There is concern that the Internet causes ideological polarization through selective exposure and the echo chamber effect. This paper examines the effect of social media on polarization by applying a difference-in-difference approach to panel data of 50 thousand respondents in Japan. Japan is good case for this research because other factors affecting polarization like huge wealth gap and massive immigration are not serious issue, thus it offers quasi natural experimental situation to test the effect of the Internet. The results show that people who started using social media during the research period (targets) were no more polarized than people who did not (controls). There was a tendency for younger and politically moderate people to be less polarized. The only case in which the Internet increased polarization was for already radical people who started using Twitter. However, since radical people represent only 20% of the population and there was no effect for Facebook or blogs, the overall effect of the Internet was moderation, not polarization.