My name is Daniela Rus.
I'm the director of CSAIL and Deputy Dean for Research
for the College of Computing, and I'd
like to welcome you to today's Hot Topics in Computing.
We are delighted to welcome today two of our very own--
Professor Esteban Moro and Professor Sandy Pentland--
to talk with us about the Effectiveness
of Social Distancing in Protecting
People Throughout the Pandemic.
Esteban is a visiting professor at MIT IDSS,
and he's visiting us from Universidad Carlos
III in Spain.
His work lies at the intersection
of big data and computational social science.
And Sandy Pentland directs MIT's Connection Science Initiative,
which is an MIT-wide initiative housed by IDSS.
He has helped create and direct the Media Lab,
and has made numerous contributions for the world
through his efforts as part of United Nations
committees and World Economic Forum committees.
So without further ado, Esteban, Sandy, please take it.
So, let me say.
I thought I'd do just a little bit of introduction
here and give most of the time to Esteban.
So, our group is over in IDSS, part of Engineering,
but also includes Sloan and Engineering and SA+P,
32 people, six faculty, senior staff formulated here.
We do data analytics-- like you're going to see,
you know, to popular words "social physics," privacy
and security, are pre-standards software.
I took over a part of ISET when it was downsized,
the part that did Kerberos.
We-- together we did Open Connect authentication methods
and things like that.
So-- and we're continuing on that.
And we're sponsored by World Bank, Inter-American
Development Bank, France, Italy, Australia, Columbia,
as well as a bunch of large organizations
around the world trying to solve their problems.
And you're going to hear something exciting about that
in just a couple minutes.
Here are some pointers--
Connection.mit, trust.mit which is
the more sort of software effort, and the newest one
So we put together an association of law schools
around the world to think about what is computational law
And you should take a look at that,
because I think it's going to be a very significant thing.
Now that we're in the middle of the pandemic,
we're doing a bunch of things.
So, we're helping support the Safepaths at MIT effort Ramesh
Raskar has started by doing some other crypto and interface
and human interaction stuff.
And personally, more focused on the economic restart.
What are we going to do after all this?
And I work with an amusing group which is called Club de Madrid.
It's all the former presidents and prime ministers
in the world formed a club, and they
are looking to get behind the right sorts of initiatives,
make sure the world doesn't sink into a economic depression.
We do labor impact and training with Sloan,
Initiative for the Digital Economy are one of the leaders
I mentioned a legal framework, and we're a big data group
and we do fine-grained analytics.
And Esteban-- who is going to talk to you in just a minute
has been doing some of the very first fine-grained analytics
for distancing policy and trying to guide
decision makers in this.
So that's all I wanted to say, just introduce who we are,
and maybe I'll hand it over to Esteban.
I'll turn myself off there.
Esteban, you're in charge.
OK, thank you very much.
Can you hear me?
Can you hear me well?
OK, I see you.
Yes, OK, thank you very much for having us.
As Sandy was saying, and Daniela.
So we are part of this collection science
and ideas as a media lab, our group.
And we have been working on all the analytics now of data.
So I'm going to explain today something that we are actually
So this is actually already some result.
We have a recent result, but keep in mind that this is
a working paper, so we have--
might be up to speed.
We have things are changing every day.
We are actually having more results by the minute.
So OK, I'm going to talk to you today
about something that we have been doing in the last two
And we are having actually-- working around the globe
with a lot to providers with who are grouping collection signs,
doing the analytics.
And the idea, guys, who tried to use high
detail probability data, high detail data vault,
how people is behaving right now.
I had nothing to inform about the social distancing
strategies that we can take in this pandemic,
but as Sandy was saying also, we are
interested in the-- how we are going to exit this situation.
If we are going to flatten the curve,
eventually we will have to reconnect our society
and discover many directions that are going to be impacted.
Not only the epidemics itself--
which is what we are going to see today--
but also the economy and the long term in our society.
So I guess you all really know all this figures,
and probably you're checking the same number I do.
But today, there is around 165,000 cases
in the US of infections of the COVID-19.
There are more than 30 countries with more than 1,000 cases.
I'm from Avila, from Spain.
Spain is one of the countries that
is suffering most right now, so the situation there
is affecting not only the economy, but also the hospitals
are over capacity and, of course,
it does it personal drama and personal situation in which
some of us are actually been died.
Some are losing some family because of this.
So what is happening in most of these countries
is that we already passed the containment phase,
and we are now in mitigation and suppression strategies.
So most of the countries are implementing right now
what is called social distance strategies, which are basically
restrictions about that--
about how people get together and procedures
about mobility restrictions and contact restrictions.
So we are basically not allowed to go to a workplace,
or people are not even allowed-- for example, like in Spain--
you are not even allowed to step out on the street.
So you have to stay in your house.
And the idea of that is that you have to meet--
they want to minimize the number of contacts
and the number of people that are actually
out there infecting each other.
But the question is whether this sorts
of strategies, social distancing strategies are working.
