The mean response latency difference (MLD) across two conditions operationalizes semantic priming, Stroop interference, task switching, and implicit association effects. We show that correlations involving MLD measures are biased by the positive association between mean latency and variance of latency. Analytic expressions for the correlation of the MLD with (a) its two components, (b) average of its components, (c) external criterion, and (d) another MLD provide a framework to illustrate correlational biases. Scale invariance and validity maximization are proposed as defining properties of admissible latency contrast measures. As MLDs violate scale invariance, we investigated scale free measures to contrast latency distributions. In comparison to the MLD, the recommended scale invariant measures are statistically powerful, have reduced correlational biases, and exhibit superior validity.