Survey researchers have dedicated increasing attention toward understanding how social networks influence individual-level outcomes like attitudes, political participation, and other forms of collective action. This work implicitly assumes that errors in individuals’ self-reported attitudes and behaviors are unrelated to the composition of their social networks. We evaluate this assumption, developing a theory explaining how social networks influence the survey response by shaping the social desirability of various behaviors and attitudes. We apply our theory to the study of political participation, examining evidence from three observational datasets and an experiment conducted on a national sample. We demonstrate that non-voting respondents’ tendency to falsely report having voted is driven by political participation levels among their close friends and family. We show that this tendency can artificially inflate estimates of social influence. This study therefore suggests that survey researchers must account for social influence on the survey response to avoid biasing their conclusions.