Along with the increase of nuclear families and the aging society, the social isolation of elderly people is becoming a big issue. Social robots are being regarded as suitable interfaces to support these elderly people to communicate with others (e.g. family members). In robot-mediated communication, it can be expected that asynchronous messaging functions will cope with differences in life styles between different generations, and that message inputs can be facilitated by natural language dialogue functions. Eliciting self-disclosure from elderly people by utilizing such robots is important to draw out social support from other people. In this study, the following three research questions were examined with the aim of exploring requirements for robots to encourage elderly people's self-disclosure. First, using a self-disclosure scale, we examined topics that were encouraged to disclose from elderly speakers by an intermediary robot. Secondly, we investigated the difference between social attributes (robot-specific traits and characteristics) of two kinds of robots with different behaviors. Last, we investigated the relationship between the robot's social attributes and self-disclosure from these two independent data.