And actually, the most important is
to try to understand what is the impact, how
this social distance is actually translated
into the outcomes of the epidemics.
There have been many people actually
addressing these questions.
We are not the only ones, but our group was actually
in a very funny situation.
We were actually modeling the influence
epidemics of every [INAUDIBLE] let's
say a fluke happened in the Boston area.
And then in the last two weeks, we
pivoted everything and started looking at the COVID situation.
And we were now already in this business situation
because we were using for the first time
these high resolution mobility data about how people move
into cities, and trying to understand how people are
getting contact in the cities.
Not at the level of, big-- let's say compartmental groups,
but at the level of actual individuals.
I'll explain to you a little bit in the rest
of the talk about this.
So we are going to use high resolution mobility
data to understand the impact of social distancing policies
in epidemics, but also in people's behavior.
Data that we have comes from recently collaboration
between GovLab and Cubic.
Cubic is one of these companies that
have mobility data that comes from mobile phone applications.
We have data that basically is the movements of millions
of people in the United States at the level of places,
and we were-- what happens.
We don't have data.
It is anonymous.
We don't have access to the raw data,
we just have aggregates, which is what you're gonna see.
But the data has typically a precision
of 5 meters, so we can tell people
where people are in one restaurant or the restaurant
And it is a typical time between events
is about around 15 minutes, so you
can have complete trajectories.
There is a lot of privacy issues with this data,
and I'm happy to discuss that.
But as I was saying, we don't have access
to the individual data.
We just have access to aggregates of data.
So the privacy considerations here
are somehow solved because we are dealing with aggregates
at the level of [INAUDIBLE].
The kind of data that we have looks like this.
Let me show you this.
This is Boston downtown.
Each of the points that you see there in the Boston area
is a person that is sustained in that particular place for more
than five minutes, and this is a typical day in the Boston
area before the pandemic.
What you can see here is that people are
saying very different places.
And as you see here, that people are staying different places.
The color, by the way, the color of the points
is basically a socioeconomic proxy for that people.
And you can clearly see that when they starts,
people get together in very different places.
We have places, for example, like in the Back Bay
or in Media Lab, or MIT in which people
can-- come from many different places, get together.
So this is the idea that yes, the data
that we have allow us to understand where
the contacts are happening, and how the mobility of people
is actually impacted by this social distances measures.
I'm gonna represent preliminary results
that I was saying, because this is an ongoing work.
Just today, we have to change a little bit the first part
because we are constantly talking to the epidemiologist,
we're talking also to policymakers,
and we're also talking to all the companies that have data.
So what you're going to see today
is our preliminary report about two works.
The first one is about the effectiveness
of social distancing studies, and the other one
is about how is actually the social distance
measures translated into people's behavior in the New
York City area.
And let me go to the first one, which is actually
what we're talking about.
The question about this work is to combine these
high-resolution mobility data that we have that inform us
about where people are getting contact--
and eventually, where people can get infected together--
with epidemics of the COVID-19 in the Boston area.
There are many questions that we want to address with his work,
but this basically, what is the expected impact
of the current social distancing strategy?
The good thing about having this high detailed mobility data
is that we can simulate the scenarios.
Like, for example, closing down only the restaurants.
Or closing down only the public parks,
or closing down, for example, the working places.
So we can actually implement-- and you will see,
I mean, we have up to six different strategies that
have been implemented in different countries
and in different moments in the past month.
The question that we have also is,
I know, there's another parameter
in the social distancing measures which
is how long should be the social distance measures, which is how
long should they be in place?
And obviously, we want to understand
this in the context of only on the infection.
I mean, how many people are going to be infected?
How many of these people are going to be ill in the end,
or how many people are going to go to the hospital?
And eventually, if we flatten the curve
and these social distance strategies happen and work,
the question is, if we're going to have a second wave--
and if this second wave happened--
how are we going to get ready for that second wave?
So that's the main thing and the main purpose of this work.
One of the problems when we try to simulate is that COVID is
actually-- you probably have seen a lot of papers out there
trying to simulate the epidemic spreading by using very simple
models like this, to compartmental models--
in which everybody basically is put into one of these buckets.
You're all susceptible, very infectious or recovered.
This is too simplistic for what is happening right now.
And actually, so simplistic that none
of the social distance measures can be implemented
in the simplistic models.
We need, actually, to specify much better
where people are, how people move around the city,
and how, for example, closing down the schools
is going to affect this infection, or how,
for example, closing down the restaurants,
are going to do this.
So we need to combine the people's behavior
with this model to really understand
how social strategies are going to work.
So this is why we're doing this work.
We are going to combine these models--
these solution models--
with mobility of people in the Boston area.
In order to do that, what we have done is actually,
we are going to do a very simple basic relation, which
is based on the mobility data in the Boston area.
So we took the mobility data that happened one year ago,
and how people move around the city.
And we divided the mobility--
divided, let's say, the contagion
of the places in which people can
get infected into four layers.
You have on the right in Panel A here.
There are four layers.
