This paper summarizes observations from four empirical studies focusing on shape replication with three input methods. The aim was to identify and assess how the components of several semi-randomly generated shapes influence how accurately untrained users can replicate each of these components. We found that the pen is the least and touch the most error-prone method when used for drawing. Additionally, the distribution of errors was analyzed. The results may be used to predict which shape properties make shape replication more difficult. Additionally, the results may be used to design shapes that are easy to replicate.