How to Read: Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models [Reading Guide]

This guide will help you effectively read and understand "Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models" by Preacher and Hayes (2008), published in Behavior Research Methods.

First pass (5–10 min)

What to skim in this paper:

• Title: "Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models"
• First paragraph of the Introduction (Page 1): This sets the stage for the problem and the paper's purpose.
• Last paragraph of the Introduction (Page 1): This explicitly states the focus on multiple mediator models and the provision of computational resources.
• First few paragraphs of the "Discussion" section (Page 8): These summarize the main recommendations and findings.
• Last paragraph of the "Discussion" section (Page 9): This highlights the advantages of the proposed approach.

What to notice:

• The Problem: The paper addresses a gap in methodological literature regarding how to assess mediation when there are multiple simultaneous mediators. Existing methods for simple mediation are often inadequate or have limitations when extended to multiple mediators.
• The Contribution: The authors provide an overview of strategies for estimating and testing mediated effects in multiple mediator models, including methods for contrasting indirect effects. They advocate for specific approaches and offer practical guidance, including an illustrative example.
• Key Methods: The title immediately signals two main types of strategies: "asymptotic" (relying on large-sample theory) and "resampling" (like bootstrapping).
• Setting: The context is behavioral sciences, where mediation hypotheses are common.
• Main Recommendation (from skimming Discussion): Bootstrapping (especially bias-corrected versions) is the preferred method for assessing indirect effects in multiple mediator models due to its power and reasonable control over Type I error rates.
• Core Idea: It's important to consider multiple mediators in a single model, as this allows for assessing unique indirect effects and reduces bias from omitted variables.

Second pass (up to 1 hour)

Using SpatialRead in the Second Pass

• Use the Screenshot tool to capture figures, tables, diagrams, and formulae from the PDF and use AI for simple explanations.
• Ask: "Explain Figure X at a high level", "Compare Table 2 vs 3", or "What does the y-axis represent here?"
• Download SpatialRead →

What sections/figures to focus on:

• "Mediation in Behavioral Research" (Page 1-2):
  • Read the initial definitions of mediation and the distinction between direct and indirect effects.
  • Examine Figure 1B to understand the basic simple mediation model and its paths (a, b, c').
  • Note the limitations of the "causal steps strategy" and the advantages of the "product-of-coefficients approach" (Sobel test) and "bootstrapping" for simple mediation.
• "Strategies for Assessing Indirect Effects in Multiple Mediator Models" (Page 2-5): This is the core methodological section.
  • Read the advantages of specifying a single multiple mediation model (Page 3, bullet points).
  • Carefully study Figure 2B, which illustrates the multiple mediation model with j mediators. Understand how specific indirect effects (e.g., a₁b₁) and the total indirect effect (sum of aᵢbᵢ) are conceptualized.
  • "Causal Steps Approach" (Page 4): Understand why it's considered of "little utility" in the multiple mediator context.
  • "Product-of-Coefficients Approach" (Page 4): Understand that this approach involves calculating the indirect effect as a product of coefficients (e.g., ab) and its standard error. Note the assumption of normality and the recommendation to allow mediator residuals to covary. Skim Equation 1 to see the complexity of the variance calculation, but don't get bogged down in the derivation.
  • "Bootstrapping" (Page 5): This is the critical section. Understand the general procedure: repeatedly sampling from the data, estimating indirect effects, and building an empirical sampling distribution. Note the preference for percentile, bias-corrected (BC), and bias-corrected and accelerated (BCa) confidence intervals due to the skewed nature of indirect effect distributions.
• "Contrasting Indirect Effects in Multiple Mediator Models" (Page 6):
  • Understand the purpose of contrasts (e.g., comparing a₁b₁ vs. a₂b₂).
  • Note that contrasts compare the unique abilities of mediators, conditional on others in the model, not just their raw indirect effects.
  • Understand that both asymptotic (delta method, Equations 5 and 6) and bootstrapping approaches can be used for contrasts.
• "Illustrative Example" (Page 7-8):
  • Read the example to see how the methods are applied to a real-world scenario (socialization agents and job satisfaction).
  • Focus on how the total indirect effect and specific indirect effects are calculated and interpreted.
  • Examine Table 1 (Page 8). This table is crucial. It presents point estimates, standard errors, Z-values, and various confidence intervals (Percentile, BC, BCa) for both indirect effects and their contrasts. Pay attention to how the significance is determined (e.g., Z-value for asymptotic, CIs not containing zero for bootstrapping).

