668. Do Taylor Swift and Bad Bunny Have Blood on Their Hands?

Hey there, it's Steven Dubner with a quick announcement.

“We've just published an audiobook called Making Messiah.”

How Handel got his mojo back and created a masterpiece. The publisher is Simon and Shuster, and the audiobook is adapted from our recent Freak Nomics Radio series about the history, legacy, and economics of George Fitteric Handel's Messiah. You can buy making Messiah now wherever you get your audiobooks.

Or more information go to makingmysiah.com. [MUSIC] Over the years, we've made a bunch of Freak Nomics Radio episodes about the risk that comes with an activity that billions of us do every day. Get in a car.

Those risks have fallen over time, but because more people drive, more miles and ever, the number of deaths from traffic crashes is very high, more than a million people per year globally. And that's just deaths. There are many more injuries and the financial costs are massive.

Traffic deaths are more likely in low income countries, but among high income countries, the US is an outlier, with more than 40,000 deaths a year. That works out to roughly one death by car crash every 13 minutes in the US. Why so many? Well, it's complicated.

Here's what we heard in an episode.

“We published a few years ago called Why is the US so good at killing pedestrians?”

The cars we're driving are bigger, at or faster. The problem of distraction has gotten much worse. In the United States, we've decided that car movement is really the supreme consideration when it comes to designing our streets. And in another episode, we made, called Why is flying safer than driving?

You learned how the aviation industry devoted itself to safety. And you go back 30 or 40 years, air crashes were not uncommon. It was something the industry spent in enormous amount of time, collaborating together, sharing information, sharing learnings, working closely with the FAA to understand best practices and how we could have an open book with our regulator.

And in our last couple episodes, our friends at the Search Engine podcast looked at the contested future of driverless cars. Personally, I believe for a long time that driverless cars will save a lot of lives.

“But until that's the norm, we, the drivers, are still behind the wheel.”

And why is that a problem? We've engineered a world where the most distracting device ever made is also the one that we use to listen to music in the car. Today, on Freakonomics Radio, we talk about a new research paper that looks at traffic deaths on a very specific set of days.

Hubham release days are basically a natural experiment.

So what they find, that's coming up for now, eyes on the road, please. This is Freakonomics Radio, the podcast that explores the hidden side of everything. With your host, Steven Dubner. The study we are talking about today was just published as a working paper by the National Bureau of Economic Research.

There are four authors, all of whom are medical doctors and one who is also a PhD economist. So let's start with him. My name is Bobujena, I'm an economist, position, and professor at Harvard Medical School. Jenna used to make a podcast with us called Freakonomics MD. Back when he got his Econ PhD at the University of Chicago, he studied under my Freakonomics

friend and co-author Steve Levitt. So Jenna is well trained in using data to find what's hiding in plain sight. In 2023, he published a book along with his co-author Christopher Worsham called Random Acts of Medicine. The random in the title isn't just random.

It refers to the kind of experiments that people like Papugena and Steve Levitt often used to make these connections. Sometimes they're called natural experiments. Natural experiments have taken a while to take hold in medicine. Their origins are actually in epidemiology and then economists several decades ago started

popularizing their use and they're almost certainly most widely used in economics compared to any other discipline. In one early paper, Jenna used a natural experiment to see what happens to high-risk heart patients when their cardiologists are out of town. And what was the random act of medicine that allowed him to measure this?

So Jenna and his co-authors looked at what happened back home during those conferences and

“they found that mortality actually fell.”

Why? The best explanation is that these conferences are primarily attended by the most senior cardiologists and that senior cardiologists are more likely to opt for more invasive and riskier treatments. Jenna also used a natural experiment to study the behavior of drivers.

Five years ago, we had this piece in the New York Times the upshot where we showed that after fast and furious movie releases, there is an increase in speeding behavior. There is this question of whether art, mimics, life, or vice versa. You need some sort of natural experiment. You do not see an increase in speeding behavior when the hunger games movies come out

or Harry Potter movies come out. The movies are called fast and furious, not slow and deliberate. We had the GPS coordinates of where the person was pulled over for speeding. And what you see is the people are being pulled over on highways that are adjacent to where the movie theaters are.

A conclusion like this requires some serious faith in the robustness of the data and of the analysis itself. The good news here is that there has come to be a significant body of research like this and the track record is pretty strong. Donald Redelmeyer, a physician and epidemiologist at the University of Toronto has produced

a number of these studies. One of them done in collaboration with Danny Koneman used data from colonoscopies to illustrate something called the peak end rule. They found that when people look back on an experience like a colonoscopy or even a vacation, they tend to judge it largely by the peak moment of the experience and the end of the experience.

