ISE 754 (Fall 2026) 2.5 Aggregate Demand
Watch on YouTube →
Overview
The lecture explains how to infer and model aggregate demand when customer-level demand data is unavailable, using U.S. population datasets in LogJam at resolutions suited to each problem. It also introduces an area-adjusted great-circle distance for population centroids, demonstrates how aggregate distances affect facility location results, and shows how to join census-tract income data to population data for a weighted location analysis.
Key takeaways
- Use actual customer or demand locations when available; population is only a first-order proxy and may fail for products whose demand varies systematically by region.
- Select geographic resolution according to the task: three-digit ZIP codes offer broad U.S. coverage, five-digit ZIP codes better represent North Carolina, and census block groups support detailed metro-area analysis.
- Use FIPS codes such as 37183 to join government datasets to Wake County reliably instead of matching potentially ambiguous place names.
- Population-centroid distances should not be zero when a facility is located at the centroid, because residents are distributed across the area; the area-adjusted estimate adds a minimum within-area service distance.
- The area adjustment is an approximation based on uniform population and circular regions, so its outputs should be interpreted as modeled average distances rather than precise routes.
- Joining census-tract income data to LogJam population data shifted the example's optimal location by only 1.42 miles, indicating little difference between population and income weighting in that case.
Chapters
- A map of 1,300 U.S. foundries shows strong Great Lakes concentration, unlike a map of cities with at least 10,000 residents.
- When actual customer locations are unavailable, population can approximate demand for broadly distributed retail products, with exceptions such as winter gear in Florida.
- Population datasets share an underlying population count but differ in geographic aggregation and resolution, which determines their usefulness.
- Metropolitan and combined statistical areas aggregate counties to represent regions such as Raleigh and Atlanta; city boundaries alone can misrepresent market size.
- FIPS identifiers provide stable geographic keys: North Carolina is 37 and Wake County is 183, making 37183 the unique county identifier.
- LogJam offers places, ZIP codes, census block groups, tracts, counties, and metropolitan areas; three-digit ZIP codes often work well nationally, while five-digit ZIP codes provide better North Carolina coverage.
- Census block groups are the finest listed level with population centroids, average roughly 1,500 people, and can support detailed metro-area logistics analysis.
- A population centroid is an abstract point, not a place where residents necessarily live, so treating a facility at the centroid as zero-distance service understates travel.
- The area-adjusted distance uses the greater of the great-circle distance to the centroid and an estimated average distance within the demand area, modeled as a circle.
- The approximation assumes uniform population and circular geography; a more elaborate fitted formula is an empirical regression estimate rather than an exact analytical result.
- Applying the adjustment to a facility-location problem changes both total cost and some selected locations compared with the earlier model using only 20 unadjusted points.
- External datasets may contain missing values, so data wrangling with the DataFrames package is needed before analysis.
- A CSV file supplies income data associated with census tracts, which can be joined to LogJam population geography.
- Re-solving the location problem with income weighting produces a location about 1.42 miles from the population-based result, suggesting income is roughly proportional to population in this example.
Transcript excerpt
0:02 The focus of this lecture is how do you actually get data to represent demand? We'll be talking about a variety of different techniques. First, I want to point out that for this lecture there are quite a number of new Log Jam functions used. There's a modification to our distance calculation. You notice the A at the end. It's basically an area adjusted great circle distance. We'll talk more about that in a minute. There's also a list of the different data sources available in Log Jam. This is data. It represents different sources of data. Its primary use is that at different levels of resolution, it allows us to represent the population centroids
0:53 throughout the United States. This data is just for the United States, but similar data could be created for other countries. Also, for this lecture for the first time in addition to the regular companion script there's a data file. The dot CSV means comma separated value. That'll be our default way of representing data. It's basically the simple text version of spreadsheet data. In this case, the data refers to income levels tied to census tracts. So, we'll have that as an example at the end of the lecture. First, there's two figures at the beginning of the lecture. They really illustrate the point of the difference between dealing with population data which for this one shows all of the cities with at least 10,000
1:41 population in the continental United States. And then comparing that to this data set here which is 1,300 US foundries. And as you can see in the figure there is much more concentration around the Great Lakes area relative to the top figure, which is the entire US population. The difference is that this is just tied to US population. So,
This is the start of the transcript. Save this video free for the whole transcript, searchable and time-coded.
Related notes
Summary, takeaways, and chapters were generated by AI from the video's transcript and may contain errors. The video belongs to its creator, Michael G. Kay.