‘Causal effect’ is a term used in the research and statistics fields to describe when something has happened, or is happening, based on something that has occurred or is occurring. It’s a simple concept that has powerful implications for understanding the incremental impact of your retargeting campaigns, especially in a post-IDFA world.
Below, we explain the causal effect. We also explore how you can use it to better optimize campaigns and successfully retarget your best customers.
What is the Causal Effect?
The causal effect relates to causal reasoning which evaluates the relationship between events. When causal relationships exist, there is good reason to believe that events of one type (the causes) are systematically related to events of another type (the effects). Establishing these relationships allows researchers to affect change.
For example, if B happened because of A, and the outcome of B is strong or weak depending on how much impact A had, then a researcher can alter the influence of A to impact B.
In other words, causal reasoning gives experimenters the ability to alter test environments in order to produce (or prevent) events by identifying their causes.
Causal Effect Types
Causal relationships between variables can take on several types:
- Necessary causes. If A is a necessary cause of B, then the presence of B implies the presence of A with a probability of 100%. The presence of A, however, does not imply that B will occur. Necessary cause implies a maximum causal effect relationship between variables.
- Sufficient causes. If A is a sufficient cause of B, then the presence of A implies the presence of B with a probability of 100%. However, another cause (C) may also cause B. Therefore, the presence of B does not imply the presence of A. For example, if it is snowing outside, then the temperature is cool. Snow is a sufficient cause for one to conclude that it is cold outside. But just because it is cold does not necessarily mean it is snowing outside.
- Contributory causes. If A is a contributory cause of B, it means the presence of A makes the presence of B possible. However, not with a probability of 100%. In other words, a contributory cause may be neither necessary nor sufficient but it must be contributory. For example, cold temperatures cause snow, but are not the only factor that can affect snowy weather. Cold temperatures are a contributory cause to snow because snow is also influenced by wind chill, storm patterns, etc.
The Causal Effect Analysis and Retargeting
For marketers, understanding the causal effect of their retargeting ad spend can help them evaluate how much of their conversions are a result of their campaigns. How can marketers measure the causal effect of their retargeting campaigns? Incrementality testing!
Applying Causal Effect Analysis to Retargeting
To make inferences about the causal effect of retargeting campaigns, use your data to make predictions about what would have happened if a user had not been retargeted by an ad. This methodology follows the principle of RCT (randomized control trials). This method compares a treatment group (users that can be exposed to the ad campaign) with a control group (users that aren’t exposed to an ad campaign) to come up with a counterfactual.
- Counterfactual: a conditional statement of something that has not happened or would have been true under different circumstances. For example: if a user was not served a retargeted ad, then they would not have converted.
Evaluating the difference between an event that occurred (for example, a conversion) with a predicted counterfactual behavior is what enables incremental lift analysis.
Using Incrementality Testing to Measure Causal Effect
A basic incrementality test starts with the random selection of a test group and a control group. Generally, the control group includes 10% of users and the test group includes 90% of users. The test group receives an ad and the control group does not. The difference in the conversion rate of both populations is then measured for incremental conversions.
Isolating a control group — or holdout group — of users that you do not serve ads to is key to determining a sales revenue counterfactual. In other words, hold out a group of users to evaluate what would have happened without a retargeted ad campaign. Comparing this data to a treatment group then allows you to measure the causal effect of your retargeting campaigns.
Evaluating Necessary Causes
As mentioned above, there are different types of causal effect relationships. Ideally, marketers should aim to identify necessary causal relationships between their retargeting activities and conversions. That means the factors that lead to conversions with a probability of 100%.
In order to identify these factors, marketers need to create a clean testing environment in which covariates — influential variables that impact outcomes — are eliminated. This is where different incremental lift methodologies that control for noisy data are useful.
One such methodology is ghost bidding incremental lift testing. This methodology places ‘Ghost bids’ — or invisible bids — on the users in the control group. Two groups are also created in the control group: users who ‘would have been exposed’ and ‘would not have been exposed’. Conversions from the exposed users in the treatment group are then compared to the ‘would have been exposed’ users in the control group to measure for incremental lift. This creates an apples-to-apples comparison between users in the treatment and control groups by considering all users that could have been exposed to an ad in a specific moment. From there, the necessary causes of the incremental lift of the campaign can be identified.
Takeaways on Causal Effect Analyses & Retargeting
‘Causal effect’ is a term used to describe when something has happened based on something that has occurred. Measuring the causal effect of your retargeting campaigns using incrementality is a powerful way to evaluate their effectiveness in a post-IDFA world.
- To successfully measure causal effect, marketers should isolate a control or holdout group and test group to create a counterfactual.
- Counterfactual: a conditional statement of something that has not happened or would have been true under different circumstances.
- Control for noisy data by eliminating covariates. These are variables that are not of direct interest but influence the outcome of an experiment.
Retarget your app
Reach out to us for help navigating the post-IDFA world for mobile engagement and ROI success!