So if a bad family vacation ends well, it may generate a better overall memory than it might deserve. In another study, Redelmeyer used three decades of driving data to show that April 15th tax day led to a roughly 6% increase in traffic deaths. The best explanation is that the psychological burden of a stressful deadline spills over

into the car. The paper we are talking about today, by Papugena, Vishal Patel, Chris Wersham, and Michael Lou also looks at a hidden risk of driving. It is called smartphones, online music streaming and traffic fatalities. So where does an idea like this come from?

We devote an inordinate amount of time.

“I think compared to most other researchers to simply brainstorming research projects.”

That is Christopher Wersham, Jenna's longtime research partner. Is there so many data sources out there that anybody can do basically anything with data? Wersham's job description also has some hyphens in it.

I'm a pulmonary and critical care physician at Mass General Hospital and I'm a health policy

in public health researcher at Harvard Medical School. I envision you having a day outfit and a night outfit, or at least a hat that you switch. Your day job is keeping people alive in the ICU. What's what I was doing yesterday, today I'm here talking to you about my research. The broader term is a physician scientist.

Most people imagine a physician scientist as someone doing something in a lab with pipettes, test tubes and petri dishes that they're then going to bring to the patients that they work with and a number of my colleagues are that type of physician scientist. And then there are folks like me who the science is more public health and data science and studying our broader health care system.

“Do your bench science colleagues think that you're a little bit kooky?”

Probably. There are a number who just don't get it at all and some who don't care to get it. Then there are some who don't get it completely but get enough of it to think it's cool.

I am lucky enough to be a pulmonary and critical care doc who also gets to do this type

of research because I am supported by people who even if they don't fully get it, understand that these types of analyses can be incredibly powerful. If you were forced to do just one of your two functions, which would you choose? Oh God, probably be the regular doctor. I spend about 20-25% of my time being a doctor who sees patients in the rest of my time doing

research and writing and trying to figure out what are the best things we can do to improve our health as a nation. When you are practicing medicine, you are faced to face with the outcomes of our deeply

Every time I'm seeing patients, it fuels a new set of ideas.

Can you give an example? I'll give you an example of something we are working on right now as an ICU doc working at Mass General, which is a big referral center for much of New England. We have patients who are extremely sick that sometimes hospitals with less resources are just, we have to get him out of here and get him to you and so we'll fly this patient

on a helicopter. I've had patients medflated who come in just unbelievably sick and I am glad that we did not waste a single minute, getting them to a hospital like Mass General where we have every resource under the sun. Then I have patients who have come in on a helicopter who are smiling and waving at me.

“When something like that happens, you just wonder, is this the best use of our resources?”

The way I have been able to avoid burnout professionally is that I have some tools to try to address some of these problems. Several years ago, Chris Wersham and Papu Jenna set up a working group to help find good research ideas. They found that a good idea can come from anywhere.

We actually had a ninth grader who has shown up to some of our ideas meetings all the way up to tenured professors and CEOs of companies who kind of drop in. Where are you guys meeting online? We were doing this before the pandemic and is this a regular set meeting? Three times a week.

Three times a week. Three times a week. What would you say is the median number of people on this three times a week meeting? Probably four or five. Most of these ideas that come in are pretty half-baked, even quarter-baked.

I was just in the hospital and this thing happened to my patient and I think there might

“be something here, let's talk it out to see, is there an idea here?”

Okay. Well, that's a good idea. Is there data available that we could use to try to answer the question? And if so, is the data good enough to actually do it? The even ideas that look good may turn out to be impotent once you dig into the data.

Here again is Bapugena. I thought that after HBO fight nights, you might observe increases in domestic violence. We don't observe that. We thought that when it snows a lot, when it's really cold outside, you see an increase ADHD diagnosis because kids are unable to go outside for recess and they're stuck in schools.

We don't see any evidence of that. You know how your phone rings when this is an envelope alert? It's crazy, right? Now, imagine being in the car when that thing goes off. You might think it would transiently distract you, a because of what it represents but also

because that noise is really loud. We looked at this. We don't find any evidence that traffic authorities increase when this number alert. We even thought to look at surgeons who are operating, imagine you're in the zone operating and that thing goes off, surgical mortality doesn't change.

I'd like to know what to make of this line of questioning about distraction. Probably some matters, some don't. We have a lot of these really interesting ideas that just don't pan out. But once in a while they do, and this was an example of one of them that did, this idea being the new paper about streaming music.

So Vashal came to us with this idea. He's talking about Vashal Patel, who is 28 years old. I am a surgery resident at Breagan Women's Hospital and a researcher at Harvard Medical School. Vashal is also a physician hybrid.

