Using Spatial Analytics and GIS to Study ‘Bespoke Neighborhoods’

[MUSIC PLAYING][Using GIS and Spatial Analytics to Study "BespokeNeighborhoods"]

JOHN OSTH: John Osth is my name.I'm a geographer.[John Osth][Senior Lecturer, Social & Economic Geography,Uppsala University]I'm based at Uppsala University in Sweden.What I'm doing is pretty much spatial analysis and GIS.I'm very deeply rooted in quantitative geographyand the relationship of space and place.[How did you become interested in spatial analysis and GIS?]

JOHN OSTH [continued]: But I've always been interested in maps and stuff.So geography, we've always been there.Actually, once upon a time, I usedto be an upper secondary school teacher,and I went for the geography part of it,and I found it interesting enough to actually goto the university.So it's a long journey through school and into the universityworld.[Tell us about your "Bespoke Neighborhood" project]

JOHN OSTH [continued]: It's an interesting one, because it'srather different from what you usually thinkwhen you've got sort of areas.If you're going for traditional, sortof the ways we depict geographies,it's usually things like [INAUDIBLE]or a [INAUDIBLE] or an area.Or it could be, I don't know, block or somethinglike that in another context and country.And the individual is to some extent in that area.

JOHN OSTH [continued]: You can say that's the bespoke neighborhood, the neighborhoodof themselves.The bespoke in this sense literally meaningwhere they belong, what roots they've got in the area.But when you have sort of a bespoke neighborhoodand you realize the individual is livingat the corner of the block and really hasgot more emotion to the next one, so to some other area,these kinds of rigid, administrative areas

JOHN OSTH [continued]: are doing little to actually help us understandwhat a neighborhood is.So the bespoke neighborhood, and perhapseven the eccentric neighborhood helpsus to understand that we can compose and constructneighborhoods that revolves around the individual,and perhaps even on multiple scales.So we can have the scale of, I don't know,the stairs, the building, the playground, the football field,

JOHN OSTH [continued]: the school neighborhood.And all these kinds of different scaled neighborhoodsthat still centers on the individualis useful, because it helps us to understandthe composition of individuals, and perhaps the opportunitieswe've got from a point of reference, usually the home.[How do you access data to study a neighborhood?]

JOHN OSTH [continued]: Traditionally, we used to have sort of governmental data.I'm from Sweden myself, so we've gotSCB, which is Statistics Sweden, that provides researchwith gridded statistics on 100-meter level.Also UK and US and many other Europeanand sort of many, many countries havegot these kinds of statistics, but they vary between.

JOHN OSTH [continued]: Now with sort of the digital times,we've got OpenStreetMaps, we've got so many different sourceswe literally could use.So traditionally, it's would have been easy to say.Nowadays, it's just a varied lot of data sources.The easy to get around or the easy to get kinds of datausually have got quality problems.OpenStreetMaps, of course, is built on anyone's contribution

JOHN OSTH [continued]: to the geography.And of course, some are doing bad thingsand writing odd words with streets and stuff like that.And on the other hand, when it comesto very controlled and high-quality data,usually it's very restricted.So it helps, of course, to be a researcher.It helps to have all these kinds of ethical forms beingplayed out.But this also varies between nations and regulations,

JOHN OSTH [continued]: and whatever.So it's very difficult to come up with one guiding rule.[What are the ethical implications of studyingBespoke Neighborhood?]Actually, far less than one mightthink, because the bespoke neighborhoods gives usa sense of the neighborhood, the context.And that means that we can leave the statistics somewhat off

JOHN OSTH [continued]: of the individual itself.Because if we know everything about the individual,if we know the level of education, gender, age,and so many other things, it starts becoming easierto understand who it actually is behind the sortof anonymous statistics.Usually, we don't have a name.But if you've got several attributes,it might be easier to find an individual.If you've got a bespoke neighborhood,

JOHN OSTH [continued]: it's easier to hide an individual.Because rather than talking about the individual per se,you could instead talk about the composition of neighbors.You could say something--within this neighborhood, it's sortof 50% of the 500 nearest neighbors are poor.It doesn't single out the individualin the center as being poor or rich or anything in between.So it's actually working in favor, I would say, for ethics.

JOHN OSTH [continued]: But again, usually, you want both the individual variableand the bespoke variables, and then you'restuck with the same ethical problem as always, I guess.[How do you combine all the different types of datacollected?]I mean, the only thing we really can sort of do anything aboutis that--think, for instance, OpenStreetMap.

