.

[Kristin] (10:35 - 10:59)
Yes, exactly. And in fact, I did notice some red flags when I looked at the paper. I'll put those in the show notes.

But Regina, do you want to hear the really dramatic twist here?

[Regina]
Of course.

[Kristin]
In this copy of the paper, there is no evidence to support those two backfire effects that everyone who cites the paper, including Schwarz, claims that the paper shows.

[Regina] (10:59 - 11:09)
What? The findings Schwarz reported do not match what you found in his own unpublished ghost paper.

[Kristin] (11:09 - 11:21)
Nope. The facts-and-myths group rated the importance of the flu vaccine on average three out of nine, whereas the control group rated it four out of nine. But this difference was not statistically significant.

[Regina] (11:21 - 11:29)
Ah, and without statistical significance, they cannot claim to have found a difference. Statistically not allowed.

[Kristin] (11:30 - 11:52)
Exactly. And it's worse for intention to vaccinate. The facts-and-myths group rated their likelihood of getting the vaccine next winter on average 3.2 out of nine, whereas the control group rated it as 3.1 out of nine, virtually identical. There was no difference between the groups in the intention to vaccinate, even though all the subsequent citations of this paper claim that there was a difference.

[Regina] (11:53 - 12:04)
So the one study everyone cites as original proof of the familiarity backfire did not actually show any evidence of this effect.

[Kristin] (12:04 - 12:23)
That's right. This paper turns out to be a myth itself. This is the problem that we talked about with bad citations in the sugar sag episode, Regina. You should never trust that someone else's summary of a study is accurate, even if it was one of the authors of the study.

You should always go back to original sources and read and summarize the paper yourself.

[Regina] (12:23 - 12:34)
Except here, Kristin, there was no source to go back to. And, OK, you cannot be sure that you actually got the final version of the paper, right?

[Kristin] (12:35 - 13:03)
That's true. I don't know if the copy that I found on the Wayback Machine is the actual final version of this paper, but it would be very strange if the results changed that dramatically between drafts. So my conclusion is this ghost paper does not actually provide any evidence of the backfire effect, despite the fact that it's cited repeatedly in the literature as proof of the backfire effect.

All right, Regina, now we're going to dig into that famous 2010 paper.

[Regina] (13:03 - 13:06)
That's the one you mentioned at the beginning that made the big splash.

[Kristin] (13:06 - 13:30)
Yes. Brendan Nyhan is the first author, as we talked about, and Jason Reifler is his co-author. And this one was actually published, correct?

Yes. This study was peer-reviewed and published, so we're already doing better than the ghost paper. They ran two different studies that are described within this one paper.

In the first study, they collected data in 2005 on 130 students at a Catholic university in the Midwest.

[Regina] (13:31 - 13:35)
Students again. Were these students getting extra credit by any chance?

[Kristin] (13:35 - 13:41)
Yes. These were psych students participating in the study to get course credit, exactly.

[Regina] (13:41 - 14:00)
Kristin, you and I have talked about how psych studies often rely on these college students, which means these samples are often not representative of the general population. And you mentioned this was a Catholic university. Maybe they weren't even representative of college students in general.

[Kristin] (14:00 - 14:14)
Right. They may have been more conservative than typical college students, for example. Regina, we're going to need to do a little history lesson here. Do you remember the whole controversy about Iraq and weapons of mass destruction back in the early 2000s?

[Regina] (14:14 - 14:46)
Ooh, this is a little memory game. I know I said I was feeling nostalgic, but, okay, I do remember. This was after 9-11 when the American government claimed Iraq had weapons of mass destruction, WMDs, and that Iraq might give those weapons to terrorists.

And those WMDs were the main reason the government gave for invading Iraq. And that was in 2003. But then later, whoops, it was revealed that they were wrong.

Those WMDs did not actually exist.

[Kristin] (14:47 - 15:27)
That's a great summary, Regina, especially for our young listeners who may not remember this controversy because they were too young at the time. All right, back to the study now. Participants read a news article that falsely implied that Iraq had weapons of mass destruction, WMDs.

The participants were randomly assigned to receive just the news article or the news article followed by a correction that clarified that Iraq did not have WMDs. After they read the material, participants had to rate their agreement with the statement that Iraq had weapons of mass destruction. They rated this on a one-to-five scale, where one meant strongly disagree with the statement, and five meant strongly agree.

[Regina] (15:27 - 15:33)
So for this outcome, higher numbers mean more belief in the falsehood. Yes.

[Kristin] (15:33 - 16:47)
They also asked the participants to report their political ideology in one of seven categories. Very liberal, liberal, somewhat liberal, centrist, somewhat conservative, conservative, or very conservative.

[Regina]
So seven categories.

[Kristin]
That's right. So that was the first study. The second study they conducted in 2006, and this involved 197 students from the same college.

And Regina, this is great. They did an internal replication. They tried to replicate the exact same experiment on Iraq and WMDs that they had conducted in 2005.

[Regina]
Ooh, that is great. We love a replication.

[Kristin]
Yes, we love replication.

They also added two new false statements. One, that tax cuts increase revenue. Not quite sure how the math works on that.

And two, that George W. Bush canceled all stem cell research. And that second false statement, I believe, was put in there to appeal to liberals more than conservatives.

