An Example of Spatial Analysis

Imagine that we operate a successful retail store in Kansas City. We're ready to expand our business by opening a second store. 

We've found a potential location in St. Louis, but would it be a good place for a second store?

In this example, we'll show how spatial analysis can help answer this question. We'll analyze sales at our existing store, and link them with demographic and economic data for the surrounding area. That will let us understand the relationships between existing sales, customer characteristics, and customer locations. We'll apply that knowledge to predict sales we could expect at the new location. That will help us make the decision whether to open a store there or not.

This spreadsheet comes from a customer loyalty card program at the existing store, and it lists the customer address and amount of each sale made at our store over the course of several months.

Many organizations have data like this, whether they're private companies keeping track of customers in a CRM system, government programs managing client databases, or utilities managing projects in a work order management system.

Geocoding is the process of converting street addresses or similar text information into coordinates that can be displayed on a map. We used  ArcGIS Maps for Office  to geocode these sales records and map them.

But we're not just interested in customer locations; we also want to know how much was sold in each location, and we want to be able to compare those amounts to other statistics like population counts, income and age.

Spatial aggregation is frequently used to summarize point-based information, like our sales points, into larger areas so that we can make comparisons between areas or link it to other information about those areas. 

We aggregated our sales data by Census block group, so that we can link it to hundreds of demographic statistics from  ArcGIS Community Analyst , government agencies, third-party data providers, and other sources.

On this map we can see that the sum of purchases made by customers varies from block group to block group. If there's a pattern to those variations, and they aren't just random, we may be able to identify some of the underlying factors that influence our sales. So is there more to this pattern than random variation?

Hot spot analysis uses statistical methods to answer that question. We used it to determine if there are clusters of block groups with significantly higher-than-average sales (hot spots) or clusters with significantly lower-than-average sales (cold spots). 

The red areas indicate hot spots, and the blue areas indicate cold spots. Hot spot analysis tells us that there is, in fact, a significant spatial pattern to our sales, and they are not just randomly distributed.

We would expect that the easier it is for customers to get to our store, the more likely they are to shop there. So travel time is one possible factor that influences the amount we sell to customers in each block group. That might explain the hot spot immediately around the store. It's easier for those customers to get to our store, so they shop there more.  

Network analysis is used to model networks such as roads or pipelines, and to analyze distances and movement along those networks. We used  ArcGIS Network Analyst  to calculate the fastest route from each block group to our store. Those routes are the red lines on this map. 

After calculating these routes, we associated each block group with its travel time and mapped the results.

Another factor that might influence sales is population. We would expect to make more sales to block groups that have more potential customers in them. 

We obtained block group population data from  ArcGIS Community Analyst , and mapped it here. We can see an area on the south side of the city where there's a greater concentration of people. This might explain the sales hot spot we saw in that area.

If our store attracts people of a particular income level, then we might also want to consider how household income influences sales. This map shows median household income by block group, from ArcGIS Community Analyst. 

If our store sells high-end merchandise, the hot spot of sales we saw in the southwest area of the city might be related to the higher-than-average incomes in that area. 

Perhaps our store appeals to a particular age group. This is a map of median age by block group. 

It's not obvious to our eyes whether there's any strong relationship between median age and sales; some of the hot spots we saw appear to be in older areas, and some are in younger areas. Spatial analysis will tell us whether there's a relationship between sales and age, regardless of whether or not we can see it here.

We've identified four things that might influence sales in each block group: 

• Travel time to our store
• Population in the block group
• Household income levels
• Average age of the population

Just looking at these maps, however, we can't be certain which of these four factors really have a significant influence, and which do not. We'll apply another analysis technique to make that determination.

 Exploratory regression  is a process that evaluates every combination of these four factors, and determines which of them are useful in predicting sales by block group, and which are not.

The exploratory regression tool tells us that travel time, population, and income are strongly related to the amount of sales made to each block group, but age is not.

Now we can evaluate the exact nature of those relationships.

 Ordinary Regression  is a statistical method that examines the way in which one or more input variables, such as travel time, population, and income, affect a single output variable, such as sales.  

We applied the Ordinary Least Squares regression tool to our three variables to quantify their relationships to sales.

This gave us an equation that expresses the relationship between a block group's population, household income, and travel time to the amount of sales we make to customers living in that block group.

We used this equation to calculate "predicted" sales for each block group around our existing store. Then we mapped those estimates so we could compare them to the actual sales from our spreadsheet, and see how well the predictions matched the actual sales. 

This map shows the sales predicted by our equation. It looks fairly similar to the map of actual sales, but we can't judge just by looking at a map. We won't go into the details here, but we used additional spatial analysis techniques, like searching for patterns in the differences between predicted and actual sales, to check the accuracy and completeness of our equation. We found that the equation is accurate enough for our purposes.

Now we could take what we learned from sales at the existing store and apply it to the new location to estimate how much we might sell if we opened a store there.

Here's the proposed second location in St. Louis again. From our analysis of sales at our existing store, we learned the relationships between travel time, population, and income and sales made at our store. Next, we looked at those same factors in the area around this new location:

We applied the equation we got from analyzing the sales at our existing store to estimate sales by block group for the area around this new location.

Here are the predicted sales by block group for the potential new store. At this point, we could total these numbers up and decide whether this amount of revenue is enough to support a second store. 

 Of course, there are other factors that would go into making this decision, such as:

• Rent and utility costs
• Tax rates
• Costs of transporting merchandise
• Traffic patterns and ease of access by potential customers
• Locations of our competitors

We can apply spatial analysis techniques to evaluate all of these factors as well, giving us more information to make a better decision about whether or not to open this new store.