I did a master's in public health when I was in medical school. A lot of those courses expose you to some of these economics methods when I came to Boston. I started my residency. I saw that Bopu and Chris were working at the intersection of ethics and real-world

data and that drew me to working with them. Okay. Let's get into this new paper.

First, just describe where the idea came from.

I'm typically the type of person who drives with the radio off. I kind of like just get lost in my thoughts, maybe think about work. But my wife, she's the complete opposite. She's a singer, so music's her whole world. A couple of years ago, the New Taylor Swift album came out and my wife texted me saying,

“"Oh, you have to listen to these songs."”

You're driving, and your wife texts you while you're driving, telling you that you need to listen to the new Taylor Swift record. I don't think she knew I was driving, but I was like, "Okay, fine. I'll try it." So I started fumbling through Spotify to try to find the specific songs you've

been talking about. I looked up and I realized I was starting to drift out of my lane. A few more seconds, and I probably would have been in a serious accident. Did you blame anyone for this near-miss? Did you blame yourself, your wife, Taylor Swift? Well, I blame myself.

I shouldn't have done that.

I was like, "Oh, I'm never going to do this again."

And then a day later, I did the same thing, probably. Everyone uses their smartphone while driving.

We've engineered a world where the most distracting device ever made is also the one that we use to listen to music in the car.

When the first things Patel did was to dig into a federal database called "Fars", that's

the fatality analysis reporting system. It's been used for decades to study traffic fatalities. There's actually like a whole genre of research just built around the data set. For example, do car crashes go up on Super Bowl Sundays, or on U.S. tax income days, or even on full moons.

Well, I had that near-miss on the road. It kind of clicked, and I was like, "Well, album release days are basically natural experiments. There are moments where millions of people suddenly now have a reason to pick up their phone and interact with it sometimes while driving."

And nobody had really looked at whether or not that spike in potential smartphone use translates into more fatal car crashes. Your suspicion was that it would. I thought so, given my personal experience, but we test a lot of ideas, and a very small percentage of them actually work out.

So I wasn't very hopeful that I would find a signal, but it was something where I had to try. When I thought of this idea, I actually didn't share it with Bopou initially. I went home and I did the analysis, because I was so curious. He sent me this picture, which showed an increase in traffic fatalities when these major

music albums were released on Spotify. I'm thinking to myself, "All right, what does he have in mind?" I immediately thought that the mechanism was big music album release people are filling

“where their phones, that's what he's thinking, but I just pick up the phone and I call”

him and say, "What did you do? How did you get to this finding and do you think it's real?" And what did he say? What did he do and how did he get that finding? So what he did is he uses really terrific data as publicly available.

It includes all traffic fatalities in the United States. He's been working with that data in a number of different projects. And all he did was he looked at the dates of major music album releases on Spotify. You can actually look at Spotify data to see how streaming volume changes when these albums are released.

He'd seen that and show that there was a large increase in streaming volume on the days these music albums come out. Then he paired that with analyses of traffic fatalities and showed that on those same days, there seems to be this spike in traffic fatalities. That's sort of the starting point.

Once you see that, you've got to do a lot more to make sure that that's a robust finding and get to the mechanism. The spike in the data was significant. Patel looked at all fatal crashes in the US between 2017 and 2022 and then he laid on

top of that the release dates of the 10 records during that time frame with the most first

day streams on Spotify. We're talking about artists like Taylor Swift, Bad Bunny, Drake, Kendrick Lamar, streaming volume on those record release dates increased by 40%. There was a 15% increase in traffic deaths, which translates into roughly 18 extra deaths on each of those days.

Correct. On a base of about 130 deaths per day nationally, when you see a finding like this, where

“the magnitude is large, you've got to ask yourself, is it implausibly large?”

To me, this feels like it could be in that area where it's larger than I would expect, then how do we know what to expect when you see the big spike and you think you understand what may be causing that, the release of these new records. I think of a spectrum of excitement on the one hand and devout skepticism on the other hand.

Where do you lie in that spectrum and then how do you go about trying to make sure that your excitement is justified? I'm like 99% excited, but 0% skepticism. In my world, when I see a paper from a journal that sends me to referee a paper, if the paper is really interesting to me, I'm usually not very skeptical, whereas a lot of people

are very skeptical. I just love the idea. I love the idea too, but I also just want to say, we've lived through a few years here where a lot of research findings, especially from social psychology, have come under question, and for a very good reason, it turned out to be their exaggerated or fraudulent

and so on. I'm just curious if it's changed the dynamic for you when you're investigating and checking the robustness of an idea, are you a little bit more skeptical or at least careful than you used to be? I would say, I have the same level of care.

I appreciate why there's a lot of concern about findings that don't replicate.

“But I think that we miss so much of in science is we're very outcome oriented.”

At the start of our conversation, I gave you two ideas that I thought were really clever for which we didn't find anything.