One is the school layer.
The one is the community layer.
The other one is a household layer,
and the final one is the workplace layer.
It's very important that we distinguish
between these layers, because the infection and the dynamics
of the infection in different layers is very different.
The good thing about having these Cubic data
and the mobility data is that the community layer--
which was typically assumed to be a homogeneous layer in which
everybody gets infected with a custom probability,
we can actually substitute that with the actual data,
because we know where people go to that restaurant, which
is the people that-- who is the people that this person gets
in contact to, et cetera.
So what we did was, absolutely use
this data to simulate 100,000 agents in the Boston area,
and the infection is happening in different layers.
And as I say, the communities they
use based on the data on the household, the schools,
and the workplaces are taken from the census area.
You have almost all the details about the simulation in them,
and about how we selected these 100,000 agents in the Boston
area in the paper.
It's very important also to distinguish,
so to know that the model has--
in different layers-- have different parameters.
So the infection, for example, in the schools
is different from the infection that
happened in the households, and the parameters
are actually different.
So what we did was to extend, a little bit,
the model for the COVID, in COVID
be there is another bucket, which is exposed--
is where people actually get exposed to some of the virus,
but they don't develop the disease,
or they actually have asymptomatic--
so that they don't develop the disease,
and there's another one in which you become infectious in which
you can actually just spread the disease.
What we did, is because we already
have the model calibrated, the layers
calibrated to reproduce the post on fluidity and influence
decisions, we know the parameters
in which infection happens in the different layers.
But finally, this is only to calibrate the layers.
But finally what we did was to use the literature-- especially
the paper, probably you have seen
the be paper by [INAUDIBLE] College
which is up to date tempo now because people are using right
now the literature for how you move from one bucket
to the other one in this model, how you move
from susceptible to exposed to supposedly infection
to suspection to recover and also to recovered.
If you come back to susceptible, in the case of this simulation,
we have it we have assumed that is non-existent.
OK, so first we calibrate the infection with the layers,
and then we calibrate them the way we use the model,
it's probably because people aren't using
the model in most situations.
First of all, let me discuss a little bit of the limitations
that we have so far, because we have not actually
done this so far.
Age is not included in our model.
So we don't have buckets by age, which
is very important for COVID-19.
It's really, really important to include age,
and I can tell you at the end what we are doing
to alleviate this problem.
The other one is that it is not a true operation.
The operation is closed, we are assuming
that this is the Boston area.
We don't have influx of people coming from other areas.
And so in this preliminary report,
we don't include the difference between mild and [INAUDIBLE]
cases, which is very important.
But I can tell you of after I present this,
what we are doing right now.
We implemented-- I don't want you to read this transparency,
but we implemented six social strategies
that people have been using and the different policymakers
coming in, implementing.
The first one is just school closures.
The second one is self-distance by teleworking.
Self-distancing means that you stay home,
around the fraction of people staying home.
And they are teleworking.
And they don't go to the restaurants.
That's self-distancing, for example.
The other one is self-distancing plus teleworking
plus school closure.
The fourth one is just restaurants
and nightlife and cultural closes.
So we remove all the places in which just a lot of people
are getting together.
The fifth one is that we close all non-essential workplace
closures, which is another study that, for example, New York is
using right now.
And obviously, the last one is total confinement.
Everybody basically is stay home apart
from the essential workplaces which is still going on.
As I was telling you, we have where people go,
so we have a category of the places, which,
we know, for example, whether this is a factory or a police
So we know the difference between,
for example, what is nonessential or essential
So let me just briefly summarize the result.
You have all of them in the report.
Basically what happens is, for example, we do
do all of the school closures.
The more you close the schools, the better it works.
The epidemics is delay, but only a school closure doesn't work.
So even if you close the schools for 90 days,
and we are now looking at, for example, I
don't think the schools are going
to be open until September, so we now
look at a number of 180 days.
Then, the problem is that only the school closures, what
is going to happen is that the kids are going
to infect the parents in the households,
and then the households are going to propagate the epidemic
throughout the working places.
So only closing the schools is not gonna work.
This is something that probably everybody was expecting.
But when you combine these with other social strategies,
for example, like social distances at 90%,
or social distance by the school closure, two things happen.
The first one is that we delay the peak.
So not only we flatten the curve very easily,
but we delay the peak.
And we buy, basically, three-four months of time.
The peak is going to happen.
You will see all these in relations.
There is a second peak happening,
but by doing these social assistance strategies,
what we see what we can do is to buy time
until the second peak happens.
But also, we can decrease the--
Sorry, could I ask a question?
Yes, you-- go ahead.
What sequential decision-making model or framework
is being used in the simulations to address the created
assignment question, that is, which closures are actually
useful and to what extent?
Yeah, so basically, there's many parameters you can use here.
Like, for example, the amount of time
that these strategies imposed, or whether you
can have one after the other.
We are actually working on that.
But right now, everything is actually done after 100 cases,
after we have 100 cases in our city,
we implement those social strategies,
and the only parameter that you see here that is varied
is the actual time in which those social strategies are
Thank you for your question.