And then the participants had to rate their belief on those false statements on that one-to-five scale. And Regina, overall, they didn't find any difference between the two groups, between the correction group and the control group, in any of the four experiments.

[Regina] (16:48 - 16:49)
Well, that does not sound very exciting.

[Kristin] (16:51 - 17:14)
That's right. How did this get big headlines? But remember, they were studying the worldview backfire effect.

And the worldview backfire effect is not about what happens overall. It's about what happens in certain subgroups, like liberals or conservatives. So now I want to talk about the results when they divided this up by political ideology.

This is going to involve something called interactions.

[Regina] (17:14 - 17:20)
Oh, I cannot wait to hear what they found and to talk about interactions. I love those. But let's take a short break first.

[Kristin] (17:29 - 17:40)
Regina, I've mentioned before on this podcast our introductory statistics course, Demystifying Data, which is on Stanford Online. I want to give our listeners a little bit more information about that course.

[Regina] (17:40 - 17:50)
It's a self-paced course where we do a lot of really fun case studies. It's for stats novices, but also people who might have had a stats course in the past but want a deeper understanding now.

[Kristin] (17:50 - 18:03)
You can get a Stanford professional certificate as well as CME credit. You can find a link to that course on our website, normalcurves.com. And our listeners get a discount.

The discount code is normalcurves10. That's all lowercase.

[Regina] (18:12 - 18:24)
Welcome back to Normal Curves. We are examining the claim that the backfire effect is real. And we were about to discuss results from the 2010 paper that made a huge splash.

[Kristin] (18:25 - 18:40)
All right, let's start with that first study, the data they collected on the 130 students in 2005. Remember, the two groups corresponding to this study had to rate their agreement with the statement that Iraq had WMDs on this one-to-five scale.

[Regina] (18:40 - 18:47)
OK, and you said before that there was no difference between the correction and the control groups on this outcome.

[Kristin] (18:48 - 19:12)
Right. The mean belief for the correction group was 2.4 out of 5 and the mean for the control was 2.4 out of 5, virtually identical. But then they asked a different question.

Does the effect of the correction depend on your politics? Like, maybe the correction works for liberals, but backfires for conservatives because it threatens their worldview. And Regina, how do you think they tested that?

[Regina] (19:12 - 19:25)
Well, we are talking here about interactions. They are looking at how two things work together, politics and fact-checking. So, interaction effects in a regression model.

[Kristin] (19:26 - 19:39)
That's exactly right. Regina, we talked about interactions in an earlier episode and just want to give you now the big picture. An interaction just means that the effect of the intervention is different in different groups. But interaction comes in different flavors.

[Regina] (19:39 - 19:55)
In the dating wishlist episode, we talked about synergistic interactions. I called it the chocolate peanut butter effect. Two good things that are even better together.

But now, Kristin, in this case, you were talking about something slightly different.

[Kristin] (19:55 - 20:09)
Yes. Rather than a synergistic interaction, we have what's called a crossover interaction. It's not about synergy. It's that the intervention has the opposite effects in different groups.

Regina, do you have an analogy for this one for us?

[Regina] (20:10 - 20:16)
I did. I thought about this ahead of time. I think I'm going to go with a sex example rather than dessert example this time.

[Kristin] (20:16 - 20:16)
I'm excited.

[Regina] (20:17 - 20:39)
We've not talked about sex yet in this episode. I feel like it's time. So let's go with the dirty talk in the bedroom.

Because if you are a dirty talk kind of person, you get a little foul language in the bedroom. Super sexy. But if you are not a dirty talk kind of person, then that has the opposite effect.

It completely kills the mood.

[Kristin] (20:40 - 21:54)
Yeah, definitely different people might react to that really differently. All right. Back to WMDs now from dirty talk.

In this study, they did find a statistically significant interaction. Let me give you now the big picture interpretation of the interaction they found. They found that if you were on the liberal end of the spectrum, the correction worked.

Belief in WMDs went down. But if you were on the conservative end, it backfired. Belief in WMDs actually went up.

[Regina]
Ooh, that is the backfire.

[Kristin]
Right. But I want to get under the hood a little, Regina, and walk through how that actually played out in the model and what it looks like statistically, mathematically.

So let's talk about the model. First of all, they used a linear regression. The outcome variable for that regression was this belief that Iraq had WMDs rated on this 1 to 5 scale.

And the model contains three main predictors. The first predictor is are you in the correction group or the control group? The second predictor is your political ideology.

And the final main predictor is the interaction between what group you're in and your political ideology.

[Regina] (21:54 - 22:16)
Okay, so three main predictors, which experimental group you're in, that's a binary, categorical. Belief in the statement on a five-point scale, and political ideology on a seven-point scale. And those are also categorical, Kristin, but they are treating them as numbers, as integer scales.

[Kristin] (22:16 - 22:26)
Yeah, that's right. Political ideology is treated as a seven-point scale with one point added every time you shift one category to another. So, it's one category to the right in the conservative direction.

[Regina] (22:27 - 22:36)
Hmm, there are potential problems with treating categories as numbers, but I think we'll save that discussion for another episode because it is one of my rants.

[Kristin] (22:36 - 23:35)
I have heard you rant on this topic, Regina, yes. But we don't have enough time to get into it today, so we're going to save this great topic for a future episode. This is, though, how they can use a linear regression model and fit lines.

