Gravar-mail: Estimating causal effects from observational data with a model for multiple bias