Module 6: Regression Bootstrapping

Purpose

Bootstrapping provides nonparametric standard errors and confidence intervals for regression estimates. SmartPLS supports regression bootstrapping for significance assessment.

Learning Objectives

After this module, learners should be able to:

• Explain why bootstrapping is used.
• Select appropriate bootstrap settings.
• Interpret bootstrap confidence intervals.
• Report bootstrapped coefficients and significance.

What Bootstrapping Does

Bootstrapping repeatedly draws samples from the original dataset with replacement and re-estimates the model. The distribution of bootstrap estimates is used to estimate standard errors, t-values, p-values, and confidence intervals.

Recommended Workflow

For initial checking:

1,000 bootstrap subsamples

For final reporting:

10,000 bootstrap subsamples

Use a fixed seed for reproducibility.

Confidence Intervals

If the confidence interval for a coefficient does not include zero, the coefficient is statistically significant at the chosen confidence level.

Example:

b = 0.42, 95% CI [0.18, 0.66]

This coefficient is positive and statistically significant.

Percentile vs BCa

SmartPLS documentation lists percentile, studentized, and BCa bootstrap options. Percentile bootstrap is the default recommendation. BCa can be considered when the bootstrap distribution is clearly non-normal.

Reporting Template

Bootstrapping with 10,000 subsamples was used to assess the significance of the regression coefficients. The effect of X on Y was positive and significant (b = ..., t = ..., p = ..., 95% CI [..., ...]).

Lab

Complete Lab 4: Run regression bootstrapping.

Next Module

Continue to Module 7: Reporting Regression Results.