The Division of Understanding: Specialization and Democratic Accountability

Referee: [Model Setup] The model defines integrators as agents who 'understand cross-domain policy consequences better than specialists' and then derives policy bias and welfare effects from that definition. The manuscript must show whether this superiority is derived from the learning-economies technology or imposed as a primitive; if the latter, the welfare gain from reducing the integrator share is at risk of being tautological rather than a genuine comparative-static result.

Authors: In our model, the learning-economies technology is the primitive: agents choose how to allocate their learning budget across domains subject to convex costs of multi-domain learning. Specialists optimally concentrate on a single domain, achieving high precision there but none elsewhere. Integrators allocate across domains, achieving moderate precision in each. Cross-domain policy consequences are state variables in each domain, and an agent's assessment of policy utility is the expectation conditional on their precision in those domains. Therefore, integrators have strictly superior assessments for policies spanning multiple domains by construction of the technology and optimization. This is derived, not imposed. The welfare result is not tautological because the market equilibrium features too few integrators (or too little breadth) due to the unpriced political externality; broadening increases the share of agents with cross-domain knowledge, improving policy choice and governance efficiency. We will add an explicit derivation of the understanding gap as a lemma in Section 2. revision: partial

Referee: [Welfare and Comparative Statics] The welfare claim requires that raising the share of broadened specialists moves the equilibrium policy vector strictly closer to the social optimum. The paper should explicitly rule out or incorporate the possibility that specialists aggregate cross-domain information through voting, parties, or delegation; without this, the asserted welfare superiority over the market allocation rests on an untested link between integrator share and governance efficiency.

Authors: The model assumes that policy evaluation requires domain-specific knowledge and that agents vote according to their individual assessments without costless aggregation or delegation. This isolates the civic externality from labor market choices. We agree that aggregation mechanisms could mitigate the issue and will add a robustness discussion in Section 4. Under the maintained assumption that verifying or communicating cross-domain implications is costly (consistent with the learning costs), full aggregation does not occur. We will show that the comparative static - that increasing the measure of agents with cross-domain knowledge improves welfare - holds as long as the social planner values the marginal integrative knowledge, which it does. This extension will be included in the revised version. revision: yes

Referee: [Political Equilibrium] The electoral-competition stage must be formalized so that the tilt toward integrators' interests is a derived equilibrium outcome rather than an assertion. If the bias is obtained only under particular functional forms or parameter restrictions, the result that broadening specialists raises welfare is not general and requires qualification.

Authors: The electoral stage is modeled as a standard probabilistic voting game in which candidates simultaneously propose policy vectors to maximize their expected vote share. Each voter i supports the candidate whose proposed policy maximizes the voter's expected utility given their knowledge vector. Because integrators have higher precision on cross-domain states, their ideal policies place greater weight on coherence across domains. The equilibrium policy is a weighted average of ideal points, with weights proportional to the measure of each type and their precision. Under standard assumptions of quadratic utility losses and concave production, the equilibrium tilts toward integrators' preferences whenever the integrator share is positive. This holds generally without further restrictions on functional forms beyond those stated. We will expand the formal characterization of this equilibrium in the main text and provide the proof in the appendix to make the derivation fully explicit. revision: partial