Predicting child language impairment from too many variables: overinterpreting stepwise discriminant function analysis

Tallal (1999) defends an earlier presentation of 160 variables predicting classification of 59 children for specific language impairment (SLI) from the statistical criticisms of Zhang and Tomblin (1998). I present here 3 additional reasons why it is wrong to conclude from Tallal's original 1985 analysis and presentation that it is specifically temporal processing variables that underlie SLI. On the contrary, these data lend no support to this position.