compareEase: Compare EASE of correlated sets of estimates

In EmpiricalCalibration: Routines for Performing Empirical Calibration of Observational Study Estimates

Compare EASE of correlated sets of estimates

Description

Compare EASE of correlated sets of estimates

Usage

compareEase(
  logRr1,
  seLogRr1,
  logRr2,
  seLogRr2,
  alpha = 0.05,
  sampleSize = 1000
)

Arguments

logRr1	A numeric vector of effect estimates generated using the first method on the log scale.
seLogRr1	The standard error of the log of the effect estimates generated using the first method.
logRr2	A numeric vector of effect estimates generated using the second method on the log scale.
seLogRr2	The standard error of the log of the effect estimates generated using the second method.
alpha	The expected type I error for computing confidence intervals and p-values.
sampleSize	The number of samples in the bootstraps.

Details

Compare the expected absolute systematic error (EASE) of two sets of estimates for the same set of negative controls.

Important: the two sets of estimates (logRr1 + seLogRr1 and logRr2 + seLogRr2) should be in identical order, so that for example the first item in each vector corresponds to the same negative control.

Value

A data frame with 4 columns: the point estimate, confidence interval lower bound, and upper bound for the difference between EASE in the two sets of negative controls, and a p value against the null hypothesis that the EASE is the same for the two sets.

The data frame has two attributes: ease1 and ease2, providing the EASE estimates (and confidence intervals) for the two sets, computed using bootstrapping. Note that these estimates may somewhat different from those generated using computeExpectedAbsoluteSystematicError, because a different approach is used to compute the confidence interval. The approach used here will more closely align with the computation of the difference in EASE.

Examples

# Simulate results of first method:
ncs1 <- simulateControls(n = 50)

# Simulate second method to be more biased:
ncs2 <- ncs1
ncs2$logRr <- ncs2$logRr + rnorm(nrow(ncs2), mean = 0.1, sd = 0.1)

delta <- compareEase(
  logRr1 = ncs1$logRr,
  seLogRr1 = ncs1$seLogRr,
  logRr2 = ncs2$logRr,
  seLogRr2 = ncs2$seLogRr
)
delta
attr(delta, "ease1")
attr(delta, "ease2")

EmpiricalCalibration documentation built on Sept. 30, 2024, 9:12 a.m.