 Hi Yeah, perfectly. Thank you great. Can you hear us? Absolutely perfectly. Yeah, great so I'll give you the floor for the presentation and everything and then there will be a Q&A moment after your presentation and I will be very strict with the time which is 40 minutes, so you have 40 minutes and I will jump in if you like run out of time So the floor is yours. Bye Great, great. Thank you. Thank you so much and it's fabulous to be here. I'm gonna share my screen here what I what I want to talk about today is Really two separate projects one is Just an idea been playing with a long time when you think about something like inequality There's so many models out there and it almost makes I think a lot of sense to look at it through many many lenses So I want to give spend about 15 minutes just walking through the many different ways That economists and social sciences have looked at inequality Then what I want to do is I want to construct a very simple model of structural inequality To show really just two points one that looking at means can be really misleading And the other is is that oftentimes we have these interventions and there's a whole science of implementation science Where we try something and it works and then we take it someplace else and it doesn't work And also in the long term it doesn't work And I think that's because these systems are incredibly robust and they sort of pull themselves back So I want to construct a very simple model of structural inequality based primarily in the work of Rob Samson out of Harvard Okay, so quick not that people need to know this because it's a reminder, right? We have kind of just enormous wealth inequality in the United States And whether you look across the US or Europe or Western Europe You see that the top 10% of earners just earn a large percentage in that money So why? What I want to do is I want to walk through 10 models very very quickly Just to kind of let them wash over you to give you a sense of all the different ways and causes that we think exist for this level of Inequality, so the first model is just a simple econophysics model And this says that random exchanges are going to produce skewed income distributions So these kind of 10 models up on top here The first one is this if we just imagine that we create a system with a bunch of people who randomly trade with one another What we won't get is a normal distribution What we will get is a Boltzmann distribution And so what these models look like and this is one by Gregolesco and Jacvenko You just assume that there's a total amount of money in the economy People bump into each other one person wins a dollar one person loses a dollar and what you get from that Isn't a normal distribution and I think this is incredibly important to know economists tend to Dismiss this sort of model But I actually think of it as profoundly important because it essentially says the fundamental physics of exchange lead to Substantial levels of inequality Okay Second model and this would be kind of like this standard econ model is that there's really kind of a race between technology and labor So what happens is a new technology comes along check GPT being an example It's going to be labor saving what that's going to do is that's going to increase the relative value of educated workers typically And so what you see over time and this is a time series for us data from 1963 to 2010 But you see the kind of like the value of going to graduate school goes up and down the value being a college graduate goes up and down And these are all due to this kind of like race between Technological advances and then people getting educated and so the models here are really straightforward You've got a constant a which comes from a standard solo growth model Then you've got coefficients on capital So k is capital then there's returns to high skill and low skill labor these beta and this gamma And there's just a simple technological advantage parameter Depending on the state of current technology And so as that shifts as beta gets bigger more people move into high skilled jobs Inequality goes down beta gets bigger again inequality goes up And it's this is constant race between people sort of increasing their education levels and changes in technology now third model Which is that well, that's that model is kind of static And if you look at who really makes a lot of money oftentimes, it's people who create new industries, you know, you think of the Jeff Bezos you think of bill gates, right? And so what you could think of is ability translating into a rate of learning And then what's happening is ability variation is explaining a long tail So if you look at any particular job one of the things is kind of amazing This is just airline pilots that we've seen a skew in the distribution So there's you know, so there's inequality at the national level, but there's also inequality among college professors airline pilots Accountants lawyers every single profession has had this increase in inequality Now these are models by a professor Jones out of stanford You used to imagine each person has an ability a But instead of thinking of that as like high skill and low skill like the previous model What you could think of is that like my in a is like my ability to learn And so my income in your t equals some constant, which is the base your income times one plus a raised to the power t So what happens is is if I just kind of learn more and continue to grow My i'm going to grow at a rate a and if somebody else has a higher rate of learning They're going to grow faster And so what we're seeing is just the compounding Of people who sort of continue to accumulate capital And again, there's strong empirical support for the fact that the people who Make a lot of money are people who continue to sort of grow and learn And one way to think about this is the rule of 72 Which is something that accountants use all the time if your income grows at a rate r it doubles