JOHN OSTH [continued]: It comes with coordinates.So it could be transformed to coordinates.It could be projected into any kind of projection systemthat we want to use in a statistical system.And then we probably have got statisticsfrom an external source saying something about the area.Could be the ward.Could be something like that.The number of voters, the compositionof something or something.So it's relatively easy from a locational point of view.

JOHN OSTH [continued]: As long as we've got either coordinatesfor the location itself or for the area,it's really simple to merge all these togetherto create a composite dataset thatcomprises of contextual or individual-level things.[What tools would you recommend using for this type of datacollection?]

JOHN OSTH [continued]: In general, when it comes to the collection of data,from a spatial point of view, sortof a wide range of GIS softwares could be applied.Open ones that are sort of free and easy to use for anyoneis actually QGIS.And sort of readily--almost always updated with the latest functions and stuff.So it's a great software.

JOHN OSTH [continued]: But also RGIS and a few others thatprovides us with the opportunity to findthese kinds of spatial things.When it comes to bespoke neighborhoodsand how those kinds of conversations could be done,it depends on what technique you want to have.If you see the bespoke neighborhoodas a matter of radius, any traditional GIS softwarecould be used.But if you want to use what I traditionally-- what

JOHN OSTH [continued]: I try to do is sort of based on the k-nearest neighborapproach.Then there's another software called the EquiPop softwarethat you can use, which literallyhelps you to collect the contextual data basedon the number of neighbors rather thanthe meters around them.[How do you analyze these data?]

JOHN OSTH [continued]: Well, the final analysis is not different from very manyof what you see in econometrics or statistical analysis,as far as I'm concerned.I mean, much of it ends up in a big, grand regressionin the end or a multilevel regression,or anything like that.The more interesting thing is ratherhow we create the dataset before we start using it.rather than having the data just readily available for us,

JOHN OSTH [continued]: all these collecting and spatiallyjoining the data material from external sourceslike the OpenStreetMap or imagery,for instance remote sensing, or all these kinds of materialthat together could be combined to build a more cool--sorry for the wording-- dataset is somethingthat we'd bring in.So I would say we spend much, much more time on constructing

JOHN OSTH [continued]: the spatial contexts around the individualsin the final dataset.And then we apply it in what becomesa very spatially dependent, if you like,dataset for the final regressions and stuff.[How does data visualization work for this typeof research?]

JOHN OSTH [continued]: I'd say whenever you've got a result, as a geographer as Iam, I mean, the visual sort of what you could seeis literally what you get.But then again, the eye could be fooled.What we could see in terms of neighborhood,it depends on how we describe the cuttingpoints between values.How you want to distribute the colorsthat depicts poverty, for instance,

JOHN OSTH [continued]: if you want to do that, if it's goingto be the median value or the standard deviations.Or any kind of things, any techniquesare going to render different sort of color valuesthat you can actually fool with maps and stuff.So essentially, map making and howyou depict the things in a map are tremendously important.But it also is more than the illustration itself.

JOHN OSTH [continued]: Many, many times I find that whenyou want to find out if the material or the datayou've got actually is useful by plotting it on a mapor after a regression-- sorry for going very technical.But usually, if you run a complex set of regressionsand you save the residuals and thenyou plot the residuals on a map, then you

JOHN OSTH [continued]: see if there's a very spatial pattern.Those spatial patterns can tell you a lot about what is missingand what is in.So space and just how we view things can tell us a lot.And of course, there are measures for those thingsas well, but a quick glimpse will tell you a lot[How have you used these methods in a project?]

JOHN OSTH [continued]: So your specific project of mine wouldbe the development of the EquiPop softwarein the k-nearest neighbor approach.For me, I've always been interested on finding outthe neighborhood composition surrounding individuals.And not only for one neighborhood,but a multiple set of neighborhoodscorresponding to what we might call functions in society.