The world will never hear about them.

This could have been an example of that as well, where we looked at music album releases and find no increase in traffic authorities. To me, that was independently an interesting question. The fact that we find a positive outcome that seems to be robust to various things is important

Of course, you've got to do the things to try to convince the world that the finding is correct, but I'm most interested in the idea personally. Coming up after the break, Jenna and his co-authors try to convince us at least that their finding is correct.

“Also if you want to hear more about some of the academic fraud I just mentioned, you”

could start with our episode number 572, which is called "Why is there so much fraud in academia?" I'm Stephen Dubner, this is "Freakonomics Radio" and we will be right back.

When I first saw this new research paper that we're talking about today, we found that

traffic deaths spike on the day of big record releases. I was curious as I expect you would be and also a bit skeptical as I also expect you would be. So I sent the paper to an elder statesman in the field, who I mentioned earlier, Donald Redelmeyer. He wrote right back.

"I read the materials and liked what I saw," he said, "best single aspect was focused on traffic fatalities. I was also impressed by nifty nuance of in-cart infotainment, objective data on volume of streaming online audio and several diligent falsification tests. If easy, he wrote, would be great to update data from 2023, which is now available from Nitsa.

That's the National Highway Traffic Safety Administration.

And here's how Redelmeyer ended his note.

“Overall, I believe the concept is exactly correct.”

Namely, a moment of inattention can cause a fatal traffic crash. Furthermore, exciting new music can be enchanting to some listeners and indirectly causes some to potentially lower their guard. I doubt this is what the artist's intended. I also doubt this is what the artist's intended.

And so with Don Redelmeyer's seal of approval, I pressed on. Here again is Vishal Patel talking about how using public federal data is a sort of insurance policy while doing this kind of research. One of the benefits of using publicly available data is, well, there's a benefit for both the researcher and for the public.

For the researcher, the benefit is that if there is any doubt that anyone has as to the validity of the results, the data is publicly available and it can be replicated by

“anyone, that's a common thread through a lot of my research.”

I actually have only had access to publicly available data. In the back of my head, anytime I'm doing something, I'm like, oh, okay, this is great. If there is a mistake, someone will catch it. The research questions that I'm interested in are research questions that a lot of people are interested in.

One of Patel's concerns with this research project is that the album release dates are not random. For years, it has been common practice to release albums at midnight eastern time on Fridays. This used to be driven by anti-piracy concerns, but now it's a way to globally synchronize day one streams. Vishal Patel understood that he would have to factor that in.

Because most of these albums dropped on Fridays and Fridays already have higher traffic fatalities and say Tuesdays, we had to design the study to compare each album release day to the same weekday in the week before and after. The window of the study period included 10 days before the album release drop and 10 days after, which would ensure that regardless of which weekday it was released on, we would include

the Friday before and after or the weekend before and after.

The main analysis by definition accounts for the Friday effect, because you can always compare

to the Friday before and after. We then also performed some falsification tests. The first falsification test that we performed is we randomly selected 10 placebo dates and then again 10 separate random Friday placebo album release dates, re-repeated the analysis a thousand times.

So a real effect was larger than what you'd get by chance alone in 98 to 99 percent of iterations of this placebo test. Then what we did, we took each album's calendar date and applied it to other studies. For example, if a album came out on let's say May 5th was a Friday of 2020. We went to the same day in 2019, 2018, 2017 so on and then the years subsequently,

where there was no album release and we did the same analysis. We picked the closest day to the same day of the week to that calendar date and the album release saw no spike in traffic fatalities. Okay then you started looking at some of the particular components of this spike in deaths. Walk me through that. One of the questions is, could it just be that people are drinking more

those with alcohol involvement and separately it wasn't more pronounced at night time.

So if this was primarily a partying effect, nightlife effect or alcohol effect, you'd expect the opposite pattern. A lot of crashes happened at night, plainly. That must have made your spike look even more interesting and anomalous. Exactly. The fact that it held for both day and night time

“crashes was reassuring. What caused you the biggest concern that your conclusion was correct?”

The main signal is really what we felt had to be backed by robustness and falsification tests, because simply we just didn't have a control group. All we had is on the days of these album releases. Traffic fatalities go up. It could be coincidence looking at who is most effective. You would expect that younger Americans who are much more likely to stream audio in the car are more likely to be affected by this album release effect. We stratified the effects by age and we

found exactly that that the effect size increased with younger age. The data that you use goes up through 2022. I wondered why you didn't include whatever is available beyond that. This is data from far as the fatality analysis reporting system. Tell me why 2022 was your cut off.

