Predicting an earthquake via an animals reactions or disappearances, in my experience, does not seem to be a reliable predictor. I have been living in California since the early 70's with house pets and earthquakes were a part of my Engineering position to protect from and mitigate any impact on my company. Over the past 50 so years, I have experienced a 100 or so noticeable earthquakes 3.0 and above, including the 1989 quake. None of our animals ever showed any pre-quake sensitivity. Our neighbors dogs (both sides) are very sensitive and will bark for hours at every real or potential skunk, possum, squirrel, shadow, bird, wind gust, flying leaves, sunspot, etc. A week ago, we were only a few miles away from a 3.+ epicenter only 4 miles deep. Normally, we don't even pay hardly any attention to a 3.5 or below, but being so close to shallow epicenter, we felt much more shaking energy than a typical 3 pointer we experience. We actually stopped the movie to watch hanging items sway back and forth. Of course, not a sound out of the neighborhood dogs before or after. Even our cat didn't wake up during the shaking.
On another note, I went to several Silicon Valley seminars where Geologist Jim Berkland spoke about earthquake potential and prediction methodology. When I spoke with him afterwards, he answered a number of questions openly and sincerely. Some of his methods were a little hard to accept but overall I think he may have been on to something that might have developed into a short range predictor. It's too bad that his conformist USGS bosses didn't like his out of the box thinking, which leaves us with no predicative capability and billions of tax dollars invested in holes in the ground.