• mvriens

When the path to impact is not direct

The surprising influence of indirect effects.

Indirect effects play an important role in marketing research and analytics though they are not always considered in commercial practice. In this blog we outline briefly what indirect effects are -- and why you should care.

As always this is not meant to be comprehensive. Still I have encountered indirect effects regularly, and the following examples will give you a good idea in what type of situations we may encounter indirect effects.

Example 1: Driver modeling

Whenever a company does research to measure customer satisfaction they usually want to do more than just measuring overall satisfaction: they also want to understand the sub-components of overall satisfaction and what the relative impact of these sub-components on overall satisfaction is. For example, say want to measure how overall satisfied students are with their college (one question). In addition, to this overall question we asked the students how overall satisfied they were with

  • the faculty

  • with course selection

  • with overall quality of courses, and

  • with the choice of major and minor programs

We could hypothesize that each of these sub-questions affects overall satisfaction ratings: e.g. the more major and minor programs the higher overall satisfaction. We might quantify these effects using some type of regression analysis. We call these effects, direct effects.

However, one might also hypothesize: If we improve the quality of our faculty, we would be improving the satisfaction with the quality of our courses? If that would happen, we would have an indirect effect. Namely, the improved faculty leads to improved quality of courses perception, and the improved quality of courses lead to an increase in overall satisfaction. If quality of faculty also impacts overall satisfaction directly, we have a direct and an indirect effect. To understand how important quality of faculty is, we need to sum the direct effect and the indirect effect.

Example 2: Marketing mix models

Say, we want to understand the effect of various marketing activities on sales. We have activities such as: weekly ad spend on TV, radio, billboards, online ads, etc. We might have also some “intermediary” variables such as website visits etc. Now, direct effects would be any direct effect that the advertising spending variables have on sales; for example, TV ad spending may have a direct effect on sales. Likewise, we might hypothesize that website visits also translates in higher sales (i.e. a certain percentage of the website visitors is converted to actual sales), also a direct effect.

However, we might also hypothesize that TV ad spending will increase the awareness of the company’s website, leading to more visitors, which leads to more sales. This would be a possible indirect effect of TV ad sales on actual sales (indirect because it operates through website visits). So, to fully understand the total effect of TV spending we need to add up the direct and the indirect effect. Just to give one example of how interesting indirect effects can be: Wiesel, Pauwels & Arts (2011) found that almost 75% of the effect of Google Ad words was achieved indirectly: i.e. Google AdWords had a direct effect on sales, but it also had an effect on requests for information and quote, and these subsequently lead to sales.

Example 3: Consumer behavior models

In consumer experiments we may vary certain “interventions” to study what impact these interventions may have on what consumers choose (e.g. buy). Say we want to investigate whether we can re-direct how much attention consumer give to 2 choice alternatives. For example, we show them a series of two snacks, and they tell us which one they would pick: e.g. a Mars bar and a small bag with chocolate covered almonds. I want to test if pointing an arrow to one of the two snacks has an impact on what they choose. I measure (1) how long they look at each snack (attention, let’s call it gaze duration) and what they choose. The results show that consumers look longer at snacks to which the arrow points to. So, there is a direct effect of the arrow on gaze duration

Now, I also capture snack choice and I find gaze duration has an impact on snack choice (i.e. the longer I look at an alternative, the more likely I will pick that alternative). I might find that the arrow has no effect on snack choice: so, there is no direct effect. However, there is an indirect effect of the arrow on snack choice because the arrow impacts gaze duration, and gaze duration impacts snack choice (See Vriens, Vidden & Schomaker, 2020).


There are various statistical approaches one can use to get some insight into indirect effects, depending a little on the exact situation: i.e. in each of the examples we would probably go about differently. For example, in the customer satisfaction driver, to calculate the indirect effect we could multiply the direct effect of quality of faculty on quality of courses with the direct effect of quality of courses on overall satisfaction. That indirect effect would then be added to the direct effect of quality of courses on satisfaction to get the total or net effect of quality of faculty. There are other ways to calculate indirect of mediation effects, which we'll cover in upcoming posts.

Disclaimer: Blogs are not scientific articles. There is a lot more that can be said about this topic - join us on the Kwantum LinkedIn page, or reach out to us at marco.vriens at teamkwantum dot com.


Wiesel, T., Pauwels, K. & Arts, J. (2011). Marketing's Profit Impact: Quantifying Online and

Off-line Funnel Progress, Marketing Science, 30. 4, 604-611.

Vriens, M., Vidden, C. & Schomaker, J. (2020). What I see is what I want: Top down attention biasing choice behavior. Journal of Business Research, 111, 262-269.