Issues in multivariate assessment of a large-scale behavioral program.

Herrnstein et al. (1979) wrote a think-piece, not an experiment. They looked at big school programs that try to measure many skills at once.

The authors asked: what hidden traps appear when you test hundreds of kids across many classrooms?

The paper lists problems, not numbers. It warns that community pressure, teacher stress, and unseen social forces can twist your data.

They say: if you only count easy behaviors, you miss the real story.

McMillan (1973) set the stage. That earlier paper told us to watch for sequence effects in small single-subject designs. J et al. widen the lens to whole schools.

Cameron et al. (1996) later proved one of the worries. They showed peer sociometrics bounce around too much to trust for one kid. This backs the 1979 warning that hidden variables wreck large data sets.

Cohen et al. (1993) and Fàbregues et al. (2022) keep the thread alive. Both shout that tools must fit real classrooms and mixed-methods must be clear. The 1979 paper reads like their early blueprint.

Next time you plan a district-wide social-skills screen, add cheap proxy measures for teacher stress and classroom climate. Track them along with student scores. If the hidden numbers drift, you will know your main data may be noisy before you make big decisions.