We need numbers to do that. So, we're just going to go with it here. Okay, that's the regression model they fit, and it's probably easiest to understand the output of the model by having me draw a mental picture.

So, imagine we've got a scatter plot. On the vertical axis, it's this one to five, how much do you believe that Iraq had WMDs? On the horizontal axis, it's political ideology coded on this seven-point scale from very liberal to very conservative.

And then what we're doing is we're actually plotting two regression lines. Our model, because it contains an interaction, is actually giving us two different regression lines, one for the correction group and one for the control group. Both of those lines, in this case, ended up with a positive slope.

Can you interpret that for us, Regina?

[Regina] (23:36 - 23:51)
Okay, positive slope, that means as people get more conservative, because you said greater values was more conservative, as they get more conservative, they are more likely to believe that Iraq had weapons of mass destruction, which makes sense.

[Kristin] (23:52 - 24:28)
Conservatives are more likely to believe this in the first place. The key to the interaction, though, is that the two lines have different slopes. The correction group has a steeper slope than the control group, and the two lines cross.

If you are on the liberal side of the graph, the left side of the graph, the correction line is actually below the control line, meaning the correction worked. People reduced their beliefs in WMDs. But if you're on the conservative side of the graph, the right side of the graph, the correction line is above the control line.

It's backfiring. If you got the correction, belief in WMDs actually went up.

[Regina] (24:28 - 24:46)
Oh, Kristin, I love this. Thank you for walking us through this. But a question, technically, if a researcher says that they found an interaction, what they're really saying is that they found the slopes of those two lines are significantly different from each other, and that was the case here?

[Kristin] (24:47 - 25:16)
Yes, the slopes of the two lines were significantly different. Correct. And, Regina, just to give you a sense of the size of the effect, in addition to the statistical significance, the model predicts that if you were in the very liberal group, getting the correction would reduce your belief in WMDs by almost one point on that five-point scale.

But if you're in the very conservative group, the model predicts that getting the correction increases your belief in WMDs by over a point on that five-point scale.

[Regina] (25:16 - 25:24)
Interesting. So not a super huge effect, but not nothing, either. It's a moderate effect.

[Kristin] (25:24 - 25:58)
Yeah. Now, I do have a complaint, Regina. They did something in the paper that's a pet peeve of mine. Uh-oh.

First, the good thing, they give us the full results of the regression model. They give us the equation of the model, and they plot the model, which, of course, is just these nice, straight, perfect lines. But the problem is that is just telling us the results of the model.

They actually never give us anything about the underlying data. And I like to see the underlying data to understand if the model is really a good representation of the data or not. And there's some really simple data that I would have liked to have seen here.

[Regina] (26:00 - 26:02)
Yes, it is always nice to undress the model.

[Kristin] (26:03 - 26:27)
Yes. This study actually had just 14 groups. There are seven political ideologies crossed with, you're either in the correction or the control group.

That's just 14 groups. And they could have shown us the mean WMD belief for each of those 14 groups. Then we really would have understood what was going on in the data.

They also could have just plotted the raw data rather than plotted the smoothed over regression lines.

[Regina] (26:28 - 26:30)
Yes, I vote for naked data.

[Kristin] (26:32 - 26:45)
Yes, especially when we have small samples like this. I want to see the actual data. I mean, here, they did have 130 people overall.

But once you split that into 14 groups, you are working with very small numbers.

[Regina] (26:45 - 26:54)
Yes, you are. And that means the model might just be chasing noise or smoothing over little quirks in the data. Exactly.

[Kristin] (26:54 - 27:36)
And let me make a little analogy here. It's like when I tell my kids to clean their room. I walk ahead and it looks clean.

Floors clear, beds made, things appear in place. That's kind of like the regression model. Neat, tidy, smoothed over.

But I'm not buying it until I check under the bed, in the closet and in the drawers. Because half the time, all the junk that was all over has just been crammed out of sight in a big mess in drawers or in the closet or in boxes in the corner. So it looks clean on the surface, but underneath it's a complete mess.

And that can happen with regression models. It might look great on the surface, but there might be a lot of messiness underneath. That's why I always want to see the raw data.

[Regina] (27:37 - 27:39)
I love that analogy so much.

[Kristin] (27:39 - 28:03)
Thanks, Regina. Now, fortunately, Nhyan, I have to admit, I'm a fangirl. And not just because I'm a Dartmouth alum, but because on his website, he lists all his publications.

And for most of them, he provides both the dataset, the raw data, and this data code. And he has data going all the way back to 2010, including for this paper.

[Regina] (28:04 - 28:11)
No way. He is precocious. He was doing open data before open data was even a thing.

[Kristin] (28:11 - 28:30)
Yeah, this is amazing that I was able to get data from a study that's 15 years old, and I was able to look under the hood. Regina, the first thing I did with the data is I simply replicated the models in the paper. I re-ran the regression models to make sure that I got the same numbers that he got.

And guess what? I did. The numbers match.

[Regina] (28:30 - 28:46)
That is actually not a small feat, I want to point out. This is a very particular type of replication you're doing called computational replication, and it did not always work out with such great results as this. So kudos to the authors, kudos to you.