“Far as data lags behind a couple of years. It's not like the mortality data that we have from”

the CDC. There's a lot more granular data in it and so it usually takes several years for all that to be reflected. Currently the data goes up to 2023. When we had done the study that was the most recently available data that was out there. Our producer, Theo Jacobs, out of curiosity used Claude code to reproduce your results from 2017 to 2022. He did and they replicated and then he re-ran the analysis using 2023 far as data and says that the effect size is went away with the 2023

album releases. The result was no longer significant. First of all, his analysis is nowhere near as

detailed as your teams. But I am curious to know what you think about that back at the envelope finding. That is interesting and I have two comments for that. One, it's possible. There's only 10 album releases that we use and I'm not sure how many more albums were actually released to that met that top 10 of all time criteria in the year 2023. But even if there were one or two new albums that were in that group, it's possible now 20 to 30% of the dates have changed and now are

probably more contemporary dates. It's possible that autonomous vehicles and driving safety might

“be changing the signal that we saw before. The other thing is I think that that is actually a very”

interesting use of Claude code and AI in general. I think it is so useful to use these technologies as they are not only to brainstorm new ideas but to go back and check old ideas and make sure that they or what we intended them to be because in the same way that humans can make errors while driving, humans can certainly make errors while analyzing data sets. We sent those Claude code results to

Patel and the rest of his team with some time to dig into our analysis. Here's what Papugena had to say.

When you replicate it with Claude, I'm happy to say that our original findings do replicate, which is good. When you include an additional year of data, both on traffic deaths and on album releases, as you said, the effect goes away. The reason the effect goes away is because when you include 2023 albums, it turns out that's like five out of the ten most downloaded albums are coming in from 2023. Three of those albums fall on consecutive Fridays. What that does is it makes the treatment

and control basically the same. If there's a Friday album release and in the previous and the following Friday, there's another album release. You're now comparing a big streaming day to other big streaming days. The way to deal with that is to say, all right, let's look at albums from let's say 2017 to 2023, which is what you did and look at the top ten albums that are not overlapping in their dates. When you do that, you get the same sort of picture as we originally found. Let's talk a little bit

about the circumstances of these fatal crashes in your data set and how they compared to the median fatal crash. What can you tell us about highway driving versus not how typical or a typical with these crashes? We didn't look at the location of the crash. That's a good thought.

We mostly looked at circumstances, did the crash involve alcohol, did it occur at night time,

where the visibility conditions any better or worse? Tell me what other potential confounding factors there may have been. Maybe these ten album releases that you measured happen to coincide with bad weather with something going on in the world, etc. Correct. Maybe these are happening around holiday weekends. Let's say it's Memorial Day or Labor Day. We account for that as well. The other thing might be maybe it's possible that weather or other things are just

happening randomly on those ten days. What you could have done and we didn't do this, we could have looked at actual weather patterns on the day of the album release compared to every day around it and shown that the weather is stable. I didn't think to do that because I thought it just on its face. It would be highly, highly unlikely that that would be the case because weather happens to be random already. Let's say you index record streams on day one of these big records. In your data set,

you pick the ten biggest out of these several years, Taylor Swift, Bad Bunny, a few others. Let's assume that those are ten on your index. Did you test this effect against releases

“that ranked maybe an eight on that index or an eight and a half or an nine?”

What we don't do is look at it album by album like the most downloaders or least download and the reason why is because it's just not enough power to do that. What we can do is say, let's look at 10, 20, 30, 40, 50 and as you move further down in terms of albums, the effect falls, which to me makes sense because I think most of the effect is happening with these very large music album releases. For small album releases, you're not going to really see an increase in streaming volume.

Another way to get it, this is that suppose that we found this effect in traffic deaths, but we didn't see anything going on with streaming. I'd be a little bit worried about what might be going on. That's a harder story to explain than one where we can clearly see that there is a large increase in actual streaming volume around the day of the album release. Okay, what did you learn about the

“characteristics of the drivers? That's a good one. So the two things that I think we're”

interesting to us. One is that the effect is a little bit larger in younger drivers than older drivers, which fits your thesis, yes, which fits the thesis. And the other thing that to me was interesting when Vishaw was talking to me and you know, you get excited when you have these insights. And I said, oh, Vishaw, we should look and see whether or not the effect is any different when there's one driver versus a driver plus a passenger under the theory that if there's a passion, the car,

maybe they are operating the device or maybe they're just talking and not listening to the music. And we actually find that the effect is larger. What I'll call the album release effect is larger when there is a solo person in the car, just the driver. It's smaller when there's a passenger in the car. You're saying that of those who do have a passenger in the car, the rate of fatality is a

little bit lower there, which suggests to you that that second person is kind of an insurance policy,

because the driver isn't having to fiddle with the phone, yeah, correct. I go one step further. What we're doing is we're looking at whether on album days, the presence of that person, is protective relative to all the other days surrounding it. And there what we find is that it might be protective. So having a passenger in the car makes you less susceptible to an increased car crash when an album is least compared to the surrounding days. Coming up after the break,

the federal safety agencies do produce a lot of good data, but when it comes to what drivers are actually doing before they get an crash, that is much trickier. We will get into that right after this. I'm Steven Dubner and this is Freakonomics Radio.

