Retail Gravity Models and GIS

1
Retail Gravity Models and GIS

• Jonathan Dorwart
• GIS Analyst / Planner
• Wikstrom Planning and Economic Consultants
• April 5, 2007

2
Retail Gravity Models and GIS
3
Typical market analysis estimates competitive
position on the basis of trade areas
Travel time
Fixed radii and breakpoints
Sales data from individuals
4
Why develop a gravity model in GIS to model
retail competition?

• Better model the impacts of competition when
  customer datanot available (basis of trade area
  has to do with character of development)
• Additional information for making zoning
  decisions ordetermining optimum amount of retail
  in a proposed development
• Allows a pro forma determination of the
  suitability of retail on a site in a given
  market
• Pedagogic in the sense that it lets you
  visualize competition and opportunities
• Data availability Census, local area population
  projections, sales tax and land use data
  (including competing retail) readily available

5
Building the Model

• Identify problem
• Conceptualize main ingredients/symbolize
• Utility Size/Distance
• Operationalize Define variables
• Test (observed v. predicted)
• Herein lies the problem
• Evaluation
• How to use results Advantages vs. Disadvantages

6
Conceptual Basis

• Idealization of consumer behavior
• In absence of empirical data to articulate
  consumer decision-making process three naïve
  assumptions are made
• Psychological trade off between size and distance
• This tradeoff can be generalized across a number
  of competing sites (all other things being
  equal)
• Little consideration of scale or regional
  context (Original Reilly model was built as an
  analogy of regional patterns between cities in
  Texas)
• Is the model well-specified???

7
Variations on the Gravity Model
Newton (1687)
Reilly (1931) Social physics
Huff (1964)Probability
Estimating the friction of distance using
multiple linear regression Gravity models
multiplicative nature allows log linear form to
be taken, which can be regressed upon.
8
Implementation

• Our model is based on Huffs and integrated into
  a raster based data model in GIS using map
  algebra
• Total gravity surface derived and then a
  specific gravity (capture rate) calculated for
  site in question
• The total number of consumers captured are
  determined by population density.
• Distance can be measured as a crow flies, along
  a transportation network or in terms of
  estimated travel time.

9
Implementation

• Large number of calculations (retail sites)
  requires scripting in either Python/VBA
• GIS software is similar to Lego blocks
• Utilize existing data on transportation network
  to arrive at travel times as a proxy for
  distance (not always current)
• Testing/Calibration? Well now, that is a good
  questionhow much do travel time/distance and
  size of retail establish influence consumers?
• OUR PROBLEM IS WE HAVE NO OBSERVATIONS!!!

10
Implementation

• How can consumer buying power be estimated from
  this model?
• Per capita spending determined from state sales
  tax receipts
• What is the relationship of consumer buying power
  to store size?
• Amount of available buying power can be
  translated in to gross leasable area of a store
• How does this relate to the site?
• FAR, parking requirements, setbacks, building
  height

11
In reality the market area circumscribed this
area
12
For the sake of simplicity many stores were
aggregated into shopping centers/areas. Strip
nature of retail development and point based
representation used in gravity model is
problematic.
13
This is an example of a simplified dasymetric map
current urban land use constrains population
density
14
(No Transcript)
15
Value of Model and Lessons Learned

• Model is pedagogic One (flexible) step in
  decision-making process
• Decisions are political in nature (within
  community/firm)
• Always tension between theory and empiricism
• This model is challenged by lack of hypothetical
  deductive spiral(i.e.. no observations to test)
• However, difficulties in operationalizing
  development process
• in retail sector are great
• Minimum of three year development process this
  is not a
• laboratory
• Lack of access to consumer behavior data. Why?
• See Birken in GIS, Spatial Analysis, and Modeling
• Community politics in entitlement process
• Other factors Land use is fragmented, consumer
  confidence changes, capital markets, etc.
• Question is what information is useful in a
  given situation. Sometimes simple models provide
  good initial guidance.