The last example in Section 8.5.5 dealt with comparing males to females regarding the desired number of sexual partners over the next 30 years. Using Student's T, we fail to reject which is not surprising because there is an extreme outlier among the responses given by males.
If we simply discard this one outlier and compare groups using Student's T or Welch's method, what criticism might be made even if we could ignore problems with non-normality?