Contents

Bad control

Examples[edit]

Simple bad control[edit]

A simplified example studies effect of education on wages ${\displaystyle W}$.^[3] In this thought experiment, two levels of education ${\displaystyle E}$ are possible: lower and higher and two types of jobs ${\displaystyle T}$ are performed: white-collar and blue-collar work. When considering the causal effect of education on wages of an individual, it might be tempting to control for the work-type ${\displaystyle T}$, however, work type is a mediator (${\displaystyle E\to T\to W}$) in the causal relationship between education and wages (see causal diagram) and thus, controlling for it precludes causal inference from the regression coefficients.

Bad proxy-control[edit]

Another example of bad control is when attempting to control for innate ability when estimating effect of education ${\displaystyle E}$ on wages ${\displaystyle W}$.^[3] In this example, innate ability ${\displaystyle I}$ (thought of as for example IQ at pre-school age) is a variable influencing wages ${\displaystyle W}$, but its value is unavailable to researchers at the time of estimation. Instead they choose before-work IQ test scores ${\displaystyle L}$, or late ability, as a proxy variable to estimate innate ability and perform regression from education to wages adjusting for late ability. Unfortunately, late ability (in this thought experiment) is causally determined by education and innate ability and, by controlling for it, researchers introduced collider bias into their model by opening a back-door path ${\displaystyle E\to L\leftarrow I\to W}$ previously not present in their model. On the other hand, if both links ${\displaystyle E\to L}$ and ${\displaystyle I\to L}$ are strong, one can expect strong (non-causal) correlation between ${\displaystyle I}$ and ${\displaystyle E}$ and thus large omitted-variable biasif${\displaystyle I}$ is not controlled for. This issue, however, is separate from the causality problem.

References[edit]