Skip to content
Scintria
ToolsGrant writerFunding
Start free
ScintriaStart free
Simulation Lab

Confounding and E-values

An observational study found an association. How strong would an unmeasured confounder have to be to explain it away? And what would a confounder you can describe do to the estimate?

Measure

More options
%
%

E-value

3.00

A confounder would need a risk ratio of 3.00 with both the exposure and the outcome to explain the estimate away.

E-value for the confidence limit

1.92

Weaker confounding than 1.92 could not move the interval to include no effect.

On the risk-ratio scale

1.80 (1.30–2.50)

Bias from the confounder you described

× 1.17

Estimate after removing it

1.54 (1.11–2.14)

Confounders strong enough to explain it away

2468468E = 3.00Confounder–exposure risk ratioConfounder–outcome risk ratio

Any confounder above and to the right of the curve could explain the whole association; below it, it could not.

For your protocol

For the observed risk ratio of 1.80 (95% CI 1.30 to 2.50), the E-value is 3.00: an unmeasured confounder associated with both the exposure and the outcome by a risk ratio of at least 3.00 each, above and beyond the measured covariates, could explain away the observed association, but weaker confounding could not. The E-value for the confidence limit closest to the null is 1.92. A confounder present in 40% of the exposed and 20% of the unexposed, with a risk ratio of 2.00 for the outcome, would move the estimate to 1.54 (1.11 to 2.14).

How this simulation works

The E-value (VanderWeele and Ding, 2017) is the minimum strength of association, on the risk-ratio scale, that an unmeasured confounder would need with both the exposure and the outcome, beyond the measured covariates, to fully explain away the observed association: E = RR + √(RR × (RR − 1)), with protective ratios inverted first. Odds ratios for common outcomes are converted by their square root, and hazard ratios by VanderWeele's approximation. The specific-confounder calculation uses the classic bias factor: (p₁(RR − 1) + 1) / (p₀(RR − 1) + 1), where p₁ and p₀ are the confounder's prevalence among the exposed and unexposed. The curve shows every pair of associations that would explain the estimate away.