While one of the purposes of this article was to show the flexibility and benefits of the MC approach to determining sample size, there is a drawback: it requires
users to know more about their studies’ variables than traditional methods. Investigators have to specify not only α and power, but they also have to specify values
for all the variables’ relations. Thus, Cohen’s (1992) concern about researchers not knowing appropriate effect size values for their particular field is amplified if
they have to know how all the variables relate to each
other. In the best situation, scholars would select the
model’s values from theory or previous research. In the
complete absence of any theoretical expectations,
Maxwell (2000) suggested starting with the assumption
that all zero-order correlations are .30, then changing
the values to see how it influences the required sample
size (i.e., sensitivity analysis). Using values of .30 gives
R2 values around 0.14 (2 predictors) to 0.24 (10
predictors), which may or may not be appropriate for a
study.