Very young infants operate on a simple assumption: what I like, everyone likes. If a 14-month-old prefers crackers to broccoli, they assume an adult will also prefer crackers. This is an egocentric projection of desire. However, by around 18 months of age, a profound cognitive shift occurs. Children begin to understand that desires are subjective—that another person might genuinely prefer the broccoli.
How do children bridge this gap? Historically, developmental psychology lacked a unified mathematical framework to explain this transition. A recent line of research frames the child not merely as growing into this capability through passive maturation, but as an intuitive statistician or econometrician solving a latent variable problem.
The Child as an Econometrician
To understand why this transition works computationally, we can look to econometrics. When consumers choose products, economists use discrete choice models to infer their underlying preferences. In A rational model of preference learning and choice prediction by children, Christopher G. Lucas et al. proposed that children utilize a remarkably similar mathematical framework: the Mixed Multinomial Logit (MML) model.
In an MML framework, an individual's choice is modeled as a probabilistic function of the latent utility (or value) they assign to an object. If a child observes an adult choosing an item, the child can invert this probabilistic model using Bayesian inference to calculate the likelihood of the adult's hidden preferences.
Crucially, the "mixed" aspect of the model allows the learner to entertain competing hypotheses about how preferences are distributed across society. The child evaluates two main possibilities:
- Universal Preferences: Everyone shares the same utility function (the egocentric baseline).
- Subjective Preferences: Different individuals have distinct, independent utility functions.
By treating preferences as latent variables within a hierarchical Bayesian model, a child can transition from the first hypothesis to the second when confronted with the right data.
Overcoming Sparse Data with Statistical Evidence
Why does this econometric model accurately capture the developmental shift? The answer lies in how children evaluate statistical sampling. In The Child as Econometrician: A Rational Model of Preference Understanding in Children, Fei Xu et al. demonstrated that the MML model accurately predicts how toddlers infer preferences from sparse behavioral data.
Imagine a box filled mostly with red toys and a few blue toys. If an adult reaches in and pulls out a red toy, it tells the child very little about the adult's desires; they might just be grabbing what is common. But if the adult consistently selects the rare blue toys, the statistical improbability of this action provides strong evidence of a specific preference.
The authors showed that children use this exact type of statistical evidence to override their initial egocentric assumptions. When the observed data cannot be reasonably explained by a universal preference model, the Bayesian posterior probability shifts toward the subjective preference model. The child rationally concludes that the adult possesses a distinct, individual taste.
Empirical Proof: Accelerating Theory of Mind
Is this mathematical framework just a neat description of behavior, or does it reflect a genuine inference mechanism that can be actively triggered? To test this, Tiffany Doan et al. conducted an infant training study detailed in Learning to reason about desires: An infant training study.
The authors targeted 14-to-17-month-old infants, who typically fail to comprehend subjective desires. By exposing these infants to explicit statistical evidence—scenarios where different individuals expressed conflicting preferences for objects—the authors successfully trained the infants to infer and act upon another person's subjective desires early.
Importantly, the infants did not just learn a simple "game of opposites" (i.e., assuming adults always want the opposite of what the infant wants). When tested with new individuals who happened to share the infant's tastes, the infants correctly aligned their actions with the shared preference. This demonstrates that the training triggered a genuine capacity to model individual-specific utility functions, accelerating a core component of Theory of Mind through targeted statistical evidence.
Key Foundational Work
A rational model of preference learning and choice prediction by children (2008) — contributed 3 concept elements to this direction.
The Child as Econometrician: A Rational Model of Preference Understanding in Children (2014) — contributed 2 concept elements to this direction.