Two-year versus one-year Head Start program impact: addressing selection bias by comparing regression modeling with propensity score analysis