We just pretty much work-- policymakers
are doing right now, after 100 cases,
they have started doing this strategies
and they keep social strategies for [INAUDIBLE]..
I'm going to mention it a minute also how
you can do active strategies.
But I was saying, I mean, even though we
flatten the curve a little bit, or we
flatten completely the curve right now,
there is a second wave coming.
And actually, the probability that we have a second wave
after implementing those things in after 90 days, for example--
is really large.
The probability that we are going
to have a second wave probably during the summer or early
in autumn is huge.
Apart from the social distance school closure,
this is just going to buy-- this is probability to find a second
peak in 90 days-- the next 90 days--
apart from the one that I was showing
you which is the school close plus 90 social distancing,
which is the one on the right.
The rest of them are going to have a peak in the following 90
So we have to be ready for the next wave.
Our simulation tells us those social strategies
are going to just buy time.
If they are the only things that we do, just going to buy that.
So we have absolutely to be ready for the second wave.
And I mean, we are going to basically restart.
We are going to come back to where we are in February.
What we are doing right now is actually--
this is very preliminary.
And since there is always a probability
for this [INAUDIBLE],, we need to actually understand
what are we going to address?
What are we going to do when the second wave comes?
Of course, we need more testing.
So one of the things that was not happening
is that we didn't know that it was actually one big--
one infection going on.
And we need to be able, naturally,
to do more testing between--
even for people which are not symptomatic.
And I'm going to comment in a little bit on that.
But of course, one of the results of our simulations
is that just passive strategies-- like, for example,
the lockdowns, or preventing people movements-- is not
going to stop this.
I mean, apart from the pharmaceutical,
let's say, measures and eventually, we have a vaccine,
this is going to solve the problem.
But if we just do this measures, like, for example,
closing the schools, or closing the workplace,
this is going to work.
It's going to buy time.
By the time we actually resumed activity,
the infection is going to spread again.
So we need more active measures, and we
are going to work-- we are actually working on these.
We need to, for example, do contact tracing.
And this is something that some people
are saying, is a safe bot application
that we are building--
we are helping building-- in our group.
It's about that.
And we are-- because we have the data about how people move,
we can actually look in the simulations
how the contact tracing is going to work.
And you can see preliminary results here
in this transparency.
But by doing 20% contact tracing after the infection
of a person, by keeping the contacts and the people
that are in the household in quarantine,
actually diminish completely the infection.
It operationally prevents the spreading
of the virus of the [INAUDIBLE].
But to do that, we need testing.
We need a lot of testing.
We need to test people whether they are actually--
even if they are known as symptomatic.
So this is what we are working right now.
I wanted to dig in a little bit on this point.
So you said only 20% contact tracing will help
flatten the curve completely.
Is that right?
Did I understand you correctly?
Well, I mean not completely, because I mean,
you can see here that also, there is--
I mean something that is happening.
But I mean, the more--
I got to say contact tracing that you do, the better.
But what we see is that from 20% onwards, I mean,
there is actually on this axis from below 20% to above 20%,
the change is dramatic.
So how do you reconcile this was what happened in China?
So China was very aggressive about contact tracing
and they, I guess, I imagine they traced everyone.
And they have that they have great success.
But it was a kind of a universal approach.
Well, the China approach, I think, is twofold.
I mean, they did a very severe social distancing measures.
People were actually not allowed to actually step
on the street from the very beginning
after they have, like, 1,000 cases.
So it was really, really strict.
We cannot do that in the US.
We are far gone from that time, because people are already--
I mean, we have--
The other thing is that people actually were--
when they were actually locked down,
they started doing this contact tracing.
But I mean, I think that where it is where they actually
prevent this spreading by having this severe lockdown
measurements in the first place.
What I understand of the contact tracing,
it wasn't nearly as good as you would hope,
because they were in flat-out emergency mode.
Most of the tools, those sort of surveillance tools,
were used to make sure people stayed quarantined.
We have, perhaps, the ability to do better contact tracing
that they were able to do.
The other thing that I was telling you,
this is the model that we are using.
And we do so, these to the people
that are actually on the CDC, they say, no way.
This is not.
This is not going to help.
Actually, the model is this one.
This is model that we are doing right now.
The model is super-complicated.
And you can see here that the model goes
from the susceptible to the people
that have symptoms to the people which are infected,
but they don't have symptoms, to the people that
go to a hospital but they don't have any symptoms,
and the people that are infected that go to hospital
who have ventilators.
So depending on this, this is actually
what the CDC is working right now,
and we're working with people that have contact with them.
This, actually, is the model that we are doing right now.
Because the important thing right now
is-- one of the most important things
is-- how the number of infections
translate into the stress of the system, into the health system.
So we need to understand not only what
is the number of people which are infected which is good,
but also how many people are going to end up
in an IUC with no ventilators, for example, or with an IUC
or in a hospital that has actually no vents, for example,
like, for example, what is happening in other countries.