JOHN OSTH [continued]: Think of where you live.I don't know, apartment or house or anything,that the things you experience when you walk outyour door, the neighbors you meetmight be the, say, 12 nearest neighbors.That's the function of saying hello, knowing them, perhapshelping them if they're going on vacation with their mails,and stuff like that.And if you double these numbers, you

JOHN OSTH [continued]: get 25, 50, 100, 200, 400, 800, 1,600,and so forth and so forth.It's not only the number of neighbors that adds on.It's also the functions or perceived functions.Of course, no society is as sort of flat as I describe now,but there are things to it.When you reach your 400 nearest neighbors,you haven't got the same knowledge about them,

JOHN OSTH [continued]: but you probably have got a joint, I don't know,football field, sports field.When you've got your 10,000 nearest neighbors,you might have a school or a local supermarket,or something like that.Those kinds of counts will also tell yousomething about the probability of meeting individualsof different walks of life.So you might have the very poor individuals never meetinganyone else but other poor people on any levels

JOHN OSTH [continued]: on any neighborhoods.And you might have others where you're living pooror you're living wealthy.But at some level, you start meeting other individuals.And depending on the frequency, how often you go to the storeor how often you go to the hospital,you will be more or less prone to understandother individuals' sort of situations.So creating these kinds of egocentric,

JOHN OSTH [continued]: built on the individual, but also multi-scaled neighborhoodsis what motivated me to bring up the EquiPop software.And literally, the software is thenonly trying to describe these kinds of things,telling us what's the share of individual at eachof these key levels that you yourself as a user define,

JOHN OSTH [continued]: and how many of the, say, 500 nearest neighbors that fulfillyour settings, if you like.And it does it for everyone in the complete dataset.[Tell Us More About the Equipop Software]The software was built from the beginning for one reason only,

JOHN OSTH [continued]: and that's because the bespoke neighborhood k-nearest neighborapproach was difficult for [INAUDIBLE],,because literally understanding the nearestneighbor and the distance to all the neighbors forcedus to do a sorting on every individual case.So start with me, for instance.I needed to sort all the neighbors that I've got.If you have a country like Sweden,you've got 10 million neighbors.

JOHN OSTH [continued]: I need to sort them according to the distance from me,and then I can start doing so for the one, two, three, four,five.And then I'll cut up all the values that make sensewhen I come to 12, 25, 50, and so forth,and see what's the share.And that's easy and fine.But then when I reach the second individual,I have to redo sorting.I have to redo everything again.And that kind of computational material is just bizarre.You can't do it.

JOHN OSTH [continued]: So I came up not with a new idea, because the idea is old.Very old.But rather, a new way of thinking of the algorithm, howyou do this computationally.So it was by gridding the space.Then you always know-- think of a checkerboard of howthings are being located.Disregard where they live within each of these sort of units,square units.

JOHN OSTH [continued]: But rather, think that they are representedby the central point of these.Then you always know the next nearest unit,and then you can make a prediction.You don't have to sort.You always know the next nearest.So that makes computation much, much more easier.So the first version of EquiPop--soon a second one-- but the first oneactually made use of that kind of an idea

JOHN OSTH [continued]: and created the bespoke neighborhood calculationsusing a k-nearest approach with that kind of technique.[What skills do you need to conduct this kind of research?]They need basic understanding of geography, because firstand most importantly, there will be gridded statistics.

JOHN OSTH [continued]: Otherwise, you can't run it.So understanding how x and y--you could leave it at that.You don't need any zs or any complex levelsof hierarchies and stuff.But coordinate plotted things that comes in [INAUDIBLE]so you've got with 100 meters separation or 50 meterseparation, you've got a value that corresponds

JOHN OSTH [continued]: to the number of individuals.Those kinds of skills you need.Basically, skills when it comes to statistics.Rather, sort of very little, I would say.So there are lots of videos.Well, I couldn't say lots, but there are videos,and there are manuals, and there are previous sort of paperspublished in a wide field of journalsnow using the EquiPop software.

JOHN OSTH [continued]: So I think it's relatively easy to use.There's literally just five variables being used,and they are detailly prescribed in the manuals and stuff.Of course, again, the basic understandingof how to use this, and also some basic understandingof statistical assumptions being made.I mean, start with the fact that I said that we've got something

JOHN OSTH [continued]: with the 12 nearest neighbors.That literally assumes that 12 nearest neighbors is the same,regardless of where you are.And of course, any social scientist would say, well, no.Social science is just about the variation.But there's always sort of a few assumptions we have to make.So perhaps think of it as a baseline.Perhaps think of it as an experiment,

JOHN OSTH [continued]: and try to do the best of these kinds of statistics.So really, again, and as always, itdepends on what kinds of questionsyou approach science with, and what kinds of thingsyou want to test.So I think one has to have a quantitative-- a little bit,at least--state of mind when you approach these kinds of things.But from a methods perspective, very little is demanded.