Agreed re:being weaselly about your number. Given that someone has to say a number first (absent weird note-trading schemes), not saying a number first is (game theoretically) tantamount to saying "I'm hoping you'll unwittingly make an offer better than what I believe I could credibly demand". In complex multi-dimensional negotiations this can be totally reasonable (when there are 20 different key terms to negotiate, saying "I think you'll make an anchoring mistake first" is not at all unreasonable). Salary negotiations are usually not multidimensional since the position/title is usually fixed and the official story is that you'll work at some fixed level of output commensurate with that title and not connected to pay (so you can't say "I'll accept $X but I'll probably slack a lot").
From the employer's position the "I think you're likely to botch it" bet is not a bad one: employees don't know the employer's value function for the work, but the employer can have a pretty decent model of the employee's value function for steady income (tip: eliminate your short-term need for steady income and many negotiations become a lot easier).
For the employee, hoping for an accidental or poorly-informed over-offer is rarely justified. Employers usually have many past negotiation experiences and many sources of information for what other employers are paying, whereas employees usually have few. Similarly, the employer's side of the exchange is primarily liquid and easily quantified (cash+benefits) whereas the employee's side is poorly quantified and illiquid (the employer usually can't directly "sell out" your past labor and partially recoup if he decides it was worth less than he expected).
Expecting employers to err on the side of overpaying with any sort of frequency seems theoretically hard to justify, and that is certainly not in line with my experience. Unless you know the employer wants you specifically (as opposed to any instance of the class you represent), I think it's almost always better for employees to open with what they consider a high number and let the employers try to talk them down.
From the employer's position the "I think you're likely to botch it" bet is not a bad one: employees don't know the employer's value function for the work, but the employer can have a pretty decent model of the employee's value function for steady income (tip: eliminate your short-term need for steady income and many negotiations become a lot easier).
For the employee, hoping for an accidental or poorly-informed over-offer is rarely justified. Employers usually have many past negotiation experiences and many sources of information for what other employers are paying, whereas employees usually have few. Similarly, the employer's side of the exchange is primarily liquid and easily quantified (cash+benefits) whereas the employee's side is poorly quantified and illiquid (the employer usually can't directly "sell out" your past labor and partially recoup if he decides it was worth less than he expected).
Expecting employers to err on the side of overpaying with any sort of frequency seems theoretically hard to justify, and that is certainly not in line with my experience. Unless you know the employer wants you specifically (as opposed to any instance of the class you represent), I think it's almost always better for employees to open with what they consider a high number and let the employers try to talk them down.