When researchers say that they are “controlling for” something, what they mean is they are removing the influence of that thing (a variable) mathematically, in order to better assess the relationship between other variables.
For instance, in our study on B2B marketer compensation, we asked participants to name the salary increase it would take to convince them to move to a new employer – their “Defection Price.” They also shared their current job tenure and total earnings (salary plus bonuses).
We wanted to know whether people who had been with their companies longer (greater tenure) would require a higher percentage increase to leave and go to a new company. But, we had also observed that people who had been in their jobs longer tended to also be higher paid.
We tested the relations of both tenure and compensation with Defection Price, and both were found to be positively correlated with Defection Price. As either factor increases, Defection Price also increases. To isolate the impact of longer tenure on the defection price, we conducted a further correlation analysis with defection price and tenure, this time “controlling for” (also termed, “holding constant”) a person’s compensation.
As shown in the table below, the correlation between tenure and Defection price remained statistically reliable after controlling for compensation, but it was reduced substantially, giving us a much more accurate picture of how tenure relates to Defection Price.
In the real world, it is often necessary to control for multiple factors at once. Even then, there is generally a substantial portion of the variability that occurs that remains unaccounted for.
