This study explores inference about assets that have survived by avoiding poor performance. The greater is the commonality across assets in prior uncertainty about parameters, the more an asset's inferred expected return should depend on its having survived. If there is no commonality, then a surviving asset's average return can possess substantial sampling bias while nevertheless equaling the appropriate conditional expected return. Survival bias as typically computed generally makes too severe an adjustment to survivors, unless one assumes that expected returns on all assets, dead or alive, are equal to one common value that is completely unknown before observing returns data.