in 72 over r years So if two people initially were earning 50 000 dollars And you looked 36 years later a person who grew it a 2 increase in making 100 000 dollars The person who grew it a 6 interest would be making 400 000 dollars So this is just kind of a simple doubling sort of argument based on talent Now a fourth model that's similar to that and comes more from biology and and I think you've seen somebody Louise bettencourt I think it's probably sewn some stuff like this Is that social status correlates with higher growth in this human capital? So that previous model just sort of says some people have high a's Some people have lower a's and that's what explains differences But what you can imagine is think of this like a biological growth model So I've got a total amount of energy and that energy can go into two things It can go into maintenance or it can go into growth And so the growth parameter here g is like the a from the previous model Now what's causing the inequality here is more structural in the sense that if I have to spend a lot of time on maintenance I can't spend that much time on growth So let's imagine two people one person's born without much social status And so they've got to pay for their car. They've got to pay for rent. Maybe they're living a long way from downtown So their maintenance costs are really really high Someone else has a lot of family support high social status So their parents buy them a car their parents buy them an apartment their parents You know pay for somebody to give them child care, right? So their maintenance costs are a lot lower Because their maintenance costs are lower their growth rate can be six Whereas the other person's growth rate is two Even though right they've got the same fundamental energy level So what this is doing is sort of showing how Inequality can manifest itself over generations Because once you've accumulated status and wealth you lower your maintenance cost and that allows your growth rate to be higher And again articles on this all the time a secret of many urban 20-somethings Is their parents help with the rent allowing them to live closer to work lowering their maintenance costs Okay model five now, let's go Sociology right so those have been kind of like physics models economics models Let's go to a sociological model and this one actually has Maybe the most empirical support out of all of this It's just a standard preferential attachment model social effects lead to big winners So a socially constructed superstar model. This is a standard preferential attachment attachment model I can buy from one product I can buy one product or another I'm more likely to buy things that other people buy right? And so this is the famous music lab experiment by sargonica and watts and others If people don't see the songs that other people download This is kind of the the songs along the axis the horizontal axis and the number of downloads on the vertical axis This if we plotted it differently we look pretty much like a normal distribution Some songs do a little bit better some songs do a little bit worse But once I know which songs other people Are listening to then we get this long tail distribution Now this is an inverted long tail of the High bars that lots of sales And so what you see is that people buy things in sort of a social setting and we're influenced by other people Then you're going to get winners and losers and you're going to get a winner take all economy I mean the story that economists like to tell in this sort of context is no one other than economist Goes to a bar and says give me your second best champagne or give me your second best steak, right? We tend to sort of want the best we want the thing that other people want And that leads to sort of extra marginal returns Stories so far have all been kind of like micro based and haven't had any sort of You know sort of elite that are manipulating a system Um model six is a model, you know put forth by Joe Stiglitz Nobel prize winning economist and it's in his book on inequality and it's a spatial voting model And the idea here is people at the high end of the distribution Do so they they they continue to make huge wages and salaries By manipulating the system and so If you look at the ratio of the ceo pay to average worker pay in the united states It's 400 times which is much higher in other countries. It's lower and when you saw the data before the u.s had an incredibly high Level of income inequality compared to other countries, especially in the top 1 percent This is this is Stiglitz argument and here's kind of quite simply how it works Suppose we have three people on a committee deciding how much the ceo should get paid as a proportion of the average worker And there's three people on the board Right, the board has two ceo's and then it has x which is an external consultant So the external consultant says look you should probably make about 35 times what the average worker pays But the two ceo's who also want their salaries to be high. They're like no no no no You should be paid 250 times 300 times what people get paid with the average worker gets paid So when we vote on this a spatial voting model will say well the median voter decides So the median voter here is ceo number one And so ceo number one varies the day And the ceo of this company gets paid 250 times the average worker Now let's go down to the bottom graph which should represent a country like germany or the scandinavian country where there's workers on the board Now there's three people on the board. There's ceo one. There's which is another ceo from a different country company There's x which is the outside consultant the expert And then there's w the worker when I replace the ceo with the worker What I do is I now make the outside expert we can assume is somewhat objective As the median voter and so now the ceo only gets ceo only gets 35 times the average pay of a worker So stiglet's argument is when we think of a lot