Tom Lenard: Hello and welcome to TPI’s podcast Two Think Minimum. It’s Friday, April 10th, 2020 and I’m Tom Lenard, president emeritus and senior fellow at the Technology Policy Institute. I’m joined by Scott Wallsten, TPI’s president. Today we’re delighted to talk to Seth Stephens-Davidowitz. Seth is an author, data scientist and speaker who studies what we can learn about people from new internet data sources. His 2017 book, Everybody Lies, was a New York Times best seller and an Economist book of the year. Seth is a contributing op-ed writer for the New York Times and has worked as a visiting lecturer at the Wharton school and a data scientist at Google. He received his BA in philosophy from Stanford where he graduated Phi Beta Kappa and his PhD in economics from Harvard in 2013. He is a passionate fan of the Knicks, Mets, Jets and Leonard Cohen. Welcome, Seth. It’s a pleasure to have you on the TPI podcast. So this is obviously not a good year for sports fans, but I guess that gives you even more time to look at data on the internet, Google data and other data. So how did you get interested in that subject?
Seth Stephens-Davidowitz: Well first, thanks for having me Tom. So when I was doing my PhD in economics, I was kind of in a PhD program where I felt like maybe I’d made a mistake because I wasn’t that passionate about traditional economics, like the interest rate, inflation, you know, I get a little bored in some of the classes. I’m like, well what am I?
Tom: So do you consider yourself an economist or a data scientist or both?
Seth: A little bit of both. I mean my PhD was in economics, but I think I kind of learned, okay, I’m not totally passionate about traditional economics, but I was really interested in people and human beings and what people are up to. And then at about the same time Google released this tool, Google Trends, which showed what people are searching in different parts of the world at different times. And I just latched onto that because that was totally my interests, like what people are really thinking about, even if they’re not telling people. Kind of a mix of economics but also sociology and some political science. And I think when I started the term data science didn’t really exist. But then I heard this term data science, I’m like, oh yeah, maybe that’s what I am. So since then I’ve been kind of exploring these issues. I worked at Google for a while, working with Hal Varian, who I know you know well, the great chief economist there and I’ve still been exploring a bit of a random hodgepodge of topics I’d say, which maybe fits my personality a little bit better.
Tom: Well Hal has been saying for a number of years that the future belongs to the data scientists.
Seth: Yeah. He’s the one who said the statistician is the sexy job of the 21st century.
Tom: Right. So, in addition to Google data what other internet data is there that is interesting?
Seth: So I think there’s something interesting in all of them. I studied Wikipedia data to learn where successful people tend to be born in the United States, and you see that cities and college towns produce way more people end up being on Wikipedia, so notable people in many different fields, than other parts of the United States, or I studied, this was more disturbing, I studied stormfront, which is the white nationalist site, and I was looking at what makes people join that site, scraping kind of all the data profiles on that site and a whole bunch of different sites.
Scott: Well, so what does make them join?
Seth: When Obama was elected, that was like the single best day for Stormfront membership than any other day. And then I looked at what they complain about, like about people of other races or Jews, you know the people they focus on and one of the things that’s very interesting is one of the main complaints was dating market competition, not economic competition. So it’s not that this person stole my job, it’s that, I saw this beautiful white woman that I loved with this African American man and walking down the street and that sent them in a rage. And that was kind of interesting cause you don’t usually think of, if anything would drive it would be more economic competition. Maybe it’s more primal based on kind of the dating competition that drives people that way. Facebook data I studied, actually that was one I did study to study sports where Facebook allows you to download for every individual age and every individual team, how many fans they have, so how many 32 year-olds like the New York Mets, how many 33 year-olds like the Mets, how many 34 year-olds like the Mets. You can actually go back and see sharp patterns based on how good the team was when those people were kids. So basically when someone’s about eight years old, if the team wins a world championship, you see consistent across all sports that when, usually men, it’s not as much strong with women, but if a team wins a championship when someone’s eight years old, those people are going to be following them, men are going to be following them the rest of their lives. And you see it across all these different sports, in that individual year of fandom data. So that was one. And then Google searches. I think what’s amazing about Google search is more than any other dataset, that any topic you study, there’s something in Google searches, because people search about everything. Stormfront has a lot of interesting information if you’re interested about white nationalism or hatred or animus towards other groups, but it’s not going to teach you at anything about other things. Wikipedia is interesting about success or notability, but it’s not going to teach you as much about other things. But Google, any topic you want, there’s probably some insight in the search data. Health or economics or politics or racism or domestic violence, I’ve been looking at that a little bit recently. Google searches can tell us a lot about that, really interesting. Pregnancy. People are looking at that. There’s just all these areas where there’s information.
