Objectives

• Compare methods of modelling big spatial data (BSD)

• Evaluate the opportunity of BSD compared to traditional sources of information

• Produce modelling workflows robust to frequent updates of datasets

What is a model?

The simplified representation of elements and processes from a more complex reality towards a specific purpose.

What is a complex model?

The simplified representation of elements and processes from a more complex reality towards a specific purpose.

A complex model has additional features:

• emergence
• local interactions
• "too much for the human brain"
• not solvable analytically
• heterogenous levels of interaction

Why different models?

Different models of the same reality can be

• confronted as alternatives
• complementary
• all wrong but all useful!

Different modelling frameworks (descriptive modelling & generative modelling for example)

• can be used for different purposes (ex. explain vs. predict)
• can be used in a workflow

Why model? (Epstein, 2008, JASSS)

1. Explain (very distinct from predict)
2. Guide data collection
3. Illuminate core dynamics
4. Suggest dynamical analogies
5. Discover new questions
6. Promote a scientific habit of mind
7. Bound (bracket) outcomes to plausible ranges
8. Illuminate core uncertainties.
9. Offer crisis options in near-real time
10. Demonstrate tradeoffs / suggest efficiencies
11. Challenge the robustness of prevailing theory through perturbations
12. Expose prevailing wisdom as incompatible with available data
13. Train practitioners
14. Discipline the policy dialogue
15. Educate the general public
16. Reveal the apparently simple (complex) to be complex (simple)

What is a model?

The simplified representation of elements and processes from a more complex reality towards a specific purpose.

What is data?

Operationally:

The simplified observation of elements from a more complex reality.

What is a model?

What is data?

Operationally:

The simplified observation of elements from a more complex reality.

data < model

The simplified representation of elements and processes from a more complex reality towards a specific purpose.

Specificity of big data

They are models built towards a purpose usually different to the one you will use them for.

ex. Mobile phone data located at cell tower

1. create voronoi polygons from scatter points
2. intersect polygons with other geography
3. interpolate distribution into intersections

> Creates two levels of spatial uncertainties

In this workshop...

• Make big spatial data usable for social study

• Compare sources of information on a specific subject: socioeconomic segregation

• Explore descriptive models of segregation with BSP

• Explore generative models of segregation with BSP (Schelling)

Big spatial data for social study

Case study: 

- Segregation analysis

- from Airbnb data

- in 3 Canadian metropolises (Montreal / Toronto / Vancouver)

Sources:

- Scraping of Airbnb listings (InsideAirBnB.com)

- Canadian Census API ('cancensus' R package)

Big spatial data for social study

> Get into the 'syllabus' folder

- copy/paste from USB sticks

- OR download zip package / clone from GitHub

> Open 'Modelling_Data_Part1.Rmd'

DON'T KNIT YET!

- Execute (later) the .Rmd file chunk by chunk

- OR copy/paste the chunks into a blanck .R file. In this case, use setwd(PATH_OF_'SYLLABUS') at the beginning

Airbnb Data (purpose/structure)

Airbnb Data (purpose/structure)

Big spatial data for social study

Listing file structure:

General info

Host info

Big spatial data for social study

Listing file structure:

Spatial info

Big spatial data for social study

Listing file structure:

Spatial info

Big spatial data for social study

Listing file structure:

Housing info

Economic info

Big spatial data for social study

Main problems:

Absence of info

Redundancy

Big spatial data for social study

Proxy definition

s

Identifying residential types   -11,639

Identifying residents vs. multi-owners   -7,168

Identifying current hosts   -7,499

Big spatial data for social study

Proxy definition

Estimating value per room in $  -150

Initial observations = 43,211

Final observations = 16,755

-60%

Big spatial data for social study

Spatial

sampling

bias

Big spatial data for social study

https://raisingthevillage.ca/social-identity-toolkit/

Big spatial data for social study

Big spatial data for social study

Descriptive models

Modelling the density of Airbnb listings based on:

• centrality of census tract in the city
• concentration of visible minorities in the tract

Modelling the relative price of Airbnb listings based on:

• centrality of listing in the city
• concentration of visible minorities in the surrounding tract

Descriptive models

Non linear relationship

between:

density of Airbnb listing

and

distance to the

city centre (City Hall)

Central

sample

Peripheral

sample

  10 km

Descriptive models

  Number of listings per tract

Price per room

Central

sample

Peripheral

sample

Complete

sample

Descriptive models

Segregation measures:

• on ethno-racial categories: Entropy divergence
• on airbnb relative price: Reardon ordinal index

Comparing cities

• Toronto
• Vancouver
• Montreal

Descriptive models

Entropy

Segregation

(minorities)

Reardon

Segregation

(Airbnb price)

0.18 

0.26

0.14

0.17

0.13

0.18

Generative models

Source: Gauvin et al., 2009

Generative models

Video: YouTube, Dan Olner.

Model implementation: Wilensky, U. (1997). NetLogo Segregation model.

Schelling model

• environment is a regular grid
• global vacancy percentage
• 2 groups of agents
   
• Each agent:
  • computes "happiness" regarding neighborhood occupation by other group agents and a tolerance threshold
  • teleports in an empty cell if unhappy

See Modeling/syllabus/Schelling.Rmd or SchelingwithOptim.Rmd

Generative model

Possible increments of the model rules:

• initialise with data
  
• change the mechanisms of relocation
• change the city (Vancouver, Montreal)
• simulate Airbnb listings