of wages are not set in markets And they're set by committees right university salaries are set by committees Then we compete with all the universities who set things by committees And that whole system can get corrupted by people within the inside and amplify salaries for elites This is certainly been true of university presidents in the united states When I first became professor 30 years ago university professors University presidents made a couple hundred thousand dollars. Now they make several million dollars Okay model seven going back now going back to sociology This one has a lot of empirical support. I mean if you if you do a horse race between them About five years ago this one when I checked this last this one kind of wins And this is a simple assortative mating argument highly educated people are now much more likely to marry Which increases inequality So if you look at educational assortative mating whether you're looking in the us or in europe What you see is the share of all couples where both have a college degree Is going way way up now. This is an incredibly simple model if you look at this I kind of love this word I presented to my undergrads It's like income of the family equals the income of one spouse plus the income of the other spouse Now this has done a kind of a heteronormative model where it's a man and a woman But in any family right if you've got if you've got a married couple the income is the sum of those incomes If highly educated people are more likely to hire marry highly educated people. You're just summing to higher incomes Okay model eight The famous piquetti model right some of the highest incomes comes from capital rents and always will And what you get is that the rents on capital are higher than the growth rate Right, so this is a very deep model. We turn on capital is higher than the growth rate And if you look at capital income Over time from 19 he picked piquetti in his book goes back to the 17th century But if you just look at sort of capital gains and capital income, there are a huge percentage half Of the top 0.1 percent income share And so this is just once you have wealth you continue to accumulate wealth And so the the more sophisticated model here is like the growth in wealth depends on the return on capital Minus the tax rate minus the consumption rate. So if you have enough money that you know, it's not getting taxed and consumed away You're going to grow faster than the growth rate of the economy and you're going to continue to Um accumulate money Model nine and now we're getting kind of more into the structural inequality stuff that I'm going to build to so slow down a tiny bit These are an income dynamics model. So what you think is that incomes persist across families Not only due to wealth But also due to transfers of skills So this is a wonderful chart by gregory clark from his book the sun also rises Or if you look at the top Look at names and their probability of attending harvard So if he names like ellen or peter simon kathar and elizabeth anglosaxon names Your three three fold is likely to go to oxford If you're limiting on then you have an irish name like jade or page Or shannon or shade your odds are like 150th is likely and so this is getting it You can't make an argument that people who have of irish descent Are less intelligent than people of british descent But you can make an argument that social capital is getting passed on in such a way That some people have a much higher chance of attending these elite schools than others And so these simple models that are going on is that you're just it's there These are written typically as just markoff models where there's parents and there's children And these sort of soft skills are getting passed from one parent to the other And so you can just kind of write this giant markoff model And then you get that these systems end up having this really really slow transition In terms of social mobility Last model and then i'm going to get into something kind of in more detail Are these persistent equality models? So this is the segregation creates poverty traps for economic sociological and psychological reasons If you look in the united states and this isn't by race. This is just by income. This is washington dc Low income people are concentrated in one area Higher income people are concentrated in another area and middle income people kind of form a belt between them If you look at the share of low income households Residing in majority low income census tracks In our major cities new york philadelphia los angeles, right? It's 30 percent Right If you look it's actually the transitions Out of poverty are actually higher in places like atlanta boston san francisco, which doesn't show up on this chart Because people who live in low income areas are aren't surrounded by other low income people And what and there's evidence that in terms of like just even like sort of environmental stress Stress from violence quality of schools Accessed opportunity all those things improve if you're not surrounded by other low income people So these simple models and this is one i'm going to go into kind of in more detail in a second Your income depends on kind of your ability your education and these spillovers and these spillover effects are Everything from like, you know health to peer effects to crime to stress to environmental effects So i'm not actually happy with the word spillover Even this is what the economists tend to use you might think instead of these as some as more systemic effects so quick summary and In my book the model thinker in that the book is basically a compendium of a whole bunch of models The last chapter looks at Income inequality also looks at coven some other stuff and walks through you can you can get a recapitulation of all these arguments But but the point I want to get across is when you look at something like income inequality It's so easy for someone to say this is the reason And in