So we need better models and this
is what we are doing right now.
I can tell you the sophistication
of these models is huge.
So it's complicated to simulate, because there
is many, many social strategies and that you can keep.
This is a very special infection,
because compared to, for example, influenza,
you can be asymptomatic but you can be infected.
And you can actually be-- this is actually the first line
that you see here in the model.
You can go from susceptible to infected, asymptomatic,
and then recover easily, but you can still
spread after being infected.
OK, so this is one part of it, and the other one
that we are doing is that notably,
and there is, in the last couple of weeks,
we have access to more data about how people
are moving in New York City.
So one of the things that we wanted to test
is whether their social distance policy is actually
working, if people are actually doing this social distancing.
If it's working or not, which groups,
which demographic groups are actually doing this more if
they can afford to do this.
And whether that social distancing
it's impacted the behaviors of people, and in what way.
So New York City is right now, they
are the epicenter of the epidemics.
They're just sick, and now where they
have more than 75,000 cases, more than 1,500 deaths.
The further out the governor and the mayor
here have imposed really severe lockdown measures.
Like, for example, by March 22, they--
all the nonessential businesses are already closed.
But this started, this study probably March 15
when the New York City schools actually
close up until May the 5th.
So there has been many, many studies.
And you can clearly see, for example,
how many, many cities are actually
reproducing the New York City model in which
try to close the school, that you close,
for example, hair and nail salons, and barber shops,
and then eventually you close the nonessential business.
You can clearly see that it goes very fast.
So in a matter of days, they are actually
changing from implementing new [INAUDIBLE]..
The question is how these policies translate
into the social business practice,
and how this actually is related to behavior changes of people
in their city.
Since we have this mobility data,
I can tell you that we have mobility data of yesterday,
so we can actually see what happened yesterday.
We can see how social distancing behavior has changed before
and after the new policies, how it varies
across the physical space of the city,
are then some places which are more affected by these.
Are still people gathering, for example, in the public parks?
Are still people gathering in other places?
And of course, there are some demographic groups
that can afford to sort of distance themselves, typically
because they can't telework at home.
But there are other people that cannot do this kind of social
distancing, so we want to understand whether this is
impacting more high income versus low income groups.
What we have seen, what we have done
is to measure a number of metrics
which we call mobility metrics.
Like, for example, the average distances
that people travel in the New York area,
or the average [INAUDIBLE] of the ratio, which
is the typical area covered by people in the area,
or the number of people who [INAUDIBLE]..
As you can see, after national emergence, the school closure
and the nonessential closure, the number
of the average distance traveled,
the reducing duration, and also the percentage
of people whose income has dramatically changed.
We can, for example, see that the average distance
travel went from 40 kilometers to roughly 15 kilometers, which
is actually a dramatic change that never happens,
that you may, for example, during holidays in the New York
The same thing with the percentage of people
staying home, which went from roughly 50% or 20%
during normal days, to up to 60% of the people staying home.
So yeah, I mean, those things happen.
Those policies have a tremendous, actually,
impact on move and behavior in your area.
You can also see that it's slightly
starting to happen before the national emergence,
and is probably related to the fact
that it was [INAUDIBLE] made, for example, NBA close
I mean, declaring that they shut down the disease on March 11th,
and also there was some banned international travel
on March 9.
So all of these things actually impacted
the people [INAUDIBLE].
This is how it looks, how New York on the left.
You have the average distance traveled
by the people living in each of the census tracts in the New
So each of the [INAUDIBLE] that you
see there is basically census tracts,
and the color is basically the average distance
traveled by the people living there.
You can clearly see that areas like [INAUDIBLE],, for example,
Manhattan, they have people driving less because they're
But you have all this have [INAUDIBLE] behaviors which
are in New Jersey on the way from the city for the New York
City people traveling more because they commute basically
And this, on the left, you have the weekend of March 6th.
And on the right, you have the weekend of March 20th.
And you can clearly see that average distance traveled
by people are really, really small all over the place.
It has affected most of the areas in New York City.
We also see that a number of visits to third places--
what I mean by third places are restaurants, shops, groceries,
schools, parks, et cetera, has dropped significantly,
which is Panel A here.
You can clearly see that it has dropped
from typically 1.5 visits per day in our population,
to basically nothing after the measures were imposed.
But there was a problem, actually,
when after the announcement of March 14,
actually, before that, in March 12th, a number of people
started piling up in grocery stores, which
is why you see in Panel B that there was a huge spike
and a huge surge of more than 60%,
more visits to groceries, which has been actually steadily
being the largest, by far, place that people visit
in the last couple of weeks.
OK, so just announcing these, preventing people
from going to the schools, for example,
because when they were closed or to, I don't know, food,
but everybody went to the grocery stores.
And they only-- it only actually appears
to normalize after the non-essential business closures
were announced on March 22.
Regarding-- because we have a mobility rate now,
how people move in this area, we can actually
calculate how close are people.
