Should I ask vignettes questions of all people or a random subset?

1. Chopit works well even when vignettes are asked of only a random subset of respondents, but chopit is only effective if your survey includes variables ($V_i$) that can predict the threshold variation across respondents. Chopit will be more effective when $\gamma_1$ (the effect of $V$ on the thresholds) is stable across respondents.
2. Our nonparametric method only works for respondents who have been asked both vignettes and self-assessments. It can be applied to the subset that have been asked both, or both could be asked of all respondents. The advantage of this method is that variables $V_i$ that are necessary to predict the thresholds in chopit are not needed here. So if you are sure you don't have $V_i$, ask vignettes of all respondents.
3. These decisions can be made best by evaluating actual pretest data.
4. Adaptive survey methods might be highly efficient here, whereby respondents are offered new vignettes depending on their previous vignette answers. In this situation, a CATI system could locate vignettes to the left and right of a respondent's self-assessment by bisecting the continuum using a small subset of a large list of a priori rank ordered list of vignettes. Each respondent would then be asked a relatively small number of vignettes. Be aware, however, that this strategy would require a new statistical method be developed.