reality, there's many reasons right there's naturally just a skew There's just the basic economics of technology and labor There is increasing returns to ability There is this structural inequality that comes in because of maintenance of growth There are superstar effects. The system is manipulated A sort of mating especially by education is causing an increase in inequality It is the fact that wealth accumulates faster than the growth rate of the economy It's also true that there's these social things passed on from families Right that enable you to sort of have these soft skills to do better Right as Gregory kark says in his book the sun also rises And there are these education and social forces and persistent inequality these poverty traps that don't let people get out So how do we think about modeling there being multiple causes and I want to just get something to go back to that last model And I want to take this some of this work by rob samson where he says look there's Neighborhood effects. There's these higher artifacts And there's individual actors operating within these different sort of structures And they all kind of feed on one another But what I want to do is I want to go beyond just having some sort of picture and construct kind of A model not as sophisticated as this but that kind of has these parts. So this is a graph I'm really fond of it's by the force side group out of england And it just shows all the causes of obesity, you know, I'll dial in a bit All those different colors are really different academic disciplines And so some of the things we've heard Earlier in this conference is that we can't really just be inside our silos Because if you're in if you look this whole graph all these colors are interconnected So you can't carve nature at its joints and solve each one of those and hope to solve the system Because this entire system is kind of feeding back on itself So let me sort of show how that can work In a model so I'm going to construct an incredibly simple model each of these dots is a person And I want to think of two broad dimensions of how well some is someone is doing One I want to think of socioeconomic The other want to think of his health and I want to think of health broadly defined So you can think of this as physical health as well as mental health So what I've got is I've got just a scatter shot to begin with right So there's a distribution on each of these two parameters and there's a mean there Now one could have separable additive thinking so I could think of things like health status Where there's lead exposure diet exercise stress disease those sorts of things And I could think of socioeconomic status depending on school quality family infrastructure local economy and crime and community policing policies, right And what I could do which people have done now for the last 60 years Is I could run a regression on health as a function of all these different things And what I could do then is I could say well if I put somebody in a poor community What's going to happen is that distribution is just going to kind of slide down Right in this direction And I'm going to end up With a community in which people are kind of poor health relatively speaking and poor socioeconomic status And then what I think and this is completely well in tension If I'm a health economist I work in community oriented health Look if I get the lead out of the water if I improve people's diets If I help people exercise and get rid of stress I'm going to lift up This society But the separable thinking here works out of the assumption that like this isn't going to have that much of a socioeconomic effect necessarily immediately because I'm just I'm only improving on the health dimension And then I think what I also want to do is I want to kind of a full slate of activities I want to also improve school quality infrastructure local economy crime and community policing And so now I'm going to move in both directions So I can move this way and then I can move this way and everything's better in my policy works The challenge here is that we've got like 50 years of showing that this hasn't worked right. We haven't been successful And the reason why is this is a system. So let me let me start out. I want to construct the system kind of slowly So let's suppose I start with this is just kind of a loop gif here. I'm starting with That scatter Scott distribution of socioeconomic status and health status If it's the case that health status affects socioeconomic status, what that does is that pivots The distribution right because if I have high health I'm up here in high health Then my socioeconomic status gets better if I'm poor health my socioeconomic status gets worse So it's just pushing the graph Like this or maybe it's like this depending on how I'm being reflected in zoom If in turn socioeconomic status affects health status That's going to push the people are high socioeconomic status into higher health people are low socioeconomic status under lower health and that's pivoting the graph this way Well, when I put both of those in They amplify each other. So one is pushing like this one is pushing like that and the system very quickly moves sideways So when you look at this and what's important to look at is look at the two green lines Those are tracking the trajectories of two random week shows and members of the population What you see is that I'm getting a narrowing. So I'm getting a huge correlation incredibly strong correlation between socioeconomic well-being And health well-being right that's I'm almost getting a 45 degree line I'm also seeing one trajectory where the person who starts a little bit ahead on each Is actually doing better the person who's starting out a little bit worse on each is doing worse and the mean doesn't change The blue dot is not moving Right. So I've set this thing up. So nothing is happening to the mean but the distribution is changing like crazy And when we look at this