Extended Data Table 2 Accounts Shown in Chronological Feed: Llama 3 Annotations
- Regression estimates of the intention-to-treat (ITT) and placebo effects on the chronological feed’s political content measured using Llama 3 annotations (see SI Section S1.5.1 for details of measures and SI Section 2.7 for the specification, namely, equation 10). The unit of observation is X post. Sample size is reported in the Table. Col. 1, 4, and 7 report estimates without controls, examining effects on Conservative content (1), political activist content (4), and Conservative political activist content (7). Col. 2, 5, and 8 add controls for pre-treatment X use frequency and purpose. Col. 3, 6, and 9 further include X use, demographic, and political controls. All specifications control for respondents’ initial feed settings. For each outcome, coefficients β1 and β2 capture content changes in the chronological feed between pre- and post-treatment periods, while \({\delta }_{1}\) and \({\delta }_{2}\) serve as placebo estimates validating the randomization. The sample is restricted to posts shown in the chronological feed setting. Standard errors are clustered by respondent. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively (all tests are two-sided).