[Kristin] (28:46 - 29:15)
Yeah, you'd be surprised how often this simple replication actually fails, but it gives me a lot of trust in the paper that I can just match the basic numbers reported in the paper.

[Regina]
Did you go further with the data? What'd you find?

[Kristin]
I did find an imperfection. There were 60 people in the correction group, but 70 in the control group. It wasn't quite even.

This tells me that their randomization scheme was probably some kind of really simple scheme, Regina, like a virtual coin flip.

[Regina] (29:15 - 29:39)
You and I have talked about the importance of randomization schemes because you can end up with unbalanced groups like this if you do simple randomization. What they really should have done is block randomization where you randomize in small blocks instead of the whole thing all at once, and that's so you don't end up with too many people in one group and not enough in the other.

[Kristin] (29:40 - 30:30)
Yeah, it would have been a little better had they done something like that because we would have ended up with more balanced numbers. It's not terrible, though. All right, the next thing I did was to actually look at just some simple descriptive statistics.

Regina, we've talked before about the importance of descriptive statistics. Here, I just looked at the 14 groups we have. Again, seven political ideologies, two experimental groups.

I simply calculated the means for this WMD belief score in each of the 14 groups so I could get a feel for what was going on. And here's what I found. Among the two most liberal categories of students, very liberal and liberal, there was almost no difference between the correction and control groups.

The effect we see on the liberal side is only coming from that somewhat liberal group. In that group, the mean belief was 2.2 in the control group, and it dropped to 1.4 in the correction group.

[Regina] (30:31 - 30:39)
It is sounding like the pattern is not quite as tidy and consistent as the regression model might suggest. That's right.

[Kristin] (30:39 - 31:27)
If we look in the middle, in the centrist and somewhat conservatives, the correction group was about a half point higher than the control on average, so it was a small backfire in those groups. But the real meat of it, where that backfire effect that we see in the model is really coming from, was actually from the conservative category. In those students, the mean jumped from 2.8 in the control group to 4.2 in the correction group. But here's the thing. There were only five people in each of those groups, correction and control, so just 10 total.

[Regina]
Ooh, that is not a lot.

[Kristin]
That's right. And if we look at the very conservative group, there were only four very conservative students in the whole study. And guess what?

None of them happened to be in the correction group. All four of the very conservatives ended up in the control group.

[Regina] (31:28 - 31:34)
So actually, for that very conservative group, we really don't have much to go on.

[Kristin] (31:35 - 31:45)
Yeah, even though the model makes predictions about that group, we don't have any data there. So, Regina, overall, the pattern is there, but it's not nearly as smooth and tidy as the model makes it look.

[Regina] (31:46 - 31:47)
Just like your kid's bedroom.

[Kristin] (31:48 - 32:23)
Yes. And when you're looking at interactions, having small numbers in some of the subgroups like this is a problem, especially when the study design relies on randomization. The randomization is really important here because they are using what's called a post-test study design.

That means they compare the correction and control groups only after the intervention. So we have to assume that the groups were similar to begin with. If they started out with different beliefs about Iraq and WMDs, then the differences we see afterward might just reflect those initial differences, not the effect of the correction.

[Regina] (32:24 - 32:30)
Right. Randomization is supposed to make the groups equivalent at the start. That is the whole point.

[Kristin] (32:32 - 33:06)
If you randomize 130 people into two groups, you'll probably get pretty good balance overall. If the sample size is large enough, you tend to get pretty good balance. But once you start slicing the data into smaller subgroups by politics, you can run into trouble.

Some of those subgroups, like very conservative participants, are tiny. And with such small numbers, it's easy to get imbalances just by chance where maybe the correction group happened to start out with stronger beliefs in WMDs than the control group. That alone could explain the difference we see after the intervention.

So overall, I'd say there's something here, but it's not terribly robust.

[Regina] (33:07 - 33:15)
Exactly. And this is why we need replication. Because, Kristin, didn't you say they did their own replication when in 2006?

[Kristin] (33:16 - 33:58)
Yes, they did a direct replication. They repeated the exact same experiment a year later in a slightly larger group of students, 197 students. And guess what they found, Regina?

[Regina]
Oh, something super exciting?

[Kristin]
Depends on your definition of exciting. They did not replicate the findings.

In fact, they found the opposite.

[Regina][
The opposite?

[Kristin]
Yeah, there was absolutely no backfire effect.

And in fact, in this data set, the conservatives actually decreased their beliefs in WMDs more when they got the correction than the liberals did. The correction actually worked better for conservatives in this version of the study, and there was certainly no backfire.

[Regina] (33:59 - 34:13)
Interesting. So now I'm wondering, though, Kristin, could it be because by 2006, the Iraq war was not as much of a hot-button issue as it had been the previous year?

[Kristin] (34:13 - 34:39)
Absolutely. Beliefs evolve, politics shift. This was a whole year later, and this was not as salient of an issue anymore. So it could be that both of these studies are right, and there was just some kind of change in the culture at the time.

It also could be that one or both studies are wrong. And, Regina, since they didn't find what they wanted to in that replication study, in that second study, what do you think they might have done next?

[Regina] (34:40 - 34:53)
Kristin, did they go poking around in their data just like the bad boyfriend post hoc ex that we talked about who refuses to give up on the relationship or the study?

