Assessing perceptions of cancer risk: does mode of assessment or numeracy matter?