Table 1.
List of selected attributes and levels, with definitions.
Fig 1.
Conjoint experiment: Example of screen seen by study participants.
Attribute levels were randomly assigned to create two hypothetical treatments. Participants were asked to express a preference for Treatment 1 or Treatment 2 and rank the likelihood of use of each treatment on a scale of 1 (not likely al all) to 10 (very likely).
Fig 2.
AMCE and marginal means for discrete preferences.
For our analysis of the conjoint experiment, we calculated both the AMCE (left panel) and the marginal means (right panel). The AMCE can be interpreted as indicators of “causal effect” coefficients, showing the average conditional effects for each attribute relative to the baseline. The marginal mean gives the overall favourability of an attribute with the mean support (0 to 1), where above 0.5 indicates a positive effect of the attribute on treatment preference, and below 0.5 indicates a negative effect. The figure plots the estimated values and the 95% confidence intervals for these estimates. These exact values are available in S2 Appendix. When confidence intervals do not overlap, we take this to suggest that the estimates are significantly different from one another statistically.
Fig 3.
AMCE and marginal means for preference rankings.
The ACME (left panel) and marginal means (right panel) were calculated for the ranked preferences of the infection treatments presented in the conjoint experiment. AMCE assesses the marginal effect of attributes on attitudinal measures whereas the marginal means gives the predicted mean ranking for each attribute level. The figure plots the estimated values and the 95% confidence intervals for these estimates. These exact values are available in S2 Appendix. When confidence intervals do not overlap, we take this to suggest that the estimates are significantly different from one another statistically.
Fig 4.
The left panel shows the average “likelihood of use” across the four framing conditions: (1) Phage therapy uses live viruses to kill cells, (2) Phage therapy uses viruses to selectively treat infections, (3) Phage therapy uses natural bacterial predators to treat infections, (4) Phage therapy uses friendly viruses to treat infections, where 1 is “not likely at all” and 7 is “very likely”. The figure compares the averages from those who completed the conjoint experiment before the framing experiment (red) and those who completed the framing experiment before the conjoint experiment (blue). The figure plots the estimated values and the 95% confidence intervals for these estimates. These exact values are available in S2 Appendix. When confidence intervals do not overlap, we take this to suggest that the estimates are significantly different from one another statistically.
Fig 5.
Marginal means by ordering of experiments.
The marginal means give the overall favourability of an attribute with the mean support (0 to 1), where above 0.5 indicates a positive effect of the attribute on treatment preference, and below 0.5 indicates a negative effect. This figure compares the marginal means from those who completed the conjoint experiment before the framing experiment (red) and those who completed the framing experiment before the conjoint experiment (blue). The figure plots the estimated values and the 95% confidence intervals for these estimates. These exact values are available in S2 Appendix. When confidence intervals do not overlap, we take this to suggest that the estimates are significantly different from one another statistically.
Fig 6.
267 Participants left responses to the open question; 213 of these were relevant to the question and categorised into the 5 answer categories shown in the graph. Each response was included in all relevant answer categories, as shown in S2 Appendix.