Unbalanced scales inflate both alphas and empirical correlations.

Evidence

Having contraits on a test will not make people more motivated, although it probably slows down sloppy, automatic responding that continual protraits might spark. But it does let you use the invalid “agrees” from one half of the items to cancel out the invalid “agrees” from the other half, when the answers get summed. The same applies to invalid “disagrees.” You do have to balance scales, for uncontrolled yea-saying and naysaying can produce buckets of error variance. And all the while they will lead you to believe, through inflated alphas and puffed up empirical correlations with other unbalanced measures, that things are great. The road to Measurement Hell is paved with protrait items. — Bob Altemeyer, The Authoritarian Specter, 820 (#expands, via linked claim)

Categorization

• XREF: connects to: Scales with high Cronbach-α may still conceal several distinct latent factors. — Ruiqian Li and Paul Froese, The Duality of American Christian…, 84; Shared acquiescence can create spurious correlations between any two measures; Unbalanced scales can show spurious significance without true construct validity (+4 more)

Broader Connections

• Unbalanced scales inflate reliability alphas and correlations through acquiescence #expands (0.95) — Bob Altemeyer, The Authoritarian Specter, 820