Hazards in Choosing Between Pooled and Separate Variances t Tests