- Model selected by hierarchical LRT is JC
- Model selected using the Akaike Information Criterion is SYM+G
- Each selected a different model. Hierarchical LRT will test if the addition of extra parameters improve significantly (depending on the input confidence level) the fit of our data, it starts from the simplest model JC and adds one extra parameter, model K80 (transition and transversion have different rates). If this extra parameter has not a better fit (likelihood) hierarchical LRT will stop at this point and return the simplest model. In our case, we can see that the log likelihood of the JC model is equal to the one of the K80 model. Hierarchical LRT will than stop at this step (local optimum) and return JC model as best solution, it will never reach the step of testing the model proposed by Akaike based method.
- In this case we should use the AIC solution as hierarchical LRT seems to get stacked in a local optimum. But anyway JModelTest also propose you the selection of model that gets rid of all uncertainty.