Placebo. There are biological bases of it (I don't believe in soul). Find these bases, study them, make a model of them. Then use proxy variables to measure it instead of trying to eliminate it statistically. Predict it in studies to avoid the need of placebo groups (and possibly of double blind methodology). Also, after it is completely measurable and its mechanisms are understood, if (very hypothetical) it has a really substantial effect, just use it to help treat patients.