USRetailCentres

Integrating the Who, What and Where of American Retail Centre Geographies.

This repo contains all the code used for the paper: Integrating the Who, What and Where of American Retail Centre Geographies. In this paper, using data from SafeGraph we develop a comprehensive understanding of the geographies of American retail centres. The paper has three sections/aims:

1. The 'Where' - Delineating the spatial extent of American retail centre agglomerations.
2. The 'What' - Developing a multidimensional American retail centre typology.
3. The 'Who' - Building a calibrated Huff model to estimate retail centre catchments.

---

Winner of the 2021 Robin Flowerdew Prize

This paper was selected as the 2021 winner of the 'Robin Flowerdew Prize for Best Postgraduate Paper' at the RGS-IBG International Annual Conference. Check out the blog post about the award and paper HERE.

Slides from the presentation are available to download HERE.

For those registered at the conference, you can watch the session 'The future of quantitative geography (1)' HERE.

---

Part One - The 'Where'

To delineate the spatial extents of American Retail Centres, we use a methodology that is based on the hexagonal spatial indexing system; H3. In essense the methodology aggregates retail locations to hexagons, and then using graph objects we are able to delineate contiguous tracts of retail locations based on how they interact with other hexagons in the dataset.

The retail locations are derived using a number of datasets:

• SafeGraph Retail Places - points
• SafeGraph Retail Places Building Geometries - polygons
• OSM Retail Land-Use - polygons

All the necessary functions can be found HERE.

For improved performance - particularly useful with larger states (CA, NJ, NY), there is also a set of the same functions written to utilise parallelisation, for improved performance. These can be found in HERE.

---

Part Two - The 'What'

We adopt the framework first proposed by Dolega et al. (2019) in developing a multidimensional classification - the approach involves gathering a series of variables about the retail centres, dimensionality reduction and then clustering using PAM.

The typology accounts for four key domains of retail:

• Composition - proportions of different retail categories.
• Diversity - proportions of differential ownership (e.g. Independents), local/national category diversity.
• Size & Function - size, shape, density (retail, roads, residential, employment)
• Economic Health - visit frequencies and dwell times, low income neighbourhoods.

All the necessary functions can be found HERE.

For improved performance - particularly in extracting optimal k values and performing PCA, there is also a set of functions written to utilise parallelisation, for improved performance. These can be found HERE.

---

Part Three - The 'Who'

To extract catchments for the centres we build a calibrated Huff model, utilising the SafeGraph 'weekly patterns' data. The model predicts patronage from census tracts to retail centres, accounting for attractiveness and distance (as below), and the model parameters - alpha and beta - are calibrated by comparing a series of Huff models against 'observed' probabilities obtained from the 'weekly patterns' data.

• Attractiveness - size, diversity, total number of visits
• Distance - road network distances computed using the hereR API.

For improved performance, the majority of the functions have been converted to work in parallel and are available HERE.