Simultaneous interpretation has to be conducted to shorten delay and reduce working memory load. Interpreters adopt a strategy that involves the generation of a target language in the word order that is similar to that of the source language. To clarify how the interpreters decide the word order in practice, we conduct a statistical study on the word order of simultaneous interpretation, based on comparison with written translation. The word order comparisons indicate that the strategy of maintaining the original word order is more likely to be adopted in cases where the word order is not consistent in different translations. We analyze the features of source languages, including dependencies of chunk pairs, length of post modifier, and distance between two chunks, and confirm that such features affect the word-order decision in simultaneous interpretation.