. 2006 May 4;6:21. doi: 10.1186/1471-2288-6-21

Table 4.

Comparison of methods for handling measurement error in a real dataset using a repeat FFQ on a 33% sub-sample, with total energy intake as potential confounder.

	Without adjustment for total energy intake		With adjustment for total energy intake		With adjustment for total energy intake

	Ignoring all measurement error	Regression calibration	Ignoring measurement error	Regression calibration assuming energy intake perfectly measured	Regression calibration allowing for measurement error in energy intake
$\hat{β}$ ₀(se)	3.69 (.02)	3.64 (.02)	4.04 (.05)	4.00 (.05)	4.19 (.08)
$\hat{β}$ ₁(se)	.48 (.20)	.38 (.32)	.42 (.20)	.32 (.32)	.32 (.33)
$\hat{β}$ ₂(se)	.41 (.03)	.51 (.04)	.45 (.03)	.56 (.04)	.57 (.04)
$\hat{β}$ ₃(se)	.88 (.27)	1.04 (.40)	.95 (.27)	1.12 (.39)	1.14 (.39)
$\hat{β}$ ₄(se) (× 1000)	-	-	-.15 (.02)	-.16 (.02)	-.24 (.03)