Anchoring is a well-known phenomenon . However, I just ran an experiment where the effect seemingly backfired. Basically, I asked people to estimate the number of jelly beans in a jar. In my anchor condition, people were given an estimate produce by the Jelly Bean Counter 5000 computer system which was purported to be highly accurate, then they were allowed to see the jar. They were then asked to estimate the number of jelly beans.
People who were not anchored were allowed to see the jar of jelly beans first, then estimate the count. Then, they were showed Jelly Bean Counter 5000's estimate. They were then allowed to re-estimate the number of jelly beans in the jar.
In all cases, Jelly Bean Counter 5000 provided the exact number of jelly beans in the jar.
The results are shown in the figure below. Basically, people who were anchored with the Jelly Bean Counter 5000 number were about 62% accurate. People who made their own estimate first (self-anchoring) were initially about 46% accurate, but after seeing Jelly Bean Counter 5000's estimate, they increased to about 72% accurate.
I'm at a loss to explain why people exposed to an accurate estimate (Jelly Bean Counter 5000) ended up being less accurate than people who formed their own inaccurate opinion first. Can anybody provide any alternative explanations? Are there any theories that might be applicable?
 Tversky, A., & Kahneman, D. (1974). Judgment under Uncertainty: Heuristics and Biases. Science, 185(4157), 1124-1131.