distribution, this is a this is actually just showing increases in health disparities and increases in socioeconomic status disparities that are mutually self-reinforcing But we also know that there's kind of self-reinforcing loops within each one of these So let's suppose socioeconomic status Kind of has a positive feedback with socioeconomic status or negative feedback with socioeconomic status So if I've got above average socioeconomic status, I'm going to do better That's like that model we had with the a being positive and if I'm below I'm going to do worse So now look at the two green lines the person who's got above social high above mean socioeconomic status Even though they've got poor health, they're moving up The person who's got very good health but low socioeconomic status. They're moving down. So what's happening here That positive feedback or that self-reinforcement is pulling the socioeconomic status line apart Which is another force for increasing inequality. So previously we saw these If these two things self-reinforced one another that's going to cause the whole system to just kind of like skew Toward a 45 degree line and increase in equality positive feedbacks within each one of these is going to push things to the side Now one of the things we talked about or I talked about You listened about I guess before is there's huge peer effects in these models as well Right. So if I'm surrounded by people who are doing well, I'm more likely to do well And if I'm surrounded by people who don't do well, I'm less likely to do well So if I throw a peer effect in just on the socioeconomic status not in the health status And I assume that like If someone's close to me, I become somewhat more like them What that's going to do is that's going to create these bundles and it creates in this particular case Just the way I've set this up. It turns out I get most of the time I get like three groups I get kind of a middle group a high group and a low group But notice again the mean it bumps around a little bit, but it doesn't change Because people are kind of averaging people near them So the effect of peer effects isn't so much to increase inequality but to create clusters of inequality But when I throw in the peer effect Plus the socioeconomic status loop watch what happens, right? I'm getting the the clusters again But now the one in the middle just kind of stays in the middle and the one at the bottom Moves to the left and the one at the top Moves to the right so the peer effect plus the socioeconomic status loop gives us a particular form of increasing inequality Let's go to health. So this is a lead, right? And if we decrease lead, you think, okay We should improve situations in the society even if we if lead is in the water things should get worse So let's suppose there's a negative environmental effect Now again, if I think of the world in terms of You know just things being kind of linear Then what's going to happen is when I have this negative environmental effect The health of everyone in that community Is going to get worse, but this is notice I've stripped away all of the other stuff And so the mean here is going down But realize that by trying to carve nature to its joints We have to think this just can't be the right way to think about this, right? So what if I throw in we'll leave out the peer effects for a second The health status affecting socioeconomic status socioeconomic status affecting health status positive feedbacks and health status Now when I throw in that environmental effect, right? I'm getting all I'm getting the The positive feedback loops between the two getting things on the 45 degree line I've got the environmental effect pulling things down Suddenly get the whole system heading towards the corner, right? So before I thought of health effect as Just a health effect as terrible as that may be It's just kind of pulling the whole system down relatively slowly But when I throw in the feedbacks in the system It's pulling the system kind of it's almost like looping it down into the corner And so the environmental effect not only has an effect on health status It has an effect on socioeconomic status Which in turn has an effect on health status and then we get in this kind of vicious loop And here we see that mean Just racing towards the left lower left hand corner Whereas when we had a linear view of things we just see that mean heading straight down Okay, now if I add in Right the peer effect Then what happens is I'm still getting this clustering right but The environmental effect has the effect of taking two of those three clusters and pulling those towards the corner But yet still allowing one cluster to kind of make it out The guy purposely shows this particular green dot in this example because this green dot is someone who starts out above average socioeconomic health and above average And socioeconomic status and above average health But yet because they make a set of friends that are a little bit below They start heading to the left and then that environmental effect pulls them down And they do worse One of the things that I'm Really want to sort of drive home and looking at these things is each one of these dots You can think of as a person who has a story that is a narrative And so it's very easy to look at a picture like this and follow a red dot That kind of makes its way out. In fact, there's a lot of red dots that make their way out So even though this is a system that starts out kind of in an okay place There's a negative environmental effect through peer effects and through positive loops There's a bunch of red dots that do really really well And so that allows us to tell stories about people who make it and say no, look, it's just hard work It's education. It's finding the right friends. You can make it. That's true You can make it but like 10 percent of the people can make it Most of the people are being sucked into this vortex into the bottom left hand corner So you can think