We can actually compute the proximity network.
And by the proximity network, we mean that the two persons are
within five minutes, they're sort of
within 25 meters for more than five minutes in the same place,
we consider them to be a social contact,
or they're in the proximity.
This is the social contacts as a function
of time in the New York area.
So typically, people in the New York area
are in contact with 75 to 100 people.
It depends also on the weekends, we get more contacts.
But after the social business metrics were imposed,
the number of contacts dropped from typically 75 to five.
And I'm actually-- that happens well
before nonessential closure, because not all the people
actually anticipated that it was going to happen,
and they started staying home.
This is very good.
This results shows that this is working
is the direction that we know that we need is that people are
actually having less contact.
So we are actually preventing more secondary infections.
But the bad news is this, is that we have changed completely
So the places in which people are having more contacts right
now, it used to be food.
Food restaurants in the New York area
were the places in which people have more contacts.
Now the places in which people have more contacts
are grocery stores.
Up to 25% of the social contacts that we have right now
in this new normal is basically in grocery stores.
So that is very important, because as I'm
going to mention right now, because one
of the problems that we have right now
is that there is a lot of social contacts, actually.
We spend a lot of time in grocery stores,
so we probably have to do something about that.
The final question is whether different groups
of different demographics have been affected differently
by these social distance measures.
You can see here the number of contacts by income group.
So the dark blue is the low income group,
and the light orange is the high income group.
And, of course, before the lock lockdown measures,
the high income group have more social contacts because they go
more to food, to restaurants, to shopping, to museums,
et cetera, and the low incomes stay more close to home,
or they cannot actually afford to go to other places,
so they have less contacts.
But after everything, I mean after the measures were
implemented socially systemically,
all, every group, actually, came back down
to the same number of contacts as any other.
Estoban, we have a question from Umberto
who wants to know, how did you quantify
the personal social contacts from a mobility data set?
Yeah, so what-- because we have a panel in the data,
we have around 600,000 people in the data.
And what we do is to take the proximity network
of that person--
so we count how many people this person
has per day within 25 meters, and this
is the number of contacts.
It's not-- we don't know whether there's an actual contact,
but it's a proxy for the number of contacts
you can have in your every day.
So there is this thing.
I don't remember if I had here, so basically, the idea
is that we measure how many people you
see within a 25-meter radius every day,
and this is your number of contacts.
We have played with the numbers here,
to decrease this number, to increase it,
and the results are basically continuing the same.
There is a follow-up question, and that
is about the absolute number of grocery contacts
versus the relative percentage.
Shouldn't we focus on the absolute numbers?
Yeah, we should focus first--
that is what I was saying, is that it's working.
So the measures are working, the number of social contacts,
on average, is very small.
As you can see here, it's 5.
And this question is, is this working in epidemics?
Is this working to prevent the transmission?
We think it's not.
I mean, the thing is now, because the epidemics is
actually happening, or the mission is happening right now,
even though the number of contacts is very small,
it's happening now in the groceries
because it's the place where most people are
So we should be careful, because there's still transmission.
The transmission is happening in these groceries.
Even though the number of contacts is five only,
this transmission is happening in those places.
So there is a follow-up question about that,
and that is, what can we do about the social contacts
at grocery stores?
OK, yeah, we actually, we had a meeting this morning
with the rest of the group we were discussing this
is that we have preliminary results about what
is the time of the day that people go to the grocery
stores, and I mean, it's unbelievable
that we haven't changed our patterns.
We go on Fridays, Thursday afternoons, Saturdays
and on Saturdays.
So we should probably change our behavior
to prevent being at the same time in the grocery stores.
We go less to the grocery stores,
but we're having a lot of contacts in the grocery stores.
Is the time of--
what is the time of day?
Are people going in the morning, at noon, in the afternoon?
What does the data show?
We just saw what is the number of the number of people
going by time, by weekday.
And we haven't actually gone through the time of the day,
but we probably--
I mean, the people in our [INAUDIBLE] group is
calculating this as we speak.
So we don't have the results for that.
One of the things that we're doing
is trying to flip the model, basically.
So if you had, say, Instacart or Amazon workers who were
recovered-- had antibodies--
and they were delivering food, as opposed
to people going out to collect food,
then you would have infection-resistant people
having the contact.
But that doesn't matter in contacting people
in their residence.
And to do that, you have to have a registry of
recovered people, with people with antibodies.
So one of the things that we're trying to do
is create a registry like that that can be queried.
So it's a little like we do already with tuberculosis,
or various sorts of inoculations or child care workers or food
But it may also require this sort of flipping the model.
As we've seen recently, that means that the Instagram guys
are going to have to get paid more,
and you get a lot more equipment for frontline workers,
But from the point of view of these passive measures, I mean,
We have to actually work on active measures,
like, for example, who is going to be making deliveries
and from the point of view to [INAUDIBLE],,
I think we can work right now, because now the infection
If there is an infection, it's happening
in places like grocery stores.
So we should probably do something about it.
We probably can do that.