Welcome back to the third and final act in our exploration of a new

working paper, just released by the National Bureau of Economic Research, it is called Smartphones, Online Music Streaming, and Traffic fatalities. One of the authors is Papugena. He is a

“long-time friend of Freakonomics, and more important. He is both a physician and a PhD economist.”

Here is more of my conversation with him. So in the case of the record release dates and the spike in traffic fatalities, I guess there is a couple ways you could look at it. 15% up one day and anything is quite a lot, but overall 18 extra fatalities and these bigger record release dates only happen one to maybe three times a year. So it's not a massive loss of life. On the other hand, you've identified what seems to be a pretty direct cause. How do you think about the meaning and importance

big enough to do something about because it only shows up when you look at the most downloaded albums. Most music that people listen to or download outside of let's say the top 10 music albums, you're not going to see an album day release effect. What that means is that when there's a lot of attention on this, you might see a mortality effect. As a source of explaining why we have traffic

fatalities in this country and whether or not this can move the needle, the answer is no,

“it can't because they're very small relative to the total number. But why I think this is”

interesting and these kind of things are interesting is because they give us some scientific insight into what are the types of factors that cause people to have problems on the road. It identifies it in a quasi-experiment to way because it really is hard to know what are the magnitudes we're talking about, like how big of a problem could distraction be. You can't randomly distract people on the road and study the effects of that on their behavior. When you see these data and you interrogate it the

way you do and you believe the effect as you do, what do you make of the mechanisms by which these crashes are happening? And maybe within that answer you could talk about the fact that we don't really know much about driver behavior in cars at this relatively advanced stage in our civilization, which you might think we would. You could certainly gather that data if it were encouraged and allowed, but it's not really. So people like you are left to guess a little bit. So what do you guess about

as to the mechanism? Yeah, so let me tell you what we guess about. We're thinking, art, is this person literally manipulating the phone or the infotainment system, like Apple CarPlay? We find that in cars with CarPlay the album release effects seems to be greater, which might be counterintuitive because you would think that CarPlay is something that should reduce issues in the car. Because it's supposed to be hands-free, yes? It's supposed to be hands-free, but it does lower the

barrier to accessing everything. You can just touch the screen very easily and go from one song to the other. So you're probably doing it more than you otherwise would have. You adapt in that way.

So one mechanism is that the drivers are just fiddling around with something. The second mechanism

might be that they're just playing it louder than they normally might play. They're not aware of other environmental signals that they might normally be aware of. Someone who is honking at them or any sound that might clue their attention into something being wrong on the road. A lot of people listen to this show and other shows while they're driving. I wonder if you have any advice for us? We already exclude the sound of just about anything that sounds like you

would hear it on the road. Like sirens, we don't use whistles, car horns, we try to eliminate all that because we know it can be distracting. But maybe, I don't know, I should speak more slowly or more monotonically or something to have any advice for how I could not join the ranks of Bad Bunny and Taylor Swift as killers. Well, you have a very nice radio voice. I wouldn't change anything there. But I mean, Steven, you're literally giving me an idea. Now, I'm thinking

myself, can I figure out which popular podcasts have sirens in the background and see what happens

“on those? Like, if Joe Rogan ever had an ambulance on his show, how many people have died?”

When you look at this constellation in its entirety, I'm curious what it makes you think. What it makes me think is bring on the friggin autonomous cars already because I think humans are just pretty bad at driving on average. What do you think of this? I think that's right. I'm going to

say two things. The Shaw's going to kill me when I say the second thing. The first thing is that

if we were to replicate this study, let's say using 2024 or 2025 did and go forward, my guess is that the album release effect will be less and less over time. And the reason why it's that the cars are changing. The safety features are changing. The solution here isn't to not have Taylor Swift albums. That might be the solution for some. But it is to make cars safer. And I think that you would find that the effect that any given distraction over time will wane as the vehicles

become better. The second thing, and this is what Bashar will tell me about is a few months ago, I had this idea related to Tesla's. Sometime in 2024 in a spring, Tesla made the autonomy feature free for Tesla owners. I presume what happened is that that time a lot more self-driving was happening. We will use this far as data, this fatal accident reporting data, which has the make a model of the vehicles of all in the crash to see what happened to traffic petalies around

“that time with Tesla's versus other electric vehicles. Why will Bashar hate that does he drive a Tesla?”

