The methodology for the semi-finals model is described here. The model is 90% accurate on ranking, with a margin of error of about 3 percentage points. Since the mid-line probability is about 42%, the model is unsure of Amber, Janelle, Adriana, and Breanna. It is reasonably sure that Tenna, Zoanette, and Aubrey will not advance, and it is fairly certain Angela, Candice, and Kree will advance.
The model finds the most probable Top 5 women are Angela, Candice, Kree, Amber, and Janelle.
Aubrey is below the probability where nobody has ever advanced. Only two people in the model’s history (seasons 6-11) advanced with a probability less than 40%, Michael Sarver and Kris Allen. That being said, probabilities being what they are, such an event should eventually happen, so her fans should have a little hope. I certainly am puzzled as to why her reviews were so middling and her Dialidol-measured popularity so low.
To the extent that the rules have changed this year (which is not inconsiderable), there is some doubt as to whether Dialidol will still be very reliable. I admit I’m a bit nervous about the notion of thousands of Worsters raining 50 votes on Zoanette Johnson, their pick for the worst girl (and who could argue?). Vote for the Worst doesn’t necessarily have a great track record of actually keeping people on, but the effect of the supervote is unclear.