Tom: So obviously the thing that’s number one in terms of people’s interest right now is the coronavirus and what we can learn from data on the internet and Google searches and other data on the internet about that. Why don’t we start out by maybe explaining to people, I mean Google had an earlier foray into this area with its Google flu tool, I guess you would call it. And that seemed to be successful for a while at predicting where outbreaks of flu would occur and then there were some glitches. You want to just talk a little bit about what happened with that?
Seth: Yeah, so Google flu was this idea, it was by started by Jeremy Ginsberg and some other scientists at Google, where they said, okay, the CDC takes about a week or two weeks to collate all their data and tell us where flu outbreaks are happening in the United States. Maybe we can see if people are searching for fever or sore throat or cough or flu, Google searches in an area, maybe we can expect flu is going to be higher in that area. And they looked back, they got data on the weekly influenza rates throughout the United States. They built a model of Google searches related to flu and they found, sure enough, there was a very, very high correlation and it was a paper in Nature. It was the front page of New York Times. It was really, really exciting. I’ve just talked to people this week who said they changed their whole life. They decided they want to be in digital disease surveillance after they saw that paper, they just thought it was the coolest, the neatest thing, and this was clearly the future of health surveillance and for a while the model was doing really well and telling people where the flu was likely be high and then pretty soon afterwards, the H1N1 swine flu epidemic came about and then all these people were searching for things related to flu and Google flu’s model said flu is going to rise to extraordinary levels and it didn’t. What really happened was curiosity about flu had risen to extraordinary levels, whereas actual flu hadn’t risen to extraordinary levels. So to some degree, curiosity and fear about flu were in the air more than actual flu was in the air. Also, there was all this tension that Google flu had screwed up and now these arrogant tech people thought they could understand the world with searches and it doesn’t work. And then quietly since then, there’s been kind of a resurgence of the flu modeling realizing that, okay, the first pass of just putting this basket with flu symptoms or this basket of flu searches isn’t by itself enough. You have to do be more cautious and picking your terms. But more recently they found that certain searches, key searches do seem to be more predictive. So sometimes one thing that might matter is particular searches. A lot of people type complete sentences on Google. It’s a little bit surprising. You don’t really know why they type things like, I’m sad, or I hate my boss, or I’m 45 years old and we don’t totally understand why everybody’s typing this instead of just boss issues or something like that which would be more common. But these searches can be very powerful related to health as well, because people type, I have a fever or I have a sore throat, or I have a headache. And those searches are less likely to be due to curiosity and more likely to be a real report of an illness. So, basically what the Google flu 2.0 I think is going, it seems like the early evidence is Google flu 2.0 is going to be effective, but it’s going to be a little more subtle than some of the earlier models, really taking into account the precise searches people are making, correlating these, having the models be ready to change as relationships between searches and flu may be changing. It’s a kind of picking up, they’d be picking up early on during the swine flu, oh some of these original searches that were working are no longer working. Now we have to drop them from the model and put these new things in the model for this time period. So a more advanced model, I think is going to prove very, very useful.
