Understanding Incrementality Measurement
Polaris uses two techniques to measure incrementality:
Media mix modeling
Incrementality experiments
It is important to understand these techniques, at least at a basic level, so you have confidence in the incrementality measurement Polaris provides. The resources linked in this article will give you that necessary knowledge.
These resources contain information about incrementality measurement in general as well as the basics of media mix modeling and incrementality experiments.
This short presentation provides basic information about incrementality, synthetically controlled incrementality experiments, and media mix modeling, in a simple, narrative format. It's the best place to start because the content is specifically focused on the precise techniques Polaris uses within the context of marketing incrementality measurement. Most of the other resources are a bit more general, especially those not produced by MetricWorks.
Format: Presentation
Audience: Everyone
These resources contain information about media mix modeling.
This web page lays out, in simple terms, how media mix modeling works, and provides practical advice on preparing for and building models at each phase of the process. It is written by media mix modeling experts at Meta.
Format: HTML
Audience: Analysts
This whitepaper lays out the structure of a basic media mix model including the underlying math, in a simplified way. While quite simplified, it can provide a good basis for understanding the technique, even though Polaris uses much more complex and powerful models.
Format: PDF
Audience: Analysts
Meta's Robyn open source project can be thought of as a fairly simple reference implementation of media mix modeling. It is written in R and includes pretty thorough documentation. While Polaris uses significantly more powerful models, browsing through the Robyn code can provide a solid intuition of the core concepts.
Format: Open Source Project
Audience: Analysts
This academic paper provides the math for media mix modeling and then discusses specific challenges that are often encountered as well as potential solutions. Polaris incorporates many of the solutions offered in the paper as well as other novel features important for adapting media mix modeling to performance marketing.
Format: PDF
Audience: Data Science
This academic paper goes into detail on how to use a Bayesian technique to build and fit a media mix model, accounting for complexity like carryover (adstock) and shape (response curve) effects, all of which Polaris models automatically.
Format: PDF
Audience: Data Science
These resources contain information about incrementality experiments using the synthetic control method.
This 30-minute video presents an easy-to-understand walk-through of synthetically controlled experiments and how Google's open source CausalImpact library can be used to run them. Polaris uses a very similar technique to run its incrementality experiments.
Format: Video
Audience: Everyone
This academic paper details how Bayesian structural time series models can be used for causal inference and what to consider when building and applying such a model. This is quite similar to the technique used by Polaris for incrementality experiments.
Format: PDF
Audience: Data Science
This academic paper details the synthetic control method and how it has been used for causal inference in a variety of fields, which types of studies it's most useful for, and how to best apply it. The synthetic control method is the foundation of Polaris's incrementality experiment methodology.
Format: PDF
Audience: Data Science
Media mix modeling
Incrementality experiments
It is important to understand these techniques, at least at a basic level, so you have confidence in the incrementality measurement Polaris provides. The resources linked in this article will give you that necessary knowledge.
General
These resources contain information about incrementality measurement in general as well as the basics of media mix modeling and incrementality experiments.
Incrementality Explained
This short presentation provides basic information about incrementality, synthetically controlled incrementality experiments, and media mix modeling, in a simple, narrative format. It's the best place to start because the content is specifically focused on the precise techniques Polaris uses within the context of marketing incrementality measurement. Most of the other resources are a bit more general, especially those not produced by MetricWorks.
Format: Presentation
Audience: Everyone
Media Mix Modeling
These resources contain information about media mix modeling.
Analyst's Guide to MMM
This web page lays out, in simple terms, how media mix modeling works, and provides practical advice on preparing for and building models at each phase of the process. It is written by media mix modeling experts at Meta.
Format: HTML
Audience: Analysts
Building and Validating Media Mix Models
This whitepaper lays out the structure of a basic media mix model including the underlying math, in a simplified way. While quite simplified, it can provide a good basis for understanding the technique, even though Polaris uses much more complex and powerful models.
Format: PDF
Audience: Analysts
Robyn
Meta's Robyn open source project can be thought of as a fairly simple reference implementation of media mix modeling. It is written in R and includes pretty thorough documentation. While Polaris uses significantly more powerful models, browsing through the Robyn code can provide a solid intuition of the core concepts.
Format: Open Source Project
Audience: Analysts
Challenges And Opportunities In Media Mix Modeling
This academic paper provides the math for media mix modeling and then discusses specific challenges that are often encountered as well as potential solutions. Polaris incorporates many of the solutions offered in the paper as well as other novel features important for adapting media mix modeling to performance marketing.
Format: PDF
Audience: Data Science
Bayesian Methods for Media Mix Modeling with Carryover and Shape Effects
This academic paper goes into detail on how to use a Bayesian technique to build and fit a media mix model, accounting for complexity like carryover (adstock) and shape (response curve) effects, all of which Polaris models automatically.
Format: PDF
Audience: Data Science
Incrementality Experiments
These resources contain information about incrementality experiments using the synthetic control method.
Inferring the Effect of an Event Using CausalImpact
This 30-minute video presents an easy-to-understand walk-through of synthetically controlled experiments and how Google's open source CausalImpact library can be used to run them. Polaris uses a very similar technique to run its incrementality experiments.
Format: Video
Audience: Everyone
Inferring Causal Impact Using Bayesian Structural Time-Series Models
This academic paper details how Bayesian structural time series models can be used for causal inference and what to consider when building and applying such a model. This is quite similar to the technique used by Polaris for incrementality experiments.
Format: PDF
Audience: Data Science
Using Synthetic Controls:Feasibility, Data Requirements, and Methodological Aspects
This academic paper details the synthetic control method and how it has been used for causal inference in a variety of fields, which types of studies it's most useful for, and how to best apply it. The synthetic control method is the foundation of Polaris's incrementality experiment methodology.
Format: PDF
Audience: Data Science
Updated on: 18/09/2023
Thank you!