We study a model of public decision-making in simple public goods economies with moral hazards and adverse selection. Economic agents must invest resources (or provide effort) to discover their own preferences. We consider direct revelation mechanisms based on sampling. A sample of agents is drawn in the population, and each member of the sample reports a preferences type to a Principal. The determinants of the "representative sample" size are studied. The structure and magnitude of effort and sampling costs affects the optimal number of representatives. If the net social value of the effort is high, first and second best optimality require a maximal sample (or "direct democracy"). If, on the contrary, effort is too costly, the recourse to samples ("representative democracy") is justified as a second best. To obtain the results, we not only take effort and revelation incentives into account, but also restrict decision rules to satisfy an additional property of robustness to opportunistic manipulation by the Principal, which forbids the use of a priori knowledge in public decision procedures.