okay, let's have a health intervention. Let's come in, you know, the system isn't doing well So let's intervene on health. The thing is if the health intervention Even if it more than offsets the environmental effect Once we're down in that corner The low health status is feeding into the economic status and into the health status And we're just not going to make our way out of it Right and you can think what if we do both In intervention for health and an intervention on socioeconomic status If it's not high enough if it's not strong enough The system is still I mean here you see the system kind of stagnates a little bit We're kind of fighting the system a little bit and we are seeing a couple people make it out But generally now here all we're doing is consolidating the entire system down in the lower left hand corner So what I want to make is that when we look at these systems, right We want to think of there being all these different effects that are in play And even though this is a very simple logic and it won't be something that's surprising to anyone in this room It's I think I found this simple model is probably the best way to communicate this to people You want to think of these the health status the environmental status the socioeconomic forces as Having this gravitational pull for the entire society, right? And so this red dot is again kind of representing the mean and I want to be thinking of us in terms of systems of agents But when I have some sort of intervention, I'm pushing things up the basin The problem is the ball sort of falls back, right? So do this fancy simulation one more time, right? I start here, but then the system just kind of pulls back And so the way to think about this Is that if I have a policy intervention in this sort of system, if I were to think of things as being separable then The larger the policy intervention the larger the effect, right? And this is the kind of thing we do in empirical research We say oh, we had this we did this simulation We did this intervention we had this effect If it's systemic Even though we might see it in effect initially It's just going to get pulled right back down to the bottom of the basin And what you see is there's going to be thresholds for the size of effects you need So if you think of these things as complex adaptive systems, I'm sounding a lot like Donatello Meadows here Right with these positive feedback loops Right and these kind of also these sort of direct feedback these direct effects that are pulling things down You're going to need policy interventions that have sufficient magnitude Kind of get out of that basin So the goal here is escape velocity So what I did again in this very simple model is I Weaked The health intervention and the socioeconomic status intervention to make them just high enough Just tie it up so that the system could make its way out But if you watch the pattern on this it's kind of funny because you get this kind of coalescing It doesn't seem to work that well at the beginning It's it's just kind of getting there and it kind of coalesces and you're kind of pulling everybody up a bit And then you get this kind of pathway Out so I continued to run this you'd see the whole system Go but if you watch like the green dot it's just moving around not a lot is happening But the thing is generating people on this 45 degree line It's slowly starting to go and then once they get above a sort of combined threshold on those things There's this kind of pathway out of poverty What I like about this particular simulation of how to go back to the rob samson story This is a metaphor that's used all the time right that there's pathways out of poverty And even if I go back to the simulations in which only a handful of people make it out Those people are still following similar sorts of pathways I want to just in kind of closing this part out I want to be really clear that This is still super. This is more complex than we typically think about policy But this is still incredibly simple compared to just that obesity graph that I showed at the beginning And that's only dealing with health status And part of the reason I gave the 10 models is when you think of the socioeconomic status and what's holding people back socioeconomically There's 10 or 11 or 15 forces going on in there And so we really want to think of is There being many many forces on health many many forces on the environment many many forces on community Many many forces happening in the in economics And those things create this gravity to a system and we've got to think about then how do we get people out So quick takeaway and systems thinking and then some summaries that are hopefully a lot of time for questions so the first is Watching these models you want to look at distributions of means because there's going to be people who make it out There's going to be people who do poorly. There's going to be clustering The implications of effects are not additive, right? We saw and this is just kind of a simple building block kind of model as I kept adding more things Weird things were happening to the distribution even though the mean isn't changing much And it does appear again not a new idea here But it's going to take a sustained multiple focus approach to get the system out pulling one lever People talk about lever points in these systems all the point all the time My colleague John Holland would have said that it's more like levers points I mean you want to many you probably need many interventions to get these things out um So last slide There's a ton of causes of inequality and therefore there's many models of inequality and it's really useful to look across all of those And think about how can I embed all of those? Inside a system and then think about how we get out of that system Because the many causes aren't just it's not a horse race between those causes like mine is a bigger r squared than yours are You're r squared What every one of those models is