What is a multicollinearity analysis?


In the context of Marketing Mix Modeling (MMM) , a multicollinearity plot is a key tool to analyze the relationship between the explanatory variables of the model. This documentation explains why and how this type of analysis is used.


1. Understand the relationships between variables


Multicollinearity occurs when two or more independent variables are highly correlated with each other (meaning they move in the same direction).
This can be problematic because:
  • Difficulty in isolating the effect of each variable : If the variables are highly correlated, it becomes difficult for the model to determine which variable most influences the dependent variable (e.g., turnover).
  • Biased interpretation of coefficients : Coefficients of correlated variables may become unstable or inconsistent, complicating their interpretation.
A multicollinearity plot, often in the form of a correlation matrix or heatmap, helps visualize these relationships and detect problematic variables.


2. Identify redundant variables

If two variables are highly correlated (e.g., > 0.8 or < -0.8), they often bring similar information to the model. This can lead to:
  • Unnecessary complexity of the model : By adding variables that do not provide new information.
  • Overfitting : The model may overfit historical data, reducing its generalization ability.
The graph thus makes it possible to identify these redundancies and make decisions, such as merging or removing certain variables.



3. Leverage strategic insights

Multicollinearity analysis is not only used for data cleaning, but can also provide insights into how your marketing mix is ​​working.
For example :
  • A strong correlation between TV and SEA spending may indicate that these channels are often activated together.
  • This can guide marketing strategy by revealing potential synergies or areas of dependency.

Conclusion : A key step for any Marketing Mix Modeling project
A multicollinearity analysis is an essential diagnostic tool which therefore allows:
To optimize the quality and robustness of the model.
To better understand the interactions between the variables of the marketing mix.
To improve the model's ability to provide actionable recommendations.