This paper is about higher-order theory of mind such as I think that you think that I think . Previous studies have argued that using higher-order theory of mind in the context of strategic games is difficult and cognitively demanding. In contrast, we claim that performance depends on task properties such as instruction, training, and procedure of asking for social reasoning. In an experiment based on a two-player game, we manipulated these task properties and found that higher-order theory of mind improved by providing step-by-step instruction and training. It also improved during the experiment when participants were explicitly asked to predict the opponents next move.