Scott: It seems like it’s kind of a subtle problem. I mean the searches need to be common enough that you can actually track the trends but not common enough that they get distorted by people’s curiosity.
Seth: Yeah, it’s a subtle problem and there are even attempts to control, like add a control, for how frequently the flu is in the news, which is an interesting attempt, kind of divide by how much you’ve been in the news. It definitely is a subtle problem. I think probably some of it is just really digging within a symptom, I mean really digging in, let’s say searches for “I have a fever” aren’t going up at all or going up a tiny bit, but searches for fever in general are going way up. That might be a clue that it’s more news than actual symptoms. Whereas if they’re both going up, then we’re maybe more likely to say, okay, this is really a symptom related rise.
Scott: You wrote about this a little bit with a loss of smell in your recent column, right?
Seth: Yeah. So initially, actually I hadn’t realized, even though I’m in this area, I hadn’t known the Google flu 2.0. I, like a lot of people, had heard about the flu success and then heard about the failure and during the early part everyone’s like, Seth, you know a lot about Google searches. Can you tell us something about the symptoms or something about kind of where the search is highest. And I’m like, during this one we have no hope because everybody is reading about coronavirus. Everybody is searching about the symptoms. No matter what search you look at, it’s going to go way up. Just from the curiosity from the news element. And I was kind of just ignoring it. And then somebody linked me an article about Michael Lewis that we could use searches for, whoa, I hadn’t even heard about this symptom. That loss of smell was the symptom. And he was saying we could do a smell test and somebody linked that, hey, are there any Google searches around loss of smell? And I was still on my skeptical mindset, but then I go to Google trends and literally you could go to Google trends now and look at the past seven days, so I go past seven days searches for “I can’t smell” was the first one I looked at and it’s just like boom, like a relationship you’d never see in data analysis where New York number one, New Jersey number two, Louisiana, Michigan, again past seven days. Longer periods, it’s not as clear because I think due to the exponential rise, the disease, the signal is only gotten really strong the last seven days and then other searches, lost of smell, also really related to, that was kind of the first time I said, okay, there does seem to be here and then other people show me some articles. Some people are doing the same thing that I did at the state level around the world and they’re also seeing, using data from Great Britain or the United Kingdom or for Spain or Italy, they were also seeing some symptom tracking, loss of smell being one them. And I think loss of smell is particularly powerful because it’s not a common symptom of other illnesses. So it’s even less likely to be drowned in noise for other reasons. And then it hadn’t been the news so much. We’ll see going forward, now people talking about loss of smell, is that gonna mess up the indicator? Did I screw it all up by writing this article for the New York Times, now everyone is going to relate to loss of smell and mess up. Well I guess it would still work because it’s New York readers.
Scott: It’s a weird Google Trends Heisenberg principle.
Seth: Then I’ve talked to people more in the health digital surveillance field. They say the news issue is really, searches around breast cancer are really, really predictive of breast cancer rates except during breast cancer awareness month. And that’s when the relationship breaks down. So it’s always, and this is kind of the issue with Google Trends data, is that curiosity or news factor that can sometimes overwhelm the symptom factor when you’re looking at diseases.
Tom: So even less well known, I guess the smell thing is getting more well known, but then you also have found that another symptom is people’s eyes hurting.