No, Bashar will hate it because part of why we are successful is that when we have ideas, we don't broadcast into the world. We do them first. But I have enough papers. I'm not worried.

whole traffic ecosystem that theoretically should have made us much safer over the years. And of course, it has. But in the last 10 years or so, US traffic crashes, contra other developed countries

“have risen again. So what do you think is going on in this country in particular?”

It's hard to know, but there is a economic explanation. Or at least there's probably many economic

explanations. The first one that I think of when I hear a pattern like that, where safety and

technology is improving. But the outcomes are getting worse. This compensating behavior of model that was initially popularized by the Chicago economist, St. Peltzmann. What happens with car play? You now have the ability to use your car in a way that you perceive to be safer and you compensate. Your ultimate safety might be actually worse because you're using your phone in a way that you thought was safe, but it turns out it's not as safe as it was. We are in this moment now,

where people are seeing waymo's driverless taxis in some cities. People are having more and more autonomous features on their cars. How do you think about this transitional period we're in and what else should we maybe watch out for as our tech becomes better, but we're still the driver. I was actually just in San Francisco watching these waymos go around and they're all over the place. The person us with made this observation, because we were talking about whether waymos are safer,

and they made the observation they said, well, I bet waymos themselves aren't just safer, but they reduce accents of those around them because they go so damn slow. You can imagine a world in which driver behavior is affected by other drivers, behaviors. If autonomous vehicles are on the road and they're driving slower, then people behind them will probably necessarily drive

“slower. They'd be the spillover effect, which I think is interesting and I'm curious to sort of anticipate.”

If you had the authority to mandate one policy change based on this research finding, maybe it's a policy change for Spotify or Apple, maybe it's for the automakers, maybe it's for

Nitsa National Highway Traffic Safety Administration. What would it be? First, I was going to say to, you know,

turn the music down. Like, what a new album is released to mandate on the phone that the volume can't exceed a certain level. Honestly, I think the number of deaths here is so small, relative to the average daily number of deaths, and it doesn't add up to a lot nationally over the course of the year, so I don't know that I would change anything. There is one policy solution that might seriously improve traffic death research. I talked about this with Chris Wersham.

Let me just admit that when I read a paper like yours, there's a lot of following a

“daisy chain here. Like, what about the telematics revolution that we've been promised for a long time”

that automobiles would be able to tell you a lot about driver behavior, like exactly what's happening in that vehicle during the whole trip and then at the moment of impact. I would much rather base an argument on data that's first order like that rather than do this really nice daisy chain analysis you've done. Do you think about better classes of data for this kind of research? Absolutely. If only we could get our hands on those data, the study design has allowed us

to create this connection between the release of these very popular music albums and traffic fatalities. Our study design inherently does not tell us exactly how that happens. You're left with what could plausibly be the link between these two, this logical explanation that it's through smartphone use. But if the folks at any of these automobiles manufacturers who have these data, if you're using Android Auto or Apple CarPlay, I don't believe that Apple or Google

holds on to that data long term, but presumably as part of a traffic investigation, one could collect the telematics data. On the flip side, I certainly understand why there would be reluctance from car owners, reluctant stemming from privacy concerns. Exactly, right? If you're a car owner and you get an offender bender and your insurance company doesn't want to pay for it and they can interrogate your car and say, "See, you were blasting Taylor Swift at the time of this accident,

I certainly understand why folks would not be interested in that." I've heard this argument for a long time, including in medical records, right? That was the whole idea with why electronic

health records will never work is because people don't want their personal health records exposed

and low and behold, they mostly do work and mostly people have gotten over it. Don't you as a physician and as a taxpayer and as someone who drives on these same roads with other people who might not be paying as much attention as you, don't you want telematics to be the norm? Doesn't

multiple versions of privacy concerns in the example you give of health records. The inherent idea is we digitize these health records so that Dr. A at this doctor's office can easily communicate with Dr. B or that when a patient ends up in the hospital with me, I can easily pull up all of their records without having to go to a file room and waiting for someone to bring it up. The data sharing is there for the benefit of the patient and then there are these concerns about what happens to a

data breach and someone could use their data maliciously and you don't have to look far for

“examples of people abusing that data. But I think when it comes to telematics, there are potential”

negative consequences to you directly from that data being kept, who's going to bother looking at your telematics until you get in a car accident? It's either going to be your insurance company or the police. What's the value to you as an individual? It's less obvious than in the case of health data. Whether it is a police insurance, I'm just in favor of good data. I'm just in favor of transparency. So when I hear these arguments about, oh, we have to protect the privacy of behavior

in, let's say, your automobile. I totally get that. I might not want video recording by a third

party in my car, but if I'm involved in a crash, whether I'm the one at fault or not, whether I'm the one that's hurt or not, I would certainly think it's a baseline expectation that you'd