I mean, there's probably, you know, about, for example,
some stores are actually opening earlier for elderly people,
like a Whole Foods or something.
You can go to 6 AM in the morning if you're over 65 years
And only from 6:30 to 7:00, they are open to specific groups.
So maybe we can work in these policies
of preventing people piling up, or packing up
at the same time in these stores.
Remember with gasoline rationing,
they took the last digit of your driver's license
and said you can only get gasoline on certain days.
You could imagine doing the same thing with food.
So Alan Edelman has an interesting comment.
His comment is that we need to close down grocery stores where
the public enters and allow ordering online, by phone
and then have curbside pickups.
Don't know how feasible it is to implement this,
but it's an interesting idea.
Yeah, it could be ideal.
I mean, let me just say that when you look at the--
I mean, what other countries are doing in this sense.
I mean, there are some people that cannot afford even
to arrange delivery.
So we need to put in place, for example,
because they don't have a connection,
they don't have any connection.
They don't know-- there are people which are, for example,
a life that cannot actually do these things.
So we need to allow them to go to the grocery deal
or have other systems to allow them.
So maybe delivery is the only option,
but as Sandy was saying, maybe they based around people.
The Instacart people are only the people that have already
gone through their infection.
Yeah, that's right.
So Jasmine, let me wrap up this part
of that saying that yes, social distance is happening
in the New York City area, but the question here is that how
this translates into the epidemics,
into the future of epidemics.
I think we are really--
it's too early to say that.
We can do the simulations that we have.
This is actually what we are doing, because we haven't now
the new contact matrices.
We can extrapolate our simulations
and see what is going to be the impact.
Or, we can wait a little bit more time,
but we don't have time, actually, to wait,
so probably have to look to the hybrid approach
to this process.
The other thing that we are doing right now
is to compare cities.
So Seattle, for example, went through lockdown measures
one week before New York.
And probably, the impact of those policies are already,
it's already seen.
Or other cities, like, for example, there
are states like, for example, New Mexico.
And the other thing is that we want to actually talk about
in the next reports is about how we can actually
do social distance policies within places, not
at the level of the city, but also in places to prevent where
they are received well, contacts that we have in the city
are happening right now, which is [INAUDIBLE]..
And I thank you for your attention,
and open to questions.
I'm going to clap on behalf of everyone.
Thank you very much.
So, let's see.
We have a few more questions.
Please send your follow-up questions on the chats.
So some grocery stores have now started
asking people to remain six feet between each other
and the cashiers.
Are these measures helpful?
Well, in New York, the grocery store aisles are really narrow.
And I live right next to the smallest Trader Joe's
in the world.
And I'll tell you, it's impossible to move around there
and stay six feet away from people unless there is, like,
one person in the whole place.
And on top of that, yeah, it's also very-- it's really hard.
I mean, the aisles, for example, are the same.
So basically, you pass by some people
in the aisle, and then you--
but also, this-- there are people working in models
about contagion by surfaces.
I mean, again, if either you go to the supermarket
and there's another number of people there,
you can get infected by just touching the same thing
that they touched.
So it's not about just being in proximity with people,
but the part that you on sharing the space with them.
Another question is about whether soft distancing
can be modeled as people reducing their contacts
with others rather than completely removing themselves
from the graph.
This-- is this captured in your model,
and what would be the effects?
Yeah, so the model, the social distancing
in the model and the simulations was like people--
self-selected people decided to, I don't know,
not go to restaurants or to their workplaces
and stay home only.
But that's right.
You can-- you can probably-- and that's actually something
that Sandy was saying, is that eventually, we will
have to resume our activity.
So maybe instead of picking a function of people
and deciding that they have to stay home, maybe we
have to select a particular type of activities that we can do
Like, for example, not going to in restaurants
every other week, or staying home one week or the other.
Another question from the audience
is about whether you have any interest in tying
your work to the Kinsa Internet Thermometer data,
so that behaviors that are connected
to the state of health?
Sorry, I think it was broken.
Another question from one of the members of the audience
is about whether there is any interest
in tying this work with Kinsa Internet Thermometer data
so you can correlate behavior with level of health.
Yeah, I've seen the data that they have.
It's a very interesting--
I mean, let me just-- you probably know about this.
Kinsa is actually selling a thermometer,
a [INAUDIBLE] thermometer which is connected to the internet.
So basically every measures of the temperature
goes to these companies so they have around, I
think they have around 5,000 500,000 thermometers
all over the US.
And they have the temperature, so they
can see the temperature in different cities
and the spike, for example, of temperature.
And you can see this.
I have to say that I'm a little bit--
there was an article just at The New York Times about this,
that every time they see a drop in temperature,
they immediately see a drop in their people
that go to hospitals.
And I don't think anyone that is working
right now in the biology of these epidemics
believe that this is so.
So because the fever and the symptoms
develop much, much before you actually
have to go to the hospital.
And actually, there's a lot of people
that battle this disease that don't go to the hospital.
