Network Analyst/Business Analyst Location Allocation Improvement by Assigning Cutoff Values in a Field for Facilities
When you want to maximize market share in a metropolitan area using a Location Allocation model to find new, optimized real estate locations, the current model requires that you specify a single cutoff value that is used for all facilities in the model. This works OK, but it doesn't reflect consumer retail behaviors or real estate service area expectations unless the area is homogenous (which is rarely the case when I'm trying to run this model). When the metro area has widely varying densities of demand, the current model options don't work very well. If you create smaller areas and run multiple models, you end up with "edge effects" where the model may select a Chosen location on the edge of the market area simply because it is not considering facilities that are beyond the arbitrary border that we've selected.
One way to solve this problem would be to allow the user to specify different cutoff values for each facility (required, competitor, or candidate) through a field in the attribute table in much the same way that we indicate different weights or capacity values as shown below.
If we could do this, the results would be more reflective of good real estate choices for the Chosen locations, and you'd get more accurate demand weights as an output. While the processing time would likely increase, with today's computer processing power, it would be good to improve the model with this capability.
In the example shown below, the red facilities (competitors) likely have a 15 or 20-minute drive time service area. The blue facility on the west side of the market area would also have a 15 or 20-minute drive time service area, but we want the blue facility near downtown Pittsburgh to have a 10-minute drive time service area because the demand weights are the most concentrated in this area and there are likely enough consumers to support stores within a smaller service area. Thus, we want to specify that the candidate locations in the high-density demand areas have a shorter drive time cut off than the candidate locations that are in more suburban and rural areas. When people live in rural areas, it is not uncommon to drive a half hour to a retail store, so we should have a model that is more reflective of that human expectation and behavior.
The purple heat map layer below shows my demand points which vary throughout the market area. The light green dots show possible candidate locations created from 1-mile centroid points covering the entire market area. This "what if" scenario can create a model that is potentially too large for efficient processing, depending on the number of demand points, but the candidate locations can be reduced through further analysis and prioritization. The green stars represent the results of several models run with different cutoff values. The results require an additional analysis step to remove locations that end up being selected in the same area (like the two locations in Washington or New Kensington). These are different chosen locations due to different drive time cutoff values. This is inefficient and still doesn't calculate a demand weight where the facilities have different cut off values, which would be more accurate.