[Kristin] (34:53 - 35:34)
They sure did. They ran a post hoc analysis, and to their credit, they clearly labeled it as post hoc. So here's what they did. They said, OK, let's only look at people who said they really cared about the Iraq war, who identified the Iraq war as the most important issue to them.

That was one of the survey questions. When they restricted the analysis to that subgroup, then they found that the conservatives did double down when corrected. Here's the thing, though, Regina.

There were only 34 people in that entire analysis, of whom only eight identified as anywhere on the conservative spectrum. And just four were in the correction group and four in the control group.

[Regina] (35:35 - 35:40)
Oh, four in each group. That is way, way too little to draw any conclusions from.

[Kristin] (35:40 - 35:59)
Yeah, and the authors clearly knew that because they wrote this great line in the paper. I love this. They say, This model pushes the data to the limit, since only 34 respondents rated Iraq the most important issue, including eight who placed themselves to the right of center ideologically.

[Regina] (35:59 - 36:05)
This model pushes the data to the limit. Oh, I love that line, Kristin.

[Kristin] (36:05 - 36:40)
Yeah, isn't that great? That actually is the author's way of saying, yeah, we know this is super shaky, but we're doing it anyway. I do appreciate the self-awareness. All right, Regina, remember, they also looked at two other beliefs in this study, and they did find a worldview backfire effect for the belief that tax cuts increase revenue.

In that case, the conservatives did double down after seeing the correction. It was statistically significant, but I'll tell you, the magnitude was small. It was even smaller than the effect that we saw in that first Iraq WMD study.

And for stem cells, they didn't see anything. No backfire.

[Regina] (36:41 - 36:50)
Overall, this is not looking very impressive. I mean, there might be something there, but not terribly robust because we're not seeing it in every situation.

[Kristin] (36:51 - 36:53)
Yeah, I'd call it interesting, but flimsy.

[Regina] (36:56 - 36:58)
But of course, people ran with it anyway, didn't they?

[Kristin] (36:59 - 37:46)
Yep, it got totally blown out of proportion. And I'm not saying that this is the authors’ fault. I think it's just that the reaction to the paper really outpaced the data.

And this idea of the backfire effect took on a life of its own. I also want to mention another researcher who got very interested in the backfire effect around the same time, Ulrich Ecker. He's a cognitive psychologist at the University of Western Australia.

And in 2012, he co-authored a review paper with Norbert Schwarz. Do you remember who that is, Regina?

[Regina]
Oh, that's the ghost study author.

[Kristin]
Exactly. And in that review, they even cite the ghost study. That paper focused on misinformation in general, but they spent a good chunk of it arguing that the backfire effect was real and they cited evidence that they believe supported the backfire effect.

[Regina] (37:46 - 37:50)
So Ecker was on board at this point.

[Kristin] (37:50 - 37:58)
Definitely. And we'll see later that he goes on to run some important backfire studies himself, but this particular paper was just a review, so he didn't collect any data.

[Regina] (37:58 - 38:00)
So just a summary of what was out there.

[Kristin] (38:01 - 38:22)
Exactly. And, Regina, around the same time, Nyhan and Reifler, the authors of that famous 2010 study, they were also pushing the research forward. They ran more studies about the worldview backfire effect.

And they found? Sometimes they found an effect. Sometimes they didn't.

For example, they did some studies in vaccines. They looked specifically at people who were vaccine hesitant.

[Regina] (38:23 - 38:27)
Vaccine hesitant. That is the group you'd most expect to double down.

[Kristin] (38:28 - 39:18)
Right. But the results were kind of a mixed bag. When they gave this group corrections like vaccines don't cause autism, it didn't seem to increase their belief that vaccines cause autism, but it did sometimes reduce their intention to vaccinate. Around the same time, other researchers were also getting on the backfire bandwagon and starting to study backfire effects.

And there were a number of studies specifically about the worldview backfire effect. And they did find some flickers. There were some isolated findings where they found something positive, but no consistent pattern and kind of a lot of null results.

Still, during this time, the mid-2010s, both Ecker and Nyhan became key public voices on misinformation. They gave talks. They were widely quoted.

And they often discussed the backfire effect.

[Regina] (39:19 - 39:28)
So it was a big part of both their public and professional identities. And that is where we are in the mid-2010s.

[Kristin] (39:28 - 39:43)
Exactly. A lot of people talking about backfire and believing in it. And Regina, I want to talk in detail about one more paper that came out in 2017 because this paper claimed to find evidence for the familiarity backfire effect.

[Regina] (39:43 - 39:44)
Oh, I can't wait. Time for a short break first.

[Kristin] (39:52 - 40:04)
And Regina, for those who are looking for even more learning opportunities, because of course, this is a podcast about learning. My department at Stanford offers a remote certificate program in epidemiology and clinical research.

[Regina] (40:04 - 40:09)
This is a fabulous program because they get, Kristin, even more of you.

[Kristin] (40:09 - 40:32)
Well, these are real, actual, you're enrolled in Stanford courses for Stanford credit on a Stanford transcript. So these are actual courses that we offer to our master's and PhD students, but we have figured out how to offer them remotely so that you can get a remote certificate. So it's a little bit more of a commitment than some of our other learning, teaching offerings, but well worth it if you're thinking of that career move.

