Response curves are key to any modelling project. They are vital in helping to determine how much should be invested in a given activity. A response curve shows the relationship between the cause, for example media spend, and the effect, for example sales.

The type of problem being solved and the nature of the data being analysed determine which response functions are suitable in a given situation. It is not a one-size-fits-all approach: different media, activities and resources call for different curves.

Variation is the name of the game
Greater variation in the input series gives a wider range against which to measure the response, and therefore more reliable results when a non-linear response is used. With little variation in the explanatory variable, there is no opportunity to measure the response at lower levels of input. In practice, it is then impossible to reliably quantify the response curve at those levels.

The most common functional forms
- Linear: the standard way to model if there is little variation, or if understanding diminishing returns is not an objective. Formula: y = α + β·x.
- AdBudg: a diminishing returns curve that may or may not be s-shaped. It is asymptotic, so it has a maximum lift, which is often useful to know. However, the turning point always lies at the average value of the input (or ad-stocked) series.
- AdBudg with γ = 1: as AdBudg, but without s-shaped relationships. It requires fewer parameters and is a safe bet where variation is reasonable.
- AdBudg with γ < 1: supports only diminishing returns, but the speed at which the curve approaches the asymptote can vary, so appropriate variation in the input is still needed.
- Power: a standard way to estimate diminishing returns. It is non-asymptotic and can tend towards a linear response, which is sometimes desirable.
- Logged: uses the natural log of the input series, y = α + β·ln(x). Not available if the input series contains zeros.
- Diminishing exponential: an alternative to AdBudg that can approximate a linear response, which AdBudg cannot. Useful if AdBudg forms do not give good results.
- ArcTan: an alternative to AdBudg with γ = 1, providing a steady diminishing returns shape. Useful if AdBudg forms do not give good results.




Choosing the right form
Each response curve has different properties, and the properties of your data and your objective determine which form to use:
| Data properties | Objective / hypothesis | Appropriate functional form |
|---|---|---|
| Little variation in the data | Diminishing returns unlikely | Linear |
| Reasonable variation | Understand diminishing returns; saturation point likely | AdBudg (γ = 1), ArcTan |
| Reasonable variation, non-zero values | Understand diminishing returns; saturation point likely | Logged |
| Large variation | Understand diminishing returns; saturation point likely | AdBudg (γ = 1 or < 1), diminishing exponential, reciprocal, ArcTan |
| Large variation | Response curve might be s-shaped | AdBudg |
| Large variation | Response might be linear; no saturation point | Power |
| Large variation | Response might be linear; saturation point likely | Diminishing exponential, ArcTan |
Want to know more?
We help brands and agencies choose, build and run the right measurement approach. Let’s talk about your questions in a 30-minute call.
