Hello,
I have an CB-SEM model with more than 700 cases about the impact of language variations on purchase intention. It does include dummy variables for several experimental treatment groups (2 dummys for 3 treatments) and 4 latent constructs. There are two endogenous reflective constructs that are actually only measured by one item as they are straight forward: purchase intention and comprehensibility of the text (both 1 - 7 likert). Marketing literature sais that this is ok if the constructs are easy to grab like in this case.
The model works as predicted in the hypotheses (language variations have impact on comprehensibility and even a small one on purchase intention) and the model fit is good. Just the HTMT is higher than 0.9 two times and this is exactly when the two single item constructs are involved. It is fine with Fornell Larcker though (which is outdated, I know).
I just ask myself: is this due to the nature of single item constructs?
Best regards
Sebastian