true to some extent, right? There's some amount of the variation They're explaining there's some strength of that cause and there's ways in which that cause is interacting with other causes Like we see in something like that obesity graph The combination of all those effects is to create a system that unfortunately is really really robust intervention and Rather than see the system in equilibrium at the moment I think I see the system as one in which you know Inequality is kind of increasing and so if you think about we're heading in a path that's increasing The size of those interventions may have to be even larger Than the ones I was showing you In the graphs. All right, so let me stop there and open it up for questions Thank you very much. You were perfectly in time. You have five minutes left. So it's uh, it's gonna be Used for cute questions. I already see some hands raised and I'm gonna go down there and then like Get back to the front And I apologize. I couldn't be we have a graduate student strike and so we're uh, all of us have had uh University's in a bit of turmoil here Thank you so much for your presentation. Um, I really really found it interesting. Um, I totally agree that I think There are a lot. There are many manufacturers which cause inequality and they tend to work in these kind of continuous feedback loops and Um That you know, the aim is really to disrupt. I think that feedback loop But some of robert samson's work actually shows that even at a structural level a very high level you might have Major change sometimes at the neighborhood scale neighborhoods persist in terms of uh, where In looking at their trajectories and in terms of um, remaining with uh, within you know in within certain poverty levels so I was wondering if it's maybe not necessarily the size of the disruption but rather thinking about it in terms of scale and thinking about policies in terms of um Yeah, I just wanted your thoughts on that and in terms of like multi-scaler interventions. This is such a great question. So, um Robert and I've been talking to people in Cincinnati about this in this project and The way I've been thinking that it's almost like a rectangle, right? There's kind of like That if you scale on this x and magnitude on this x right like if you go smaller, right? It can work the question is, you know, what can you draw a box around And still have it work, right? So like could we go into one like you want to go into Cincinnati? What's the minimal size of a neighborhood that you could go into and actually have a positive effect? That's a really interesting kind of empirical and theoretical question to ask right because you couldn't We know you can't the household is too small Right and the entire since city of Cincinnati might be just like Too big in the sense that like, you know, you need just things need to be targeted to specific communities And I think you know some of the stuff that louis shows bet and court, you know, they do this much more micro level data of You know what's going on in cities and who makes it out where and who doesn't Is the kind of thing we're trying to look at to think about okay, where could you make? The right sort of intervention, but the other thing that I think is really important And we're using some some cool platforms like one called crowdsmart where we're having conversations with people in the community Is trying to get conversations between people who are ex felons and people who drop down and people who are doing well and people who are not to figure out What is each person's you know each little dot in that thing? What is each person's path like what's preventing them? From being successful, but no, but you're absolutely right. I think this has to be play space and and strategically and also I think the The set of seven interventions have to be done in this neighborhood are probably very likely different than the nine interventions That have to be done in that neighborhood, but it's a it's a really interesting question Where you can draw those boundaries and I think often if you're politically constrained Thank you, right Yeah Thank you for the presentation I just want to ask for the inequality models that we have presented Uh, which models should be used when like in other ways when when is the econophysics model a bit often the other models And then secondly being new to the models. I also want to understand like which datasets are used to analyze such models Yeah, so this is a great I mean this is a really great question and when I was when I was writing the the many model book I mean one of the things that I Was just trying to start with is a very sort of simple observation Which is that if you look at sort of the total number of variables we have to measure inequality or anything Um is enormous and so the entropy of that entire system that we're trying to explain is huge Any model that we can understand is going to be lower entropy. It's kind of less information content and so what you want to have is You want to think of those models as kind of overlapping in all sorts of ways So, you know, I referred to play to carving nature to its joints Instead we want to think of those models as not carving nature to its joints like They're kind of overlapping and capturing different things And so the econophysics model is literally just looking at things like, you know, the distribution of income, right? It's sort of the mating is looking at just kind of like, you know, who is marrying whom is a function of education level, right? So what you want to think about then is Is in some sense constructing ensembles of models That are looking at sort of different dimensions. Now, there's a here's here's what I'm Optimistic, but there's a lot of I think fascinating work to be done If I construct ensembles of models, let's take a random forest sort of approach looking at different datasets And if they look at these different If they're looking at different sort of variables that