Science without (parametric) models: The case of bootstrap resampling