[Regina] (40:33 - 40:36)
And very convenient because you can do it from anywhere in the world.

[Kristin] (40:37 - 40:46)
You can find out more about the certificate program on online.stanford.edu. Just search up the epidemiology and clinical research graduate certificate program. Or on our website normalcurves.com.

[Regina] (40:53 - 41:05)
Welcome back to Normal Curves. Today, we're talking about backfire effect and we're just about to talk about a 2017 paper that claimed to have found evidence for familiarity backfire effect.

[Kristin] (41:06 - 41:33)
This paper was published in PLOS One. The first author was a PhD student and this paper got a lot of attention because it claimed to have found this elusive familiarity backfire effect. Up until now, there was really no evidence for that effect other than that ghost paper.

And in a lot of ways, the familiarity backfire effect is actually harder to find because with a familiarity backfire effect, the correction makes everybody backfire, not just the people who are super invested in the issue.

[Regina] (41:34 - 41:37)
Hmm, so that makes sense. How did they study this then?

[Kristin] (41:37 - 42:36)
They studied 134 students. These were both undergraduate and graduate students. Everyone filled out a baseline survey and then they were randomly assigned to one of four conditions.

There was a myths and facts correction group.

[Regina]
Hmm, kind of like the ghost paper.

[Kristin]
Yes, exactly.

Except that it was not about the flu vaccine. It was more about the MMR vaccine. The participants received a booklet of statements that were labeled as myths or facts.

For example, one of the myths says a 1998 study showed that the MMR vaccine causes autism, but that's juxtaposed with a fact that says there is no evidence of a link between the MMR vaccine and autism. There was also a control group. The control group saw a flyer about medical errors, so nothing about vaccines.

And then there were two other groups that got different corrections. One got a graphical correction and one got a fear-based correction. I'm not going to focus on those.

I'm only going to focus on the myths and facts correction group versus the control group because that's where the backfire effect was claimed.

[Regina] (42:37 - 42:44)
Okay, so it's a myths and facts group versus a control group. Again, similar to the ghost study.

[Kristin] (42:44 - 42:58)
That’s right. So, after reading their assigned flyers, participants were surveyed immediately, and there was no difference between the myths and facts group and the control group at that point. Then they built in a delay. They tested the participants again seven days later.

[Regina] (42:58 - 43:08)
A delay because with this kind of backfire, familiarity backfire effect, it's about faulty memory, bad memory. It's not about doubling down or defensiveness.

[Kristin] (43:09 - 43:37)
Right, so they wanted to leave time for memory to fail. Even though they started with 134 participants, only 120 of them filled out the follow-up survey seven days later. So, the paper contains data only on those 120 people.

And as I mentioned, they claim to have found a backfire effect. They said that after the delay, the myths and facts group more strongly agreed with the statement that vaccines cause autism, and they reported lower intent to vaccinate than the control group.

[Regina] (43:37 - 43:49)
And this is evidence of the familiarity backfire effect because they did not segregate participants by political ideology or other worldviews. Exactly. So, this looks promising then.

[Kristin] (43:49 - 44:14)
Yes, except as with the Nyhan paper, these results really hinge on the randomization being done properly. It also used a post-test design. We're only getting their opinions after the intervention.

So, in order for us to be able to draw conclusions, we need to know that the groups were similar to start. Otherwise, what we're seeing in the results could just be an artifact of one group had more concerns about vaccines than the other at baseline.

[Regina] (44:15 - 44:21)
And that is why we randomized to make sure the groups are balanced in their belief right from the start.

[Kristin] (44:21 - 44:25)
Exactly. But, Regina, the paper gives us no details about the randomization.

[Regina] (44:25 - 44:38)
Hmm, not good. You mentioned that the journal is PLOS ONE, and that journal requires authors to submit their data. So, Kristin, did you go data dumpster diving?

[Kristin] (44:39 - 45:36)
I sure did. And to the author's credit, they did provide the data. Yay. So, I downloaded the data, and as with the Nyhan paper, the first thing I did was just to replicate some of the values that they reported in their paper.

And I was able to replicate their results.

[Regina]
Okay, good check.

[Kristin]
Yep, very good.

All right, the next thing I did was to compare the four groups on some basic descriptive statistics. The first thing I looked at was just something very simple. I looked at the sample sizes for the groups, because unfortunately, the paper never reports the sample sizes for the different groups.

When I looked at those sample sizes, Regina, I found something really odd. I found that there were exactly 30 people in each of the four groups. And the reason that this is odd is because, remember, they randomized 134 people on the first day of the experiment, but only 120 came back.

So it's really coincidental and strange that they were able to end up with exactly 30 in each of the four groups when we had lost 14 participants.

[Regina] (45:37 - 45:56)
Ooh, that is just so perfect that it's suspicious, isn't it? It's normal to have people drop out of a study, right? That happened, but they shouldn't all be dropping out randomly.

They are not going to drop out in just the right pattern so you get a perfect 30 in each of the four groups. That's just unbelievable.

[Kristin] (45:57 - 46:08)
Yeah, and it gets worse, Regina, because I also looked at gender. Turns out that one of the groups had 19 women in it, but the other three all had exactly 18 women.

[Regina] (46:09 - 46:18)
Also, too balanced for a randomized trial with these 14 dropouts. Something feels not quite right here. Right, it's too perfect.

