Control variables and causal inference: a question of balance

A common motivation for adding control variables to statistical models is to reduce the potential for spurious findings when analyzing non-experimental data and to thereby allow for more reliable causal inferences. However, as I show here, unless all potential confounding factors are included in an analysis (which is unlikely to be achievable with most real-world data-sets), adding control variables to a model in many circumstances can make estimated effects of the variable(s) of interest to the researcher on the dependent variable less accurate. Due to this fact, in some circumstances omitting control variables, even those that affect the dependent variable and are correlated with the variable(s) of interest, may allow for more accurate estimates of the effect(s) of the variable(s) of interest.