Seth: Yeah. I want to clarify that might be a rare symptom. I don’t know. It might be 1% of people. It might be 0.5% of people. What happened is I kind of looked at this, and it hasn’t risen as much to other ones, so I looked at this in the other way where I just looked at, once I saw that the loss of smell were so correlate the state level with rates of disease, I said, let me just go the opposite way. Let me just look at like a list of, I’ve put together dozens of symptoms based on WebMD, knowing nothing about them. Even if they had nothing to do with or never even mentioned with this disease and let’s just see the past seven days, which of them are in the same region? So how highly correlate are they with rate? So basically are they really high in Louisiana and New York, New Jersey, Connecticut, Michigan, which thankfully is a very odd group of States. You’re not going to really, it’s not like it’s all the Northeast or it’s all the educated areas. It’s kind of a weird mixture of States that now have this. If all the searches have risen in a high level in those areas that might be a clue. And one of them that came up very, very high was eye pain, which there’s been some discussion around the eyes. I haven’t heard too much about it. So then I kind of look more into this. Then I found, well what about around the world and you see that eye pain, there’s a topic, Google Trends allows a topic which allows you to see many different, search in many different languages which can be useful for around the world comparisons. You put the eye pain topic in Spain and you see that they rose way, like I think five times or four times their rate or whatever, in Spain in the end of March and they seem to have risen a lot in Iran. And then the eye paint topic didn’t really rise in Italy but then I just saw recently that burning eyes, the search for burning eyes rose six-fold in Italy in March and when you put together all the different evidence, I think there is some evidence, again, I don’t know, when you look at how far it actually rises, it’s not like I can’t smell where it just shoots up to levels never seen. Part of that made because the media or some of the other symptoms, fever, shortness of breath, cough, they have bigger rise and much higher levels. So may be eye pain’s 1% of people or something, but it did make me think maybe there’s something here and actually someone also showed me that Chris Cuomo, the CNN anchor, who’s been kind of chronicling his experience after getting a positive test of covid-19 had a tweet where he said the horrible thing about it, the worst thing about it is the eye pain. And then a whole bunch of people responded to that. Some people said, eye pain, that’s not, I’ve never heard that symptom. Then some people responded, me too, I know, eye pain, eye pain, so I don’t know. Part of it I’ve also learned is I think you could say any symptom at this point and people will latch on and say, I had that, I had that, and have that theory. There’s so many people that have had the disease, are so focused on disease right now that they kind of latch onto to any theory that went out now.
Scott: It seems like there’s also a risk of spurious correlations doing it that way. Google used to have this Google correlate and it would show you these hilarious correlations, which they perfectly matched the trends but they had nothing to do with each other.
Seth: That was also an issue with Google flu where, well I think one of the, if you just went backwards a bit, the what searchs are most correlated with flu rates. I think one of them was like NBA season or something like that, something around the NBA and probably cause the flu has a strong seasonal component matching the NBA component. That’s one of the reasons I limited it to like dozens of symptoms. I think when you do dozens of symptoms to get a correlation that high, if you do dozens of symptoms only one them really should be statistically significant. These were things that were coming out with T stats of seven and eight and nine and 10.
Seth: And then going to other countries, I think you do see it as well. Again, I don’t think it’s a slam dunk. I said it should be for further research. I told a lot of doctors because I’m like, look, I don’t know anything about public health. Am I saying anything stupid? And they’re like no, you know, worst comes to worse. Someone who has eye pain’s a little more cautious, wearing mask going out to a supermarket and maybe they don’t spread it to someone who otherwise would have gotten it. So kind of the worst case scenario is probably the medic might help a few people. You’re not really hurting anybody by knowing the correlations and they’re telling people to explore
Scott: The unintended consequences might just be good ones.
Seth: Yeah, I mean you never know. Anytime you do something, unintended consequences can move in all different directions. I got an argument the friend of mine, cause someone said that nobody who’s not an epidemiologist should write anything about this topic. Like you’re literally getting people killed by your stupid math, like physicists or something, I’m going to write a new model of the disease. I’m like, to some degree, if you have something that’s interesting or different that might be right, you’re getting people killed by not writing that. Just saying you couldn’t because the story is so much in the news and it related to death and disease. Like anything you do, the unintended consequences could lead to more loss of life or less loss of life. We don’t really know. And I don’t think this is a situation where the experts have it totally, A lot of my best friends are epidemiologists, they’re brilliant people. But this is one where it doesn’t seem like the experts have it totally under control and know exactly what to do and we just need to wait for them to deal with it, to solve. Listen to their them and then the problem solved. This does see more a situation where it’s all hands on deck. We don’t really know what’s going on. We don’t have a great solution.