[Kristin] (46:19 - 47:33)
To get that kind of balance, you would need to use block randomization on gender, which they never mentioned in the paper. Plus, even if they did use block randomization, they would have used it on that original 134 people, and that would have given them gender balance for the 134. But when we have 14 people drop out, how in the world would we end up with exactly the same number of women in each of the four groups?

That's too suspicious. The final thing I noticed is that there's something fishy about the control group. Before they were randomized, everyone filled out a questionnaire about general vaccine hesitancy.

So everybody had a score on this vaccine hesitancy scale where higher scores meant more hesitancy and fear about vaccines. And participants also reported their ages. And here's the issue.

The control group does not look like the other groups. First of all, the control group is significantly younger than the other three groups. Even more striking, the control group had a much lower vaccine hesitancy score, on average, than the other groups.

Plus, there's much less variation in the scores for the control group than there is for the other three groups. That is weird. And the p-value for the difference across groups on this measure is less than one in a thousand.

[Regina] (47:34 - 47:54)
Oh, so to interpret that p-value, that would mean if things had been perfectly randomized, we would see differences at least as extreme as these in only one study out of a thousand, which, of course, is pretty rare. And that makes me a little suspicious about whether this was properly randomized.

[Kristin] (47:54 - 48:50)
Yeah, something is off with the randomization here. And unfortunately, these results just don't tell us anything unless we know that we can trust the randomization. You know, maybe something happened in the study that's not explained in the paper or is explained wrong in the paper, but Regina, it looks suspicious.

I did email the first author twice to try to get clarification, but she never responded. It appears that she's no longer in academia. She was a PhD student when she wrote this paper, but I wasn't able to track her down.

So I don't know what happened here, but this definitely was not a properly randomized study. Before she left academia, that first author also published another paper in 2019 that made similar claims about the familiarity backfire effect. It was an even smaller study, just 60 parents, and this study was not in PLOS One, so I didn't have any access to the data.

So I have no way of checking it. But given the problems with the first paper, I just don't trust that 2019 paper. I would need to see the data to really believe it.

[Regina] (48:51 - 49:14)
Yeah, yeah, I agree. Okay, Kristin, maybe now we can recap where we're at. We have had some evidence for the worldview backfire effect, but it's been really spotty, not terribly robust.

And now we've had a few papers on the familiarity backfire effect, but those have tended to fall apart once you look at them carefully.

[Kristin] (49:15 - 50:16)
That's a good summary, Regina. All right, now I want to get into what's really great about this topic. It's one of those rare cases in science where we've actually had direct replication studies. The topic got so much attention that multiple teams have tried to replicate the findings.

I'm going to start with the one that had the biggest splash. There was a huge replication study published in 2019 in the journal Political Behavior, the same journal that published the original Nyhan and Reifler study. The title is The Elusive Backfire Effect, and it was done by two PhD students.

They ran five large experiments with a total of over 10,000 participants, and they set out specifically to replicate Nyhan and Reifler's 2010 study on the Iraq WMD myth. But they didn't just stop there. They tested that correction, and then they tested factual corrections across 52 different political issues covering claims from both the left and the right.

And they were specifically looking for worldview backfire effects. They used a similar statistical approach as Nyhan and Reifler.

[Regina] (50:16 - 50:27)
Okay, so over 10,000 participants and 53 political issues. That is huge. I hope they found something.

[Kristin] (50:28 - 50:40)
No, they found no backfire effects, none. Across all five experiments, factual corrections consistently reduced belief in false claims regardless of political ideology.

[Regina] (50:41 - 50:50)
So nobody doubling down. That is amazing. But this is so different than the initial studies.

[Kristin] (50:51 - 51:11)
That's right. Nobody doubled down. Even when the correction challenged their own side, people generally updated their beliefs. Now, they did find that sometimes people updated their beliefs less if the statement went against their worldview, but they didn't find any case where there was a backfire. The researchers in this paper called backfire effects stubbornly difficult to induce.

[Regina] (51:12 - 51:17)
Maybe the effects are difficult to induce because they are not real.

[Kristin] (51:18 - 51:32)
Yes, exactly. And here's my favorite part of the story, Regina. Nyhan and Reifler, when they were presented with this evidence, they themselves updated their beliefs. They even went on to collaborate on a new study with these two students who had done this huge replication.

[Regina] (51:32 - 51:38)
So they partnered up with them. Ooh, that is amazing. So what did they go on to find?

[Kristin] (51:38 - 51:54)
They found no worldview backfire effects. And this really had an influence on Nyhan and Reifler. They began shifting their entire narrative.

Even though backfire effects had been central to their work for years, they were willing to update their beliefs and acknowledge that the evidence wasn't there.

[Regina] (51:55 - 52:09)
We have talked about this, Kristin, and it is really rare for researchers to openly admit when previous research was wrong or didn't pan out, especially if it had become part of their research identity. So, wow.

[Kristin] (52:10 - 52:29)
Yes, I admire these researchers so much for being able to update their beliefs. And I love this quote. This is Nyhan in a radio interview.

He says, it would be a terrible irony if evidence contradicting the backfire effect provoked me into doubling down on the backfire effect.

[Regina] (52:31 - 52:39)
I love that. That is a fabulous quote. And he also seems like he has a sense of humor. I love it.

