Spatial Demography with Open Source Tools

class: center, middle, inverse, title-slide

# Spatial Demography with Open Source Tools
### Kyle Walker
### May 5, 2021

---

## Software and spatial demography

---

## Problems with traditional workflows

* Point-and-click software can inhibit reproducibility and collaboration

* Even if an analysis is scripted:

- The use of commercial software makes workflows inaccessible to analysts who cannot afford the license fees;
  
--
  
  - Data access is rarely automated and is difficult to reproduce; 
  
--
  
  - Data wrangling, modeling, and visualization are "siloed" in different commercial software packages

---

## Open source workflows

* __Data sources__: open data APIs

* __Data models__: GeoJSON, simple features (attributes + geometry in single file/object)

* __Modeling & analytics tools__: sf + spdep (R), GeoPandas + PySAL (Python)

* __Presentation tools__: cartography (ggplot2, tmap) + dashboarding (Shiny) in same software environment as analysis

---

## Open data APIs

.pull-left[

* Open data APIs allow analysts to programmatically access data resources

* Examples include official government APIs (US Census) and third-party APIs (e.g. Canada's CensusMapper)

* Software tools like __censusapi__ and __tigris__ (US) and __cancensus__ (Canada) allow for programmatic access to data resources

]

.pull-right[

```r
library(cancensus)

vancouver_extract <- get_census(
  dataset = "CA16",
  regions = list(CSD = "5915022"),
  vectors = c("v_CA16_1364", 
              "v_CA16_2397",
              "v_CA16_244"),
  level = "CT",
  geo_format = "sf",
  labels = "short"
)
```

]

---

## Data models

.pull-left[

* Frameworks like the __tidyverse__ (R) and __pandas__ (Python) are widely used for data representation in open-source workflows

* Geospatial extensions to these frameworks (__sf__ for R, __GeoPandas__ for Python) allow spatial demographers to analyze data in similar ways to a regular dataset without dedicated GIS software

]

.pull-right[

```r
library(tidyverse)
library(sf)

vancouver_data <- vancouver_extract %>%
  transmute(
    tract_id = GeoUID,
    pct_english = 100 * (v_CA16_1364 / Population),
    median_income = v_CA16_2397,
    pct_65_up = 100 * (v_CA16_244 / Population)
  ) %>%
  st_transform(3005)
```
]

---

```r
vancouver_data
```

```
## Simple feature collection with 117 features and 4 fields
## Geometry type: MULTIPOLYGON
## Dimension:     XY
## Bounding box:  xmin: 1202113 ymin: 468924.5 xmax: 1217107 ymax: 481725.6
## Projected CRS: NAD83 / BC Albers
## First 10 features:
##      tract_id pct_english median_income pct_65_up
## 1  9330017.02    44.10434         63936 15.153800
## 2  9330008.02    74.76190        125184 19.047619
## 3  9330014.01    34.37095         64128 19.455253
## 4  9330014.02    35.86957         53029 19.481605
## 5  9330059.13    69.74542         72082  4.839049
## 6  9330059.14    72.17253         78822  8.207857
## 7  9330005.00    65.11915         44160 10.202684
## 8  9330009.00    61.66409         60544 19.428411
## 9  9330011.00    46.93878         63334 17.762661
## 10 9330012.00    56.65268         71851 15.490002
##                          geometry
## 1  MULTIPOLYGON (((1215538 473...
## 2  MULTIPOLYGON (((1206463 470...
## 3  MULTIPOLYGON (((1214371 471...
## 4  MULTIPOLYGON (((1214432 472...
## 5  MULTIPOLYGON (((1210995 477...
## 6  MULTIPOLYGON (((1210995 477...
## 7  MULTIPOLYGON (((1210212 470...
## 8  MULTIPOLYGON (((1208479 472...
## 9  MULTIPOLYGON (((1212078 471...
## 10 MULTIPOLYGON (((1212063 472...
```

---

```r
plot(vancouver_data$geometry)
```

![](index_files/figure-html/plot-data-1.png)

---

## Modeling & analytics

.pull-left[

* Models fit to spatial data require modeling _spatial dependence_, often represented with a neighborhood spatial weights matrix

* Example workflow: regionalization with the SKATER algorithm, equivalent to ArcGIS's Spatially Constrained Multivariate Clustering

]

.pull-right[

![](index_files/figure-html/spdep-1.png)

]

---

## Modeling and analytics

---

## Presentation tools

---

## Sharing and reproducibility

.pull-left[

* This entire workflow is fully reproducible - [just take a look at the source code](https://github.com/walkerke/paa2021/blob/main/code/paa_code.R) and run it for yourselves!

* Adoption of open science practices in spatial demography facilitates _collaboration_, _education_, and _innovation_

]

.pull-right[

]

---
class: middle, center, inverse

## Thank you!