Tom: How much predictive value to these, as in contrast to just telling people what’s going on right now, how much into the future, how much predictive value do these data have?
Seth: Yeah, I think that’s remains to be seen. So the early version of Google, before even Google flu trends was Hal’s Google analysis, which I don’t know how much you know about, but he called it nowcasting, predicting the present,
Scott: That’s one of my favorite titles.
Seth: Yeah, predicting the present is a great title. Hal’s like Mr. Wit. Like just always has witty ways to say things. So nowcasting, predicting the present that he did what Google flu did for the economy, which is, let’s look, with a coworker of his, Choi, I forget the first name, but it was 2009, it was basically, okay, when people search motorcycles, do they buy more motorcycles? Well, lo and behold, yes they do. The searches for motorcycles and sales for motorcycles are very, very highly correlated. Do a Google search for auto insurance, they buy more auto insurance and you kind of go through one category after another, unemployment benefits. When people search on unemployment benefits, they collect more unemployment benefits and the idea is these economic variables sometimes take one week, two weeks, three weeks to come out, four weeks, we can actually know what’s going on now. And I think that is, almost all the research I’ve seen on Google searches has been predicting the present. I haven’t seen much like let’s try to predict the future using this data. It would be interesting. I don’t know. I think, I think the predicting the present hasn’t been nailed down yet. You have these issues like Google Flu where the model kind of blows up a little bit, where everyone’s kind of like let’s first show that we can predict the present and then maybe see what we can do. First of all, predicting the present, we haven’t nailed predicting the present with Google trends and we haven’t nailed predicting the future with traditional data sets. I think people are a little scared saying, are we gonna be able to use this to predict what people are going to do a week, two weeks from now. But you definitely could imagine, I was recently talking to NABE, National Association of Business Economics, and they said, can you see how long people think this is going to last from the Google trends analysis and I was talking to Taylor Schneider from Adobe Digital Insights who also has limited digital data and he said very intelligently that you have to be more clever. It’s not like on these internet sources, people are saying this is how long something’s going to last, but they’ve been monitoring economic activity for office supplies or back to school stuff or what about like wedding dresses or wedding planning? Like there are probably ways through the analysis you can have a sense of what people expect the country to look like in a few weeks and that kind of gets to the idea of predicting the future. Maybe wedding dresses or wedding supplies or something is a way to predict the future. Another one, so there’s been this question, is there going to be a baby boom from the covid-19 pandemic? And you can imagine going either way, because either you could say, okay, now all of these couples are locked in their apartment together. They’re going to be having more sex. Nine months from now we’re going to see a rise in babies. That’s the type of thing you’d see after a blizzard when nobody could go outside. What happens nine months later, baby boom.
Tom: Well the famous one was the Cuban missile crisis, where nine months after that there was an uptick in births.
Seth: Yeah, I didn’t know that. Thanks Tom. But that’s one side, but then the other side is, well no, economy’s crashing. Everyone’s unemployed. What happens when the economy does bad, births go down, because people feel like they can’t support another child. So this wasn’t me, it was a company called ARK Invest who also had been monitoring trends and they put together a chart where they’d been looking at seasonally adjusted searches for pregnancy tests and they found a big drop so far. So far it looks like, that might be an early indicator that we’re not going to get a baby boom. We’re more likely to get a baby bust out of this.
Scott: But what are the divorces or other negative things?
Seth: Divorces does seem to also have gone down, divorce serches seem to have gone down. You know that was little bit surprising to me. Maybe people are so busy, they don’t have time to think about what they think about their spouse. They’re like in survival mode or something.
Tom: But they’re trapped in a confined space with their spouse. So that could go either way, I guess.
