site_outliers {bulkQC}

Identifies site level outliers

Description

Discovers potential site level outliers by using unadjusted and adjusted regression models and standardized difference calculations.

Usage

site_outliers(d0, exclude = c("pid"), siteID = "site", covs = c("age"), threshG = 0.001, 
thresh2 = 0.05, threshS = 0.5, n_uniq = 10, n_dec = 4, n_decS = 2)

Arguments

d0

A data frame with columns as variables and rows as observations

exclude

A vector of names of variables to exclude in outlier identification

siteID

The name of the variable in the data frame that identifies sites

covs

A vector of covariates to adjust for in the adjusted regression models

threshG

P-value threshold for global test equal means across sites

thresh2

P-value threshold for comparison of reference site vs. all other sites

threshS

Standardized difference threshold above which a site difference is deemed meaningfully large

n_uniq

Number of unique observations of a variable needed for outlier identification to be performed

n_dec

Number of decimals to display for p-values in output

n_decS

Number of decimals to display for standardized differences in output

Details

The function compares the distribution of a given variable across sites by first conducting a global test of equal means (without and with adjustment for covariates of interest). Among those variables where the null hypothesis of equal means across sites is rejected, the function then compares each site vs. all other sites using unadjusted and adjusted comparisons. The unadjusted comparisons include a two-sample t-test with equal variance and a standardized difference calculation. The adjusted comparisons include a linear regression model with an indicator variable for reference site and user-specified covariates, and an adjusted standardized difference calculated as the model coefficient for site divided by the model estimated root mean squared error.

Value

overall

A matrix with rows as variables where global test of equal means is rejected and columns as the corresponding p-values from the unadjusted and adjusted statistical tests

sitewise_P

For the variables identified by the global tests (columns), the unadjusted p-values (from two-sample t-test) comparing each site to all other sites (rows). Values above threshold printed as missing.

sitewise_P_adj

For the variables identified by the global tests (columns), the adjusted p-values (from linear regression model) comparing each site to all other sites (rows). Values above threshold printed as missing.

sitewise_StDf

For the variables identified by the global tests (columns), the unadjusted standardized differences comparing each site to all other sites (rows). Values below threshold printed as missing.

sitewise_StDf_adj

For the variables identified by the global tests (columns), the adjusted standardized differences comparing each site to all other sites (rows). Values below threshold printed as missing.

References

Yang D, Dalton JE. A unified approach to measuring the effect size between two groups using SAS. 2012;6

Examples

data(iris)
iris2 = iris
iris2$temp = rnorm(dim(iris2)[1]) #for covariate adjustment
site_outliers(iris2, site="Species", covs=c("temp"))