[Kristin] (52:39 - 53:17)
Yeah, it's a great quote. And another person who updated his views is Ulrich Ecker. I mentioned him earlier. He was once a strong proponent of the backfire effect, but over the years, he and his colleagues ran a series of studies.

This was mostly focused on familiarity backfire, and they kept coming up empty. And they started publishing those null results around 2017, around the same time. And even better, in 2023, Ecker ran a direct replication study of that 2017 plus one study on vaccine myths that we talked about.

He used the same experimental methods as they did, but with a larger sample size, 383 participants, and he made a few design improvements.

[Regina] (53:17 - 53:20)
Oh, no. Don't tell me they didn't find anything either.

[Kristin] (53:21 - 54:38)
There was no backfire. The results completely did not replicate. This actually doesn't surprise me all that much, given the problems we talked about earlier with that plus one paper.

And, you know, Regina, at this point, I think Ecker sees the backfire effect as more myth than reality. Like Nyhan, he's completely walked it back. Oh, good for them.

Yeah, you know, some of his colleagues had published a guidebook in 2011 called The Debunking Handbook. And this was supposed to be a resource for science communicators and journalists. In the original version in 2011, they explicitly warned about backfire effects, and they treated them as real risks.

But they updated that handbook in 2020 with Ecker as a co-author. And in that update, they completely backpedaled.

On the familiarity backfire, they write, early evidence was supportive of this idea. I think that's referring to the ghost paper. But more recently, exhaustive experimental attempts to induce a backfire effect through familiarity alone have come up empty.

And then on the worldview backfire, they say, while there was initially some evidence for the worldview backfire effect, I think that's referring to the Nyhan and Reifler paper, recent research indicates that it is not a pervasive and robust empirical phenomenon.

[Regina] (54:39 - 54:45)
Oh, that is well stated, and sounds like a pretty good summary of all the evidence we've seen so far in this episode.

[Kristin] (54:46 - 55:49)
Yes, I agree. And Regina, I didn't have time to look at every paper that's ever been published on the backfire effect for this episode. But the good news is, someone else did that work for me.

There's a 2020 review paper in the Journal of Applied Research in Memory and Cognition that basically summarizes all the studies on the backfire effect that had been done at least until 2020. And there are two major tables, one with all the studies on worldview backfire effects, and one with all the studies on familiarity backfire. And they highlight in those tables any studies that reported positive findings.

And what they find is that for the worldview backfire effect, there were a few scattered studies with positive results like that Nyhan and Reifler study, but overall, not much. And then for familiarity, there are actually only three studies that claim positive findings. Those three studies are the original ghost study, the 2017 PLOS ONE paper, and its 2019 follow-up.

As we've already discussed, I don't think any of those actually support the backfire effect. So coming up empty feels like a pretty fair way to sum up the studies for the familiarity backfire.

[Regina] (55:49 - 55:59)
Mm. This is not looking good for backfire, but it is looking good for fact checkers and people who like to go out there and correct misinformation, people like you and me.

[Kristin] (55:59 - 56:00)
Yes, absolutely.

[Regina] (56:00 - 56:36)
I think now, Kristin, we are ready to wrap this up and rate the strength of evidence for this. What do you say? Yeah, let's do it.

Yeah. Okay. The claim we are looking at today is that the backfire effect is real, that correcting misinformation might reinforce the misinformation instead of dispelling it.

And the way that we rate the evidence for claims is one to five smooch scale. One means little to no evidence for the claim. Five means very strong evidence for the claim.

So, Kristin, you go first. Kiss it or diss it. What do you say?

[Kristin] (56:36 - 56:52)
I'm going to go with 1.5 smooches, Regina. I'm pretty convinced that the backfire effect is not a real phenomenon, but maybe there are certain situations where people are really stuck on something where they might double down. So I won't rule it out entirely.

What about you?

[Regina] (56:52 - 57:16)
I'm going to go with one smooch on this because everything that I've just heard from you, it feels like it was a case of overblown attention or a study that just randomly happened to get a fluke finding for something that's not really there. We talked about this before in the red dress effect episode. So I'm going to go with one smooch.

What about methodological morals? Do you have a good one for this, Kristin?

[Kristin] (57:17 - 57:24)
I'm going to pick on the ghost paper. So mine is never cite a study you haven't read, especially if no one seems to have a copy.

[Regina] (57:26 - 57:44)
That is like a great rule of life, I think. Okay, mine is don't just double down on your research results. Stay open to updating your beliefs because I love that is what the researchers here did.

[Kristin] (57:44 - 57:49)
Yeah, we have another great science story here where researchers were actually open to new evidence.

[Regina] (57:49 - 58:05)
New evidence, updating their beliefs because this is the way science is supposed to work. And it's actually very meta, isn't it?

[Kristin]
Yes.

[Regina]
It's meta because there is no backfire effects. So do you think that we dispelled the myths or did we just reinforce them?

[Kristin] (58:06 - 58:09)
I think we did some good myth dispelling here, Regina.

[Regina] (58:09 - 58:14)
All right, Kristin, this has been fabulous. Thank you so much. And thanks to everyone for listening.

[Kristin] (58:14 - 58:16)
Thank you, Regina. And thanks, everyone.