Seth: Yeah, it’s again, one of those questions could go either way. And so far, the search data, I looked at it a week ago, I haven’t looked at the updated, but I thought there might be a rise in divorce searches and they seem to have been dropping. Yeah, it did surprise me cause I kinda thought, I have a darker view of marriage. I have Tom’s view that if your confined with your spouse for extended period of time, they’re going to drive you totally insane.
Scott: I also heard a joke, sounds like a Daily Show or Colbert thing, I can’t remember. But that we may not have a baby boom, but we’re gonna have a huge podcast boob.
Tom: So basically the Google data is not probably gonna give us a lot of help in trying to figure out when to restart things.
Seth: Yeah, not here maybe. But, so one of the reasons I was looking at that symptom stuff is my epidemiologist friends said that it can be really helpful in parts of the world where testing is bad to actually know how many people have it the disease
Tom: Like the US, right?
Seth: Yeah. Well yeah, what part? Yeah, how much in the US, how great is it in the US but definitely as bad as it may be in the US, some parts of the world, it’s just a total mess. I talked about in the piece how Ecuador is searching for loss of smell more than anywhere else in the world. Even though if you actually look at the rates of positive tests per capita, it’s higher than other places in South America, but it’s not up there with Europe or the US or Australia or Canada. And they could be an area where they might look at their official data and say, okay, it’s not that bad. We can loosen the reins and you look at the search data and say, no, everybody’s complaining that their losing their smell. Let’s wait this one out a little more. So if the testing data were bad enough, you could potentially use this as another indicator to know, I guess, opening up. I mean you could also figure out other problems. I’m just like brainstorming, but I’m sure there are others. I think there are going to all these second and third level effects from shutting down our economy. And I do wonder if Google Trends could be helpful in uncovering those, not even the ones we expect. Like it may be just like looking what searches are rising and we may find out a problem that we didn’t even know about that is caused by this situation that could be worth exploring. That wouldn’t shock me cause we’ve never really done an experiment where we just said nobody go out, everyone has to stay at home. That’s just such a dramatic change in society and it wouldn’t surprise me at all if there a whole bunch of second and third order effects that nobody’s predicting right now. And maybe Google Trends, we have one place to uncover them among other places.
Tom: Well, aside from the health effects, have you uncovered any changes in other things? For example, is there any change in the amount of political polarization in the country or have you tried to look into that?
Seth: Yeah, I haven’t looked at that, the Google searches. That’s one, you may just look at the, there are some questions where I think the poll, the survey may just be the better way to do it. And I don’t know that Google Trends the best way to measure polarization. I’m not sure. Maybe it could be. I haven’t looked at that. I looked a little bit at, Brandon Nyhan, he’s a professor at Dartmouth, reached out to me and suggested what about anti-China sentiment in Google searches. We didn’t see as much as we were expecting, which was almost interesting. We didn’t see like a big rise in searches. Like sometimes you see these really nasty searches people make, like people make, not even search for Chinese flu, replace Chinese with like a five-letter epithet for Chinese and they search them like that. But the rise was pretty moderate. Like a few people were searching that, but it wasn’t like, oh my god. Like there are other periods where you see like a real explosion of racism in Google searches. I wrote about a little bit how many people searched for really nasty jokes about African Americans when Obama was elected president. And that was just a huge factor in the data. But China’s stuff is not really like, oh my god, there’s some widespread anti-Chinese sentiment.
Tom: That is interesting. Is there anything else that might be surprising to listeners in terms of types of searches that seem to be associated with this particular period?
