How to combine Incentives analysis and Behavioral Insights to change behaviors

While it’s true that Behavioral Insights by themselves can modify individual behaviors, don’t forget that in the real world, Incentives are much more powerful in terms of driving impactful changes. Of course, using both in a combined fashion yields the best results.

Incentives trump Choice Architecture in terms of impact

If you take a look at my ICAR Behavior Analysis framework, you’ll see that the 2 main dimensions of analysis are Incentives and Choice Architecture.

Incentives is the first one on purpose: by modifying the Incentives for a given agent, you are much more likely to change its behavior than if you simply modify its Choice Architecture context.

Take the example of the behavioral intervention where researchers managed to increase apple’s consumption by students at school simply by replacing whole apples by pre-sliced apples, playing on the Behavioral Tool of Ease.

While the result is real and striking, it cannot overcome Incentives.

Imagine that students haven’t eaten for 12 hours and this randomized trial is repeated, with 50% getting whole apples and 50% getting pre-sliced apples (and nothing else to eat but apples). We could safely predict that most students will eat the apples regardless of their form, and the boost in consumption resulting from pre-slicing the apples, so from Choice Architecture, would tend towards zero.

So, in every situation, first change Incentives if you can, before changing Choice Architecture. But there is a second point to this.

You cannot assume you know the Incentives of an agent

The main difficulty in changing incentives is to change them in a way that produces your desired results without negative externalities and unforeseen consequences.

Once key aspect of that is to get the Incentives right, and this is easier said than down.

Say we repeat the experiment above, reasoning correctly that students will be starving after 12 hours and will eat apples regardless of their form. But this time, we see vast groups of students choosing not to eat anything instead of eating the apples, regardless of format.

We may conclude that after 12 hours, they are not starving enough, and we may choose to test 24 hours next time. We may conclude that we missed something and do qualitative research with those students, trying to understand and map their incentives. We would interview students that haven’t eaten the apples and may get few things out of them (except that they’re pissed off about the whole thing).

If we persevere and interview students that did eat the apples, we may learn that there is some social dynamics going on at this school about apples and that the social “leaders” (cool kids that others want to imitate, or on the contrary bully that others don’t want to displease) have decreed that eating red apples will get you sick (or is just not cool), by opposition to green apples. If you used read apples in your experiment, that’s your explanation.

The lesson from this is: you cannot guess Incentives and you cannot assume you know them without doing your due diligence and researching them.

Trying to change behaviors without deep Incentives analysis is hopeless

In terms of real world project, the key takeaway is that researching and understanding the Incentives faced by the agents whose behavior you want to change should be the very first task you need to do.

You need to understand what motivates them, what constraints them, what they value, their margins for action and change, their web of power relationships, etc. Then you can plan on changing their behaviors through changing their Incentives and Choice Architecture.

The absolute best way, in my opinion, to do this Incentives is to use the tools of strategic analysis of organizations, the organizational theory created by Michel Crozier and Erhard Friedberg (disclosure: I was trained by them at the Center for the Sociology of Organizations at Sciences-Po Paris). This approach is the single most effective one I found to date, and I have looked at many others, but I always come back to its tools. There is a good short overview in this working paper, and most unfortunately for non-French speakers, it seems their two key books have been translated in English but only in hard to find academic editions.