simInfl: Simulates significance reversals and calculates their...

In anspiess/reverseR: Linear Regression Stability to Significance Reversal

Simulates significance reversals and calculates their influence parameters

Description

This function simulates linear regressions and stores the parameters and influence measures of all simulations that resulted in LOO significance reversal, developed for research purposes.

Usage

simInfl(x = 1:10, slope = 0.02, intercept = 1, error = 0.05, nrev = 1000, ...) 

Arguments

x	the x values to be supplied to lmExact.
slope	the slope β_1 to be supplied to lmExact.
intercept	the intercept β_0 to be supplied to lmExact.
error	the \varepsilon value to be supplied to lmExact.
nrev	the number of desired significance reversals.
...	other parameters to lmExact and lmInfl.

Details

Loops over an undefined number of EXACT regressions (lmExact) with incrementing random seeds, stores all models and in case of significance reversal, parameters and influence measures (lmInfl). The simulation terminates when nrev reversals are counted.

Value

A list with the following two items:

models	the linear models of all reversals.
mat	the stored matrix with the resulting parameters and influence measures for all nrev reversals.

Author(s)

Andrej-Nikolai Spiess

Examples

## Example with slight slope, intercept = 0.5 and 10 reversals.
res <- simInfl(x = 1:10, intercept = 0.5, slope = 0.02, error = 0.05, nrev = 10)

## Plot Cook's D versus delta-P values
## and insert common cut-off.
plot(res$mat[, "cook.d"], res$mat[, "dP"], pch = 16, cex = 0.5,
     xlab = "Cook's D", ylab = "delta-P")
thresh <- qf(0.5, 2, 8)  # threshold value for Qf(0.5, npar, df)
abline(v = thresh, col = "darkred", lwd = 2)  

## Plot dfbeta slope versus delta-P values
## and insert common cut-off.
plot(res$mat[, "dfb.Slope"], res$mat[, "dP"], pch = 16, cex = 0.5,
     xlab = "dfbeta Slope", ylab = "delta-P")
thresh <- 2/sqrt(10)  # 2/sqrt(N)
abline(v = thresh, col = "darkred", lwd = 2)  

## Plot dffits versus delta-P values
## and insert common cut-off.
plot(abs(res$mat[, "dffit"]), res$mat[, "dP"], pch = 16, cex = 0.5,
     xlab = "dffits", ylab = "delta-P")
thresh <- 2 * sqrt(2/10)  # 2 * sqrt(nPar/N)
abline(v = thresh, col = "darkred", lwd = 2)  


## More illustrative with more reverser samples!
## Example with slight slope, intercept = 0.5 and 10 reversals.
res <- simInfl(x = 1:10, intercept = 0.5, slope = 0.02, error = 0.05, nrev = 200)
plot(res$mat[, "cook.d"], res$mat[, "dP"], pch = 16, cex = 0.5,
     xlab = "Cook's D", ylab = "delta-P")
thresh <- qf(0.5, 2, 8)  # threshold value for Qf(0.5, npar, df)
abline(v = thresh, col = "darkred", lwd = 2)  

anspiess/reverseR documentation built on May 14, 2022, 9:43 a.m.