The evidence

What the research actually says about paying kids for schoolwork

Written 10 September 2026 by Daniel, who built this

The settled part is narrower than either side of this argument usually admits. Paying a child for a specific thing they can go and do moves that thing, and paying for a grade or a test score mostly does not. Whether the money quietly damages a child's own interest is not settled, and the largest experiments that went looking for that damage did not find it. The real gaps are in what happens after the payments stop and in whether any of it transfers to math.

Start with the result that holds up

In the 2007 to 2008 and 2008 to 2009 school years, the economist Roland Fryer ran randomized incentive experiments in public schools in Chicago, Dallas, New York City and Washington DC, distributing $6.3 million to roughly 38,000 students across 261 schools. Four cities, four different designs, and the useful part is that the designs disagreed with each other.

New York paid fourth and seventh graders for their scores on interim assessments, and Chicago paid ninth graders for their grades in five core courses. Neither moved achievement. The New York seventh grade math estimates sat between −0.018 and −0.030 standard deviations, which is a polite way of writing zero, and Chicago's grade point averages rose slightly while test scores did not budge.

Dallas paid second graders $2 for each book they read, confirmed by a short computer quiz they had to score 80 percent on. The average child earned $13.81 for the year, and reading comprehension rose between 0.180 and 0.249 standard deviations, a statistically significant result. Washington DC paid sixth through eighth graders for attendance, behavior, uniforms and homework, and got a smaller reading gain that was only marginally significant.

Fryer's term for the split is inputs versus outputs. Pay for the behavior and you get the behavior with some learning attached. Pay for the result and you get nothing. His own explanation is the one I find most convincing: a second grader told to raise their test score has no idea what to do about it on a Tuesday afternoon, and a second grader told to read a book and pass the quiz knows exactly. The paper is Financial Incentives and Student Achievement: Evidence from Randomized Trials, published in the Quarterly Journal of Economics in 2011.

What that licenses is fairly small: if you are going to attach money to schoolwork, attach it to something the child can start doing today. It does not tell you how much to pay, whether it survives a parent running it at the kitchen table rather than a district, or whether it works for math.

The part people are still arguing about

The standard objection is the overjustification effect, and the study behind it is real. In 1973, Lepper, Greene and Nisbett watched preschoolers at the Bing Nursery School at Stanford and picked out 51 children who already loved drawing with markers. Some were promised a Good Player Award certificate for drawing, some got one as a surprise afterward, and some got nothing. In later free play, the promised group spent 8.59 percent of their time drawing, the no-award group 16.73 percent, and the surprise-award group 18.09 percent. That last number is the detail that usually gets dropped: the promise did the damage, not the certificate. The paper is short and worth reading.

The strong version of the case is Deci, Koestner and Ryan's 1999 meta-analysis of 128 studies, which found that tangible rewards significantly undermined free-choice intrinsic motivation, with effect sizes around −0.28 to −0.40, and that the effect ran worse for children than for college students.

Then it gets messy. Cameron and Pierce had published a competing meta-analysis in 1994 finding that reward does not generally decrease intrinsic motivation, and their reply to their critics was titled Protests and Accusations Do Not Alter the Results, which tells you the temperature of the field. When Cerasoli, Nicklin and Ford ran a meta-analysis across 183 samples and 212,468 people in 2014, they counted nine previous meta-analyses on this one question. Their conclusion: crowding out shows up when money is tied directly to performance and much less when it is tied indirectly, and incentives and intrinsic motivation are not necessarily antagonistic.

So anyone who tells you the science has settled this is telling you which meta-analysis they read.

The boundary condition that keeps getting skipped

Every one of those classic experiments studies an activity the person already enjoys. Deci and his co-authors excluded boring tasks from their main analysis on purpose, and said why: you would not expect a reward to undermine motivation that was not there to begin with. In their own table of the few studies that ran both an interesting and a dull version of the same task, the reward effect usually flipped sign on the dull one.

That does not settle anything in my favor, but it tells you which question you are asking. If your child already loves math, the undermining literature is about your child. If your child regards times tables as a punishment, most of it is not, and far fewer people have studied what happens next.

Fryer measured this directly as well. He gave the Intrinsic Motivation Inventory to students in all four cities and found nothing: input experiments trended slightly positive, output experiments slightly negative, every estimate too small and too noisy to call anything but a null.

What nobody actually knows

Whether it lasts after the money stops

This is the question every parent asks and the one with the least evidence behind it. Fryer could only follow up in Dallas. A year after the payments ended the reading effect had fallen to roughly half its size and was no longer statistically significant, which he compares to the fade-out seen after Head Start or a year with an excellent teacher. That is one follow-up, one city, one subject. Anyone claiming to know what happens three years later is guessing.

Whether math behaves like reading

Here is the study I would rather not have found. In the 2010 to 2011 school year, Fryer, Devi and Holden paid fifth graders in fifty low-performing Houston schools $2 for each math objective they mastered in practice software, with parents paid as well and teachers eligible for large bonuses. The results are genuinely mixed.

The children mastered far more objectives and math scores rose by 0.081 standard deviations. Reading scores, which nobody was paying for, fell by 0.084 and canceled the gain out. Split by prior achievement it gets worse: the stronger students gained 0.228 in math with no reading loss, while the weaker students gained nothing in math and lost 0.165 in reading. A year after the money stopped, those weaker students were down 0.223 in math and 0.170 in reading. The authors read that as children learning something discouraging about their own ability rather than motivation being crowded out; their survey measure of motivation barely moved.

That study is not this app. A district ran it on a school curriculum, it paid parents and teachers too, and the schools were picked for low performance. But it is the closest anyone has come to studying paid math practice, and two things in it are worth carrying: paying for one subject can pull effort out of the others, and the kids who most need help may be the ones it helps least.

Where this app sits in all that

PiggyBankMath pays for a completed practice session. No grade, no test score, nothing that talks to a school. That is a deliberate extension of Fryer's input finding, and I want to be precise about the word extension: he paid second graders to read books, and I am applying the same logic to timed math practice because finishing a session is something a child can decide to do on a given afternoon, the way finishing a book is. Nobody has tested that. There is no study of this app and there will not be one soon.

Two details came out of the research rather than around it. Payment is triggered by finishing a round and not by getting the answers right, because the input finding is about the behavior. And the daily cap stops the money while mastery, the streak and the level all keep recording, which is the closest I could get to showing a kid that the work still counts once nobody is paying for it.

If someone in this argument sounds certain, check what they are certain about. Two claims have real support: paying for grades does very little, and paying for a specific behavior can move achievement. Everything past that, including the version of it I have built, is somebody's reasonable guess.