When education rewards rise, parents face a sharper trade-off between siblings

Parents dividing educational resources between siblings may want both children to succeed without leaving either behind. Research by Esther Gehrke and Michelle González Amador examines how that balance changes when higher levels of education bring disproportionately larger rewards. Their experiment finds that the same incentive structure produces different choices depending on parents’ willingness to accept unequal outcomes.

Why the shape of educational rewards matters

An extra investment in education need not bring the same reward at every starting point. When returns are convex, gains become larger at higher levels of human capital—the skills and knowledge that contribute to earning potential. This can make concentrating resources on an already advantaged child more rewarding than helping a sibling catch up.

In their November 2026 preprint, Esther Gehrke and Michelle González Amador develop a model of how parents respond to that trade-off. Parents divide a fixed educational budget between two children while balancing combined outcomes against inequality between them.

The model does not predict a universal response. Strongly equality-minded parents may compensate the less advantaged child more as rewards become steeper. More inequality-tolerant parents may concentrate resources further on the advantaged child. Choices also depend on how easily additional investment can make up for a lower initial endowment.

An experiment with real educational stakes

The researchers recruited families through 16 high schools in San Luis Potosí, Mexico. Eligible families had two children enrolled at the same high school. Of 278 participating parents, 246 entered the main analysis after exclusions for failed comprehension checks or unvarying allocations.

Each parent divided nine tokens between two children in six scenarios. Tokens determined access to sessions of a purpose-built artificial intelligence course offering up to 15 sessions. One scenario was randomly selected for implementation, giving the choices real educational consequences.

The scenarios combined two payoff structures—linear and convex—with three levels of substitutability between initial endowments and investment. Tablet graphics showed how moving tokens changed each child’s course access.

Children were identified only by colors, concealing which sibling would benefit from a particular allocation. Although children completed a problem-solving test, the numerical endowments used in the scenarios were controlled by the researchers, not taken from actual test scores. Test rankings were used to assign course access when implementing the selected choice. This distinction matters: the experiment manipulated payoff rules, not children’s underlying abilities.

Three patterns of parental choice

Using a structural model fitted to each parent’s six decisions, the authors classified parents into three groups:

Rawlsian parents: 12.6%. Their choices were consistent with prioritizing the worse-off child’s outcome, even at substantial cost to combined payoffs.
Interior parents: 76.8%. These parents balanced equality against total benefits rather than consistently choosing either extreme.
Utilitarian parents: 10.6%. Their choices were consistent with maximizing combined payoffs. When several choices produced the same maximum, they often selected the more equal option.

These are model-based classifications of experimental behavior, not independently established personality types. The estimates also depend on a separately elicited measure of how parents value additional course sessions.

The authors report an average absolute prediction error of 0.84 tokens on the nine-token scale and a correlation of 0.88 between predicted and observed allocations. Those figures describe fit to the experimental choices, not predictions of future earnings.

Steeper payoffs shifted resources—but not uniformly

Moving from linear to convex payoffs increased allocations to the higher-endowed child by an average of 1.88 tokens. It also increased the probability of reinforcing the initial advantage by 26 percentage points.

The average concealed substantial differences. Allocations to the higher-endowed child rose by about 2.02 tokens among interior parents and 3.16 among Utilitarian parents. Among Rawlsian parents, the estimated change was essentially zero and statistically indistinguishable from zero.

Yet greater convexity also reduced the probability of choosing the allocation that maximized combined payoffs by 29.7 percentage points. These findings are compatible: many parents moved toward the advantaged child without accepting the extreme concentration required to maximize the total.

What the findings can establish

The experiment shows responses to controlled educational incentives, not the effects of an actual change in labor-market wages. Course access was the payoff; later learning and earnings were not the reported outcomes.

The sample also cannot establish how common these parental types are across Mexico. Participants could keep two children in high school, and the eligible population was unknown, preventing calculation of formal participation rates.

The authors suggest that convex returns could widen sibling inequalities in some families while making equality more costly in others. Their broader arguments about inequality between families and social mobility remain theoretical implications, rather than outcomes demonstrated by this experiment.