“have transparent data about what may have contributed to that. We just don't have it. And I find”

that to be a bizarrely massive failing. So when we talk about the magnitude of traffic fatalities around the world, I love what you've done with this paper, but I feel like you are using sort of no offense caveman tools in a world where things are running much bigger, faster, and deeper, I would like better tools and better data. That's all. I absolutely hear what you're saying. This in many ways is part of a larger story about public health research. Why we do studies

like this is we are very often making lemonade out of lemons, because there's only lemons as far as the I can see. It's hard to find telematics research anywhere. When we were doing the background research for this study, there's occasional academic papers published by universities that work closely with auto manufacturing, those groups who published some of these studies about distracted driving, and they are usually on test tracks because of the obvious ethical concerns there.

We're always saying we have real world data. The real world data is out there. And so much of it

is beyond our grasp because it is proprietary. We asked multiple times for data from the federal government and will get fully over a quest denied on absurd grounds. There's plenty of data out there that if it was available to researchers like us, we could do tremendous public health research and we can't. We try, but we can't. You're saying the federal government doesn't give you what you

“want. What about private firms? I remember years ago Steve Levitt and I we spent a bunch of time”

with an insurance company and they as part of this research that we were doing, they gave each of us a little fob that plugged into a port in my car that I did not know existed. You plug in this little thumb drive essentially and it produced telematic data on acceleration, breaking, idling, et cetera. And I loved it and I wanted to know like am I a decent driver, a bad driver, am I changing lanes too quickly, am I breaking too frequently? I thought that was the beginning of a movement. I assumed

that we were just entering this wave where the data would be much much much better in where traffic crashes would fall and fall and fall. And instead, as you point out in your paper, we're one of the few rich countries where traffic crashes have been rising lately. So it's a really big problem traffic fatalities and I guess I'm just sharing with you my frustration. I share it completely. If researchers like us could work with a lot of data that is hiding behind obvious

proprietary barriers, we could do a lot with it. I reach out to companies all the time saying, hey, I've got a cool research idea. Your data would be perfect for it. There have been a number of companies who are actually willing to have that conversation. This is what we're willing to share with you because some companies actually do want to use their data for things like this and try

to be helpful with it. But there's so anxious about their business concerns that they never share

So, what is the real message of this paper that we've been talking about today? Here is Vichal Patel,

“the man who had the idea. The real message here is not to change album release days or to prevent”

artists from releasing their albums. The bigger message is it's really all about distraction. And so understanding how new technologies interact with human behavior on the road, it's probably going to be one

of the most important traffic safety questions for decades to come. There's an evolving landscape

of smart phone and smart car technologies. For example, autonomous semi-autonomous driving features they're being rolled out at scale. There's some safety implications that may be protective from these smart technologies. There's also a lot of potential to create distraction and a false sense of security. My thanks to Vichal Patel, Papugena, Chris Worsham, and the fourth

“co-author Michael Lu. Thanks also to Don Redelmeyer. If you want to let us know what you are thinking”

about this episode, our email is [email protected]. And if you have any ideas for

Papugena's Mary Band of Health Policy Researchers, he is also happy to hear from you. His email is Jenna. That's J.E. N.A. at hcp.med.harvard.edu. Coming up next time on the show, 95% of urban made in the world is made within a 45-minute drive at the house I live in. And why is urban production so concentrated? You could say it has benefited from some soft protectionism. Well, I can defend every single one of those components as to why they are positive for flavor.

“And is that whole new chart oak barrel idea really that important or is it more of a marketing message?”

You do have to admit, it's a bit convenient. The urban trade uncorrect. That's next time on freakonomics radio. Until then, take care of yourself. And if you can, someone else too. Freakonomics radio is produced by Renbud Radio. You can find our entire archive on any podcast app. It's also at freakonomics.com where we publish transcripts and show notes. This episode was produced by Teo Jacobs and edited by Ellen Frankman. It was mixed by Jasmine Klinger with help

from Jeremy Johnson. The freakonomics radio network staff also includes Augusta Chapman, Dalvin Abwajji, Eleanor Osborn, Elsa Hernandez, Gabriel Roth, Alaria Montenacourt, Zach Lipinski, our theme song is Mr. Fortune, with hitchhikers and our composer is Luis

Guerra. As always, thank you for listening.

I had this paper a year ago which showed that when marathons happen in a city, more people die because ambulances can't get the people's homes. Whenever I threaten to write a paper about what happens when Teo a swift comes to town, everybody tells me, I can't write that paper. The freakonomics radio network, the hidden side of everything.