Can cheap talk mitigate anchoring bias? Evidence from a choice experiment