irstly by using game model we find that individuals selfinterest will lead to low ex-ante efficiency of innovation investment in the complete information environment, but the herd behaviors occurring under the condition of incomplete information may be a possible way to solve this problem because they can make the attempt for free riding come to nothing. Then we parameterize the model, carry out laboratory experiments with senior undergraduates as subjects, and test the theoretical predictions. The final results show that: i) the ex-ante efficiency is a gradual process of reducing under the condition of complete information, it decreases towards the theoretical equilibrium but does not stabilize on it; ii) the herd behaviors actually have a positive incentive effect on improving this puzzle, but there are also some limitations.