Seth: So one of the things is when the National Association of Business Economists, they’re like, what’s going on in the economy? Their question was what’s going on the economy with Google searches. They had the same question as Taylor Schneider, who’s at Adobe who’s monitoring, they built a purchase dataset. I forget exactly how they put it together but their monitoring purchases, high-frequency data. And both of us were like, there are huge changes to the economy, but they’re almost all what you’d expect. Like teleconferencing is way up. Like sports tickets is way down. It’s like unprecedented levels of changes. We’ve never seen a 200% rise in teleconferencing or 80% drop in sports tickets, he was talking about the purchase data as the same thing. Like you know, toilet paper, right up, home exercise equipment up. That one was maybe a little bit more surprising but still you could. I think the big changes were all things that we would have pretty much guessed what would happen when everybody’s home all the time and a little less had to work. But it’s definitely an area to monitor, particularly, I think it might be over time, like the more interesting ones. We were kind of focusing early on like the big changes and the 60, 70, 80 to 200% rises or the 60, the hundred percent drops are all things you’d expect. I think the more surprising things will be the 10% rise or the 10% fall where it could’ve gone either way, but those take maybe a few weeks to really kind of settle out that it’s not just noise. Like the surprising thing is I don’t think you’re going to see like toy purchases. You could say like it goes either way. There’s an economic hit, but then kids are around, kids are home a lot, so they may be demanding more toys. You could imagine it goes either way, like the search is so far about flat, but like maybe over a few months it’ll settle into, okay, this has led to more toy purchases or this had led less toy purchases, but it’s not going to be like one of the big changes. And it’ll be interesting to see some of these categories that are maybe more uncertain. It’ll be interesting over the weeks as the noise kind of dies down to see kind of where they settle in this new normal.
Scott: Have you seen any interest or gotten any requests from government agencies trying to track the economy? So in the financial crisis, the Fed looked for all kinds of ways to monitor things in real time faster than the economic data normally came in. Are people talking to you about that or do you see them reach out?
Seth: I think they’ve reached out to Google a lot, so there definitely are people looking and monitoring this one. This one is just, to some degree, like we know that changes are enormous. We know, kind of, where the changes are. It’s more about how is this going to stop. Again, we know nobody’s buying sports tickets right now. We know bars and restaurants, people aren’t visiting them or people aren’t flying as much or looking up cruises. Like all these changes are kind of a little bit of a note. If we know the changes that are enormous, we know it’s like, okay, when are we gonna stop it? They have had some success predicting the unemployment rates using searches for file for unemployment. A friend of mine, Paul Goldsmith-Pinkham, and Aaron Sojourner have been working on that and they built a really nice model of predicting unemployment benefits, but it’s another one where the Fed knows that unemployment’s going to unprecedented levels, it’s not really stopping or anything. That’s already built into the models. More the question is how do we stop this? How do we fix it? How do we get people up and running? When do we get the economy up and running again?
Tom: You say you talk to a lot of epidemiologists. Are they incorporating any Google search terms in their models?
Seth: Yeah, so there are some people are doing this, there is actually a field, digital disease surveillance, and it’s a growing field and there are, again it’s a young field so it’s more in the exploratory phase, although I think this crisis is kind of pushing it to the emergency phase.
Tom: You know if, the models they talk about that seem to be well publicize, University of Washington model, Pennsylvania model, and I’m sure a couple of others. Do you know if they incorporate Google search terms?
Seth: I don’t think they are incorporating Google searches or you know there are other tools to use. I just read a paper on using the uptick of influenza like illnesses. You see an uptick in influenza like illnesses reported in different parts of the United States, seasonally adjusted, since the crisis started with most of the upticks in places that have had big outbreaks like New York, New Jersey, Louisiana, Michigan, that could also be used in the model, which to the best of my knowledge aren’t being used in the models.
Scott: That seems to be the kind of data that New York city collects. They report number of people hospitalized with flu like symptoms or something like that. So it’s exactly the data that they seem to want.
Tom: Well, I think we’re approaching the end of our time. Scott, any last questions?
Scott: Nope, this was great.
Tom: Well, this was a pleasure, Seth and we will have to do it again sometime.
Seth: Yeah, definitely. Hopefully with this all behind us.
Scott: Thanks so much. Bye Seth.