Gravar-mail: Bias analysis applied to Agricultural Health Study publications to estimate non-random sources of uncertainty