are simple decision trees We have nice theorems and these like boosting algorithms and stuff that show these ensembles can predict Really really well even though each tree is simple But that's prediction What we want to do here is we want to design systems We want to come up with interventions and so the question is is there some sort of analog of ensemble theory right for design And for intervention and so sort of getting back to your question in terms of which datasets do you look at in some sense, I think you want to think about What's all the data we currently have and then I think, you know, as my mom would have said, you know People like luisa doing kind of like the lord's work and creating new datasets Right because if we have more datasets and then we have more models looking at different parts of those datasets Then I think we can, you know Get something like an ensemble theory of intervention But you're right. Each one of those is looking at, you know Is looking at very very different sets of things. So like Stiglitz stuff is looking at kind of like To what extent is your pay insulate? Let's look at the Difference people by whether their pay is insulated from the market or not And then let's look at kind of the increases in their pay, right? So each one of these is kind of You know think of a house with many windows. Each one has got the shutters down and a whole bunch of those windows Thank you. There was an extra Anyone else wanted to ask a question? No, just one. Okay. So I will give you the floor right now Yes, hi, thank you. Thank you for summarizing all these these models and effects So from the pictures you showed it I have the impression that The way you are modeling mathematically Is somehow assuming some linear feedback or some linear relationship between these variables And as a consequence if you want to achieve twice The outcome you need twice the intervention and I'm wondering and understand it's a simplification in this setting But I'm wondering in which extent this is A good modeling approach or whether there are not non-linear effects that could maybe Hint on on interventions that can have a non-linear consequence where we could maybe through negative loops or non-linear Instabilities by four occasions we one could not reverse Tendencies with very little interventions if if any of these effects appear in more sophisticated models if that is relevant at all I think this is a This is a great point, right? And I what what I was trying to do with this work is, you know Rob and I have been trying to go out separately but also occasionally together and just get people to try and understand systems effects and my Impression was is that Equations aren't a good way to do that, right? Nor are nor are just pictures with cogs. And so what I tried to do is just I wanted to For fun just trying to create some simple visualizations that people could understand Kind of like how systems effects aggregated and how they made and how they created robustness To your point if I made if I were able to have much more accurate data and fit a model Better in terms of like, you know, what's really driving these things Then there is the possibility of Identifying lever points. So a really good example is there's a ted talk involving the There's a famous complexity graph, you know, one of the spaghetti graphs on afghanistan Like where it shows all the things Interconnected in afghanistan looks a lot like the obesity graph and this physicist basically said, well, look Let's look at the things that we actually can change Like some of those things you can't change at all And then you can look at sort of which two or three places what I have to go and intervene in In order to have an effect relatedly like john miller has this wonderful paper called ants active non-linear testing Where he takes the world three model of population growth and says, okay, where would I have to intervene? In order to, you know, have the largest effect. So you're absolutely right if one could construct the Better model a richer model And I think again you and this is I mean it's a wonderful question because it's getting to my question of ensembles of design So let's I think this is a fascinating open question Should I construct one big model that's perfectly accurate and then look for As accurate as I can get and then look for sort of like can I do some sort of like, you know Intervention based on these non-linear terms or should I be constructing ensembles of models that kind of overlap? I love the the metaphor of kind of like a markoff blanket where I'm kind of like I've got a blanket covering different parts and I'm trying to Cut out parts that may be made a little bit less, but I've still got enough overlap And then looking at interventions that seem to be working across a variety of those models I think that's an open theoretical question. We've done because ai's been so focused on prediction We know a lot more about using ensembles to make good predictions my gut And and also all the science based on predictions suggest there's got to be a parallel ensemble theory of design and an ensemble theory of Intervention And I think that you know people have you know people like yourself who training in physics and systems Are probably going to be able to go and lead that and it's exciting right because I think the possibility to Come up with really useful policies is very very high and much higher Then if I'm running a regression and looking what I call like big coefficient thinking in my book I think big coefficient thinking is the wrong way to go when you have a system by that I mean looking for the you know running regression looking where the big coefficient is and spending your money there All right, let's thanks. You're right. You want to look for the big levers Okay, so our next speaker is highlon who who is a neuroscientist? And is going to be telling us about social hierarchies. I think so quite a different set of data and hopefully