So they're just saying that they see
an immediate thing, like in two days, the drop in temperature
propagates into the number of cases which are detected,
which I don't think is the case.
But it's very interesting to have this data,
because this is the other proxy for the symptoms
of this disease.
One of the things that we're getting interested in
is the effects of all of this on retail business.
So obviously, if you shut everything down,
there's a lot of people out of work.
There's a lot of things that are not happening.
But we are working with several countries, and with MasterCard
to be able to look at, how can you--
when you release the social distancing
and get the maximum regeneration of the economy,
or getting them out of exposure, this
has to do with sort of optimizing recovery.
And we also are working with a couple of countries
to do incentives to get people to support recovery
efforts, in other words, to patronize things that
are slowly being reopened, but to do it in a way that
But there's a lot of space to explore there.
But that-- once we get through sort of the first one
or two of these pandemic humps, it's
going to be the question of the day.
How do you bring things back without bringing back
In fact, this is a good transition,
because we have a number of questions
about the second peak.
And in particular, people would like to know,
did the intuition behind the appearance of the second peak,
and whether social distancing should be practiced
after the disappearance of the first peak
in order to avoid the creation of the second peak?
All right, so I think that the second peak is--
well, at least what we see in the simulations
is that there's always going to be a second peak,
because pretty much like we're having with the flu,
although the flu is not always.
We always have a second peak, because even though we
stop our implementing of social distancing,
there's always going to be a group of people that are
actually interacting with you.
Emergencies, essential businesses, things like that.
So yeah, there's this report by the Imperial College
in which they were assuming that you can implement, for example,
social distance on and off, like, for example, one week on
and then one week off, et cetera.
But I think what is going to happen is that probably we
have to keep working on our social distancing
for a long time until we have these pharmaceutical measures.
And we will have the vaccine.
So I don't know how it's going to be,
how our societies look like the next fall,
but probably we are going to have less people going
Some people are going to stay home.
I mean, their estimation that some of us
are probably going to stay home working
because we can afford it.
Some people cannot.
We probably have to do this to prevent the magnitude
of the second peak.
But one thing that could happen, there
are already antibody tests that are being distributed.
There's a number of them.
Some of them very sophisticated, which means you
can certify that a particular group of people
are not likely to be infected.
We'd know exactly how much so you can imagine reorganizing
things so that it's only safe groups that meet physically.
You can also imagine other sorts of strategies
where you don't open centralized things,
but you do ferment some gatherings
in segmented, separated, distributed things.
Because the disease will spread much more
slowly if there are no hubs.
We are coming close to the top of the hour.
I wonder if both of you would be willing to summarize
the key recommendations and takeaways for the audience?
Well, I mean, it-- as I was saying,
we are still working on having the model--
I mean, better model to produce this.
But one of the things is that just passive measures are not
going to work unless we have a pharmaceutical solution
There's always going to be a second peak,
so we have to be ready for that.
And we have to work on active measures.
We have to work on contact tracing.
We have to work on localized target quarantine, et cetera,
to prevent the magnitude and the time of the second peak.
Right now, the re-infection time that people are actually--
the CDC is working with is 50 weeks.
So that means that by next January,
we are going to have another one no matter how many people are
infected and go through the disease.
So the second biggest coming is going to come,
so we have to work on better strategies
to actually work on the second peak
and to exit this first one in a way
that the society recovers and economy gets
back to what it was before.
Regarding the images that we have,
it's actually nice to see, for example, that in--
there has been number of people doing the same kind of analysis
in other countries.
And I tell you that the number--
the numbers that we get for the New York City, for example,
are much better than what you get in other countries,
like, for example, Italy.
The social distance, for example, in Italy,
has been reduced too, but is a much smaller magnitude
than we have from somebody in New York City.
So also, this comparison can tell us
something about how we're gonna implement
these policies in different countries,
over time, et cetera.
One of the other things is that we
need to think about how we're going to do this long-term,
because it's going to happen again,
not just the second peak, but other things.
And I think that the idea of having a registry of people who
have antibodies is key and then reorganizing things
so that people who are in contact
with large numbers of people are preferentially people
with antibodies is a, for instance,
really important thing to do.
Another thing is, is that it turns out that different--
there's a whole--
there's millions of different COVID viruses.
Some of them may reduce relative immunity for other ones.
That's true of many of the COVIDs.
So there's research going on now to ask, can you do--
like people did with smallpox, which is--
smallpox, you've got cow pox--
which wasn't a bad disease but gave you partial immunity
So there's a whole bunch of different things
that depend, in this case, on very large data analytics where
you're looking at all of the immune responses in all
the different situations and figuring out
what combination that prevents further spread or minimizes.
So there's a lot of measurements and analytics
that need to be done here, pretty urgent.
It seems that you have began, and you
have a great opportunity in front of you
to learn so much more about how social interactions combine
with policies to help prevent disasters.
So Esteban, Sandy, thank you so much for sharing.
And we wish you continued great success.
Looking forward to the next round of results.