We report the results of a laboratory experiment based on a citizen‐candidate model with private information about ideal points. Inefficient political polarization is observed in all treatments; that is, citizens with extreme ideal points enter as candidates more often than moderate citizens. Second, less entry occurs, with even greater polarization, when voters have directional information about candidates' ideal points, using ideological party labels. Nonetheless, this directional information is welfare enhancing because the inefficiency from greater polarization is outweighed by lower total entry costs and better voter information. Third, entry rates are decreasing in group size and the entry cost. These findings are all implied by properties of the unique symmetric Bayesian equilibrium of the entry game. Quantitatively, we observe too little (too much) entry when the theoretical entry rates are high (low). This general pattern of observed biases in entry rates is implied by logit quantal response equilibrium.