« Beeminder home

Beeminder Blog

A bee carrying a marshmallow that's bigger than itself

By popular demand — specifically, being the winner of our poll — we’re catching you up on the latest research on the marshmallow test!  

The Stanford Marshmallow Experiment concluded that preschool kids who could resist gobbling a marshmallow for 15+ minutes in order to earn two marshmallows went on to become more successful adults. The original explanation was that kids with the ability to delay gratification (like by coming up with beeminder-y tricks to distract themselves from the tempting marshmallow) are served well by that skill for the rest of their lives. This was based on the original experiments in 1970 and especially 1972 plus multiple followup studies over decades to track the kids’ life outcomes. For example, in 1990 Mischel and colleagues concluded that the kids who were able to delay gratification as preschoolers got better SAT scores in high school.

Then in 2012, a study by Kidd et al called that explanation into question. It suggested that those seemingly impulsive kids are just the kids who don’t trust the adults who promise the second marshmallow. They’re thinking “yeah, I’ve heard that before” and gobble while the gobbling’s good. In other words, they’re not failing to delay gratification, they’re responding to the situation perfectly rationally based on their past interactions with adults. [1]

So flaking on your kids ruins their lives?

Well, those diligently patient 2-marshmallow kids did grow up to be markedly more successful. If that can be attributed to kids learning very early in life to rely on promises from adults, then heck yes, be diligent about following through on everything you promise to your kids. [2] It seems obvious that that’s important for a host of reasons so I suppose we don’t need to quibble about, say, all the research suggesting how little parenting actually impacts long-term outcomes. Maybe go ahead and be reliable just in case?

Wait, does any of this replicate?

Ah, good, you’re learning (that we’ve been learning much less than we thought from the social sciences). In this case, it sort of kind of does, a little! In 2018 a study by Watts et al came out aiming to disentangle Mischel et al’s claims — learning to delay gratification as a preschooler makes the rest of your life better — from Kidd et al’s claims, that the kids with the impressive outcomes are merely the ones brought up by more impressive adults.

If the second claim is true then the whole cottage sub-industry in education involving teaching kids self-control would be a waste of time.

The good news is that Watts et al used a much bigger sample size and focused on more disadvantaged kids. The bad news is that the conclusions are all quite nebulous and mostly not statistically significant. Also the way they separated their sample into kids whose mothers completed college or not sure looks suspicious (they have excuses for doing that; I’m not sure if they’re legit). They conclude that the correlation between self-control and life outcomes is there but at most half as strong as Mischel et al claim. They don’t seem to address Kidd et al’s theory that the 1-marshmallow kids are behaving perfectly rationally given their history with unreliable authority figures.

The biggest problem with all of this research is that there are no randomized controlled trials anywhere to be found. No causation can be established. Maybe it’s all just that kids delay gratification simply because they’re literally smarter and everything else follows from that. Other research suggests self-control matters as much or more than intelligence in predicting life outcomes, but either way, we just don’t have good evidence for the efficacy of any particular interventions.

My tentative conclusion from all this is that, first of all, sure, self-control as a kid is predictive of good life outcomes. Second — and this is based more on common sense than any of the research — parents and educators should give kids tools to augment self-control. (Like Beeminder! At least when they’re old enough to grok money.) Finally, modeling reliability for your kids sure can’t hurt.

Related reading

If you liked this, you may like other Beeminder blog posts about psychology research: “Smoking Sticks and Carrots”, “Social Reality And The Canard About Keeping Your Goals To Yourself”, “Negative Reinforcement ≠ Punishment”, “Ego Depletion Depletion”, and “What Is Willpower?”.

For more on self-control research, in my opinion George Loewenstein, Angela Duckworth, Katherine Milkman, and David Laibson are among the most impressive researchers.

(Scholastic tip: For all the research papers cited in this post, I’ve used DOI links that can be pasted into Sci-Hub. Bonus scholastic tip: Reading news articles about research papers is a telephone game. The actual papers have abstracts and intros and conclusions that are perfectly readable and tend not to egregiously misrepresent the actual findings. The New Yorker’s “Don’t! The secret of self-control” is pretty entertaining and informative though, if hopelessly out of date.)



[1] I discussed this with my own kids and my 11-year-old astutely pointed out that the “gobble while the gobbling’s good” rationale doesn’t hold too much water. As long as there’s even a probability of getting the second marshmallow, that’s worth the wait. Unless you view the adult as so unreliable that they’re liable to snatch the first marshmallow back from you if you don’t gobble it. I’m glad that possibility didn’t occur to my 11-year-old. Look how much trust and reliability we’re modeling!

[2] For the sake of technical correctness (the best kind of correctness) I should point out that correlation doesn’t imply causation and even if that interpretation of the marshmallow experiment is correct, it needn’t be that flaking on your kids ruins their lives. It could be that flaky parents just have flaky kids and it’s all genetic so there’s nothing you can do about it! But maybe be accountable for things you say to your kids just in case, is my point.

(I originally made this point (and in fact the seed of this whole blog post) in a Beeminder forum thread in the context of my side project, commits.to.)


Image credit: Faire Soule-Reeves