Why Consumers Don't Treat Every Product as a Substitute
October 2026 | Applied Economics | Industrial Organization | ← Back to Blog
An elasticity tells you how many buyers a product loses when its price rises. It does not tell you where they go. This post explains why that answer is rarely “to everything else in proportion to size,” what a basic logit assumes about it, and how modeling differences across consumers produces more realistic substitution.
The Core Problem: Where Do Buyers Go?
When a product's price rises, two questions follow. The first is how many buyers it loses. An elasticity answers that. It measures how much the quantity demanded of that product falls for a given percentage increase in its price.
The second question is where those buyers go. An elasticity says nothing about it. A diversion ratio does. It is the share of the product's lost sales that each alternative picks up, and the shares add up to everything lost, including buyers who stop buying altogether.
Buyers tend to move to the closest alternative, or to something cheaper. The simplest demand models do not build that in.
What a Basic Logit Assumes
The simplest demand model, a basic logit, builds in an assumption called independence of irrelevant alternatives, or IIA. The relative odds of choosing one product over another do not change when a third product is added or removed, however similar that third product is to either of them.
The classic illustration is the red bus and the blue bus. Suppose commuters divide evenly between driving a car and taking a red bus. Now add a blue bus that is identical to the red one except for its color. Because driving and the red bus start out equally popular and the blue bus is identical to the red one, a basic logit gives all three equal appeal, so each gets a third of commuters. In reality almost nobody chooses a blue bus over a red one because of paint. The two buses would share the red bus's riders, and driving would keep its half.
The same logic applies to price changes. When one price rises, the buyers who leave spread out across the remaining products in proportion to their market shares, and the most popular product gains the most, whether or not it looks anything like what they were buying.
"A basic logit sends lost buyers to other products in proportion to their size, whether or not those products look anything like what they were buying."
Why That Fails in Real Markets
Take airlines. When a low-cost carrier raises its fares, the travelers who leave tend to move to another low-cost carrier more often than to a premium one. Someone who booked the cheapest fare to save money is not likely to trade up because the budget airline got more expensive. Proportional substitution sends those travelers to the biggest airline anyway, whether or not it resembles what they were buying.
The people who leave are not a random sample of that carrier's customers. They tend to be the ones who cared most about the fare, so they are more likely to land on another cheap option than on whichever airline happens to be biggest.
How Heterogeneous Preferences Fix It
A random-coefficients model, the approach behind the Berry, Levinsohn and Pakes (BLP) framework, starts from a different assumption. In a basic logit every consumer weighs price and product characteristics the same way. In a random-coefficients model those weights differ across consumers. Some travelers care mostly about the fare, others about schedule or service, and the model estimates from the data how much those tastes vary.
That one change alters where buyers go. The travelers who leave a low-cost carrier after a fare increase are disproportionately the ones who care most about price, and they still care about price after they leave. So they land on other products that are cheap, not on whatever is largest. Substitution now follows how similar products are on the dimensions consumers care about, instead of how popular they are.
Why This Matters for Pricing and Competition Analysis
Any decision that depends on where lost buyers go depends on getting this right. A pricing team deciding how far to raise a price needs to know how many buyers leave and how many stay within its own product line. An economist assessing competition between two firms needs to know how close they are as substitutes. A model that sends lost buyers to other products in proportion to their market shares gets both wrong, and any pricing or competition conclusion built on it inherits the error.
Modeling differences across consumers lets the predicted price effects follow the close substitutes instead of the whole market.
This flexibility is not free. The model needs variation in the data that reveals how tastes differ. When that variation is limited, the estimate of how much tastes differ is imprecise, and the substitution patterns that depend on it become less reliable. That is a reason to test how sensitive the conclusions are to that estimate.
Further Reading
Berry, S., Levinsohn, J. and Pakes, A. (1995). Automobile Prices in Market Equilibrium. Econometrica. The foundational paper for random-coefficients demand estimation in differentiated-product markets.
Nevo, A. (2000). A Practitioner's Guide to Estimation of Random-Coefficients Logit Models of Demand. Journal of Economics & Management Strategy. A step-by-step guide to implementing the approach.
Train, K. (2009). Discrete Choice Methods with Simulation, 2nd ed. Cambridge University Press. A standard textbook treatment of logit models and their extensions, including IIA.
