] replicated 20 experiments and found a high level of concordance between estimates obtained from MTurk-based samples and national probability samples. Along with the supporting evidence confirming the utility of web-based convenience samples in experimental studies, we carefully implemented several methods to improve our ability to generalize causal inferences across different settings and contexts.
Our study methods were designed to enhance our understanding of the theory-based causal process and ecological validity . A strong replication pattern across samples will improve the credence of the findings.Our findings should be interpreted in light of the limitations imposed by the study designs and settings.
Second, for our path analyses, we generated dummy-coded exogenous variables to examine persuasion pathways from the conditions and also from the 5 thematic-level variables, as shown in. However, we did not compute dummy-coded variables for the message-level constructs in the path model. This finding was partially due to the sample size. If dummy coding were generated at the message level, the model would entail 25 exogenous variables, leading to overfitting problems and unspecified model fits.
Finally, although most of our analyses were based on a sample of over 900 participants, the final path model was based on a subset of the sample with children aged between 9 and 14 years, who did not receive any HPV vaccines at the moment, to reflect the age range for the HPV vaccination most recommended by the CDC; that is, we did not ask a hypothetical question to all participants .