Longitudinal sample

The bios in the longitudinal sample are those for users observed in every year 2015 through 2020. There are over a million accounts that meet this criterion. Another way to imagine the longitudinal sample is as the intersection of users from the six annual cross-sectional samples.

The longitudinal sample is valuable, because it allows for inference concerning individual change. We may observe how the same set of users described themselves at six points in time over half a decade. If a token rose in prevalence in this sample, we may safely conclude that more individuals added the token than removed it. Personally expressed identity may be observed changing within individuals. In this sample, about 52% of users edit their profile in any given year. 88% of users edited their bio at least once from 2015 through 2020.

To illustrate the sampling processes, assume we observed User A tweet once in 2016 and three times in 2018. User A’s profile biography from 2016 was included in the cross-sectional sample for 2016. One of the three tweets from 2018 was chosen at random, and User A’s profile biography accompanying that tweet was included in the cross-sectional sample for 2018. User A was not included in the longitudinal sample, because there were some years in which we observed no tweets for User A.