Testing, Testing: The Economics Of Social Conversions, Part 1

With so many variables at play, how can we dissect our data to give us new insights as to what is actually driving conversions and, ultimately, ROI? How can we measure all the way through the customer journey in order to truly understand what’s working, rather than making biased assumptions based solely on first or last click attributions. After studying Economics & Applied Mathematics in college, I understand the value of truly grasping the effects of campaigns in order to provide insights that can be built on for optimal growth. Here is my stab at looking at analytics from an economic standpoint:

Correlation vs. Causation

In economics, you cannot make conclusions on data simply based on correlation. Correlation does not imply causation. We cannot make significant attributions to factors that are associated with events without a controlled experiment, due to evident bias. The relationship between conversions and different variables is invaluable in providing a clearer direction to which your marketing campaign can become more effective in maximizing ROI. This is where testing comes into play.

The first step in looking at causation is defining your goal. What is it that you actually want to measure? Do you want to look at how cost per lead varies across different marketing channels? Do you want to analyze at the effects of a certain campaigns, content, or channels on ROI? Pick a variable to measure, and test it while controlling all your other variables. Through this understanding, you will have the ability to draw more accurate conclusions from your data and how it is affecting specific goals. Then you can focus a larger amount of energy on what is working and less on what is not working, in order to reap the rewards.

Social ROI is primarily and directly affected by five factors:

1.  Awareness
2.  Engagement
3.  Consideration
4.  Conversion
5  Customer Retention

That being said, we must look at the entire customer journey in order to make valid conclusions about our data. We must examine every single campaign a customer touches to understand their behavior, rather than simply using first or last click attribution. We must use that data to consecutively understand how these social campaigns impact sales, revenue, and cost. From that point, you will have the ability to compare ROI across all of your marketing channels in order to minimize costs and maximize profit.

The Model 

The base model is a regression of the conversion rate on the various digital variables involved in the customer journey, focusing on the influence of digital advertising efforts on ROI.

ROI_i = B₀ + B₁Awareness_i + B₂Engagement_i + B₃Consideration_i +
B₄Conversion_i + B₅CustomerRetention_i + X_i + u_i

In our multiple regression model, the dependent variable measures your return on investment with such predictors as awareness, engagement, consideration, conversion, and customer retention. However, there is existing multicollinearity, since these explanatory variables are highly correlated with each other, exhibiting a linear relationship. In this model, we will get valid results about how the entire group of predictors influences the outcome variable. But if you want accurate data about individual predictors, you will have to isolate your variables. By using fixed effects, we can account for certain variations and omitted variables when comparing across different populations, marketing channels, or assessing the effectiveness of certain campaigns.

An essential variation of the baseline model is controlling for variables such as campaign, marketing channel, content type, etc. Differences in such variables as brand awareness through advertising campaigns and reach can reinforce disparities across social channels and lead individuals to convert at different rates. Therefore, we included X_i as a vector of control variables. Different factors result in diverse social environments that ultimately impact brand perception and conversions. By using fixed effects, we can account for certain variations and omitted variables when comparing across different populations, marketing channels, campaigns, and therefore assessing the effectiveness of a certain digital advertising strategy. Examining the relationship between these metrics is essential to understanding the effects on ROI.

Testing:

Pre Campaign ROI: Conversions_i1= B₀ + X_i1 γ + u_i1

Post Campaign ROI: Conversions_i1= B₀ + β₁ Campaign_i + X_i2 γ + u_i2

δROI_i = B₁Campaign + δX_i γ + u_i

In our testing model, Campaign_i is an indicator variable for a sample population that views a particular advertising campaign and X_i is a vector of control variables. In order to interpret a variable as the causal effect in the model, we must think of it as an exogenous shift. In order to suggest a causal relationship, we must have an exogenous shift that we could use to compare the data because it holds everything else fixed and accounts for possible biases evident in the data. An exogenous shock would help us to compare cohorts just before or after in order to make a causal inference. In the example above, you could compare pre and post campaign ROI, while isolating all other variables in order to determine the total effect.

Part 2

Come back tomorrow, when we’ll dive into how to take a deeper look at social ROI. We’ll consider and look closely at Bias, Control Group Testing, User Analysis, Campaign Analysis, and Link Tagging. Ultimately, the primary goal of digital analytics is to achieve the greatest possible results with the least amount of effort, so stay tuned for Part 2, where the focus will be on working smarter, not harder!

In the meantime, post your comments and questions below.