Key terms/questions to note:

• Simple vs. multiple mediation.
• Specific indirect effect (aᵢbᵢ) vs. total indirect effect (Σaᵢbᵢ).
• Direct effect (c').
• Asymptotic strategies (e.g., product-of-coefficients, delta method) and their normality assumptions.
• Resampling strategies (bootstrapping) and their advantages (nonparametric, handles skewed distributions).
• Types of bootstrap confidence intervals: Percentile, Bias-Corrected (BC), Bias-Corrected and accelerated (BCa).
• The meaning of contrasting indirect effects (comparing unique contributions).
• Why mediator residual covariances should be freed.

Third pass (1–5 hours)

Using SpatialRead in the Third Pass

• Highlight complex keywords, dense paragraphs, equations, or proofs and ask for a plain-English explanation that preserves notation.
• Use AI to explain the concepts in a way that is easy to understand.
• Ask for alternative phrasings or analogies to confirm understanding without changing technical meaning.
• Connect concepts with each other using AI to build a comprehensive understanding of the paper.
• Download SpatialRead →

Deep reading guidance:

• Methods (Page 4-6):
  • Re-read the "Product-of-Coefficients Approach" and "Bootstrapping" sections. If the mathematical details are relevant to your work, delve into the derivations of the standard errors (e.g., Equation 1 for total indirect effect variance, Equations 5 and 6 for contrast variance). Otherwise, focus on the conceptual implications of the normality assumption for asymptotic methods.
  • Thoroughly understand the bootstrapping procedure, including the number of resamples (5,000 recommended), how the empirical distribution is formed, and the calculation of percentile, BC, and BCa confidence intervals.
  • Pay close attention to the discussion on mediator residual covariances (Page 5) and why freeing them is recommended to avoid model misspecification and biased standard errors.
  • Re-read "Contrasting Indirect Effects" to fully grasp that contrasts compare the unique contributions of mediators, which is especially important when mediators are correlated.
• Results ("Illustrative Example", Page 7-8, and Appendix B):
  • Work through the "Illustrative Example" step-by-step, cross-referencing the text with Table 1. Understand how the Z-values and confidence intervals lead to conclusions about the significance of total, specific, and contrasted indirect effects.
  • Critically examine Appendix B (Page 12-13). This appendix provides the detailed output from the example analysis. Study the "NORMAL THEORY TESTS FOR INDIRECT EFFECTS" section to see the asymptotic results, and then compare them with the "BOOTSTRAP RESULTS FOR INDIRECT EFFECTS" section, noting the differences between Percentile, BC, and BCa confidence intervals. This will solidify your understanding of how to interpret the output from these analyses.
• Discussion and Recommendations (Page 8-9):
  • Reflect on the authors' strong recommendation for bootstrapping and the reasons behind it (e.g., simulation evidence from other studies).
  • Consider the implications of mediator collinearity and how specific indirect effects and contrasts should be interpreted in that context.
  • Review the "Extensions" section for potential future research directions or applications of these methods (e.g., latent variables, mediated moderation, multiple outcomes).
• Limitations: While the paper primarily focuses on advocating for specific methods, consider the inherent limitations of mediation analysis itself (e.g., the assumption of causal order mentioned on Page 1). The computational intensity of bootstrapping, though less of an issue now, was a practical limitation at the time of publication.

Replication pointers if applicable:

• The paper provides an "Illustrative Example" (Page 7-8) with specific variables and a sample size (N=141).
• Appendix A (Page 11) includes detailed syntax for two common statistical modeling environments, demonstrating how to implement the bias-corrected bootstrapping for specific indirect effects and contrasts.
• Appendix B (Page 12-13) presents the full output of the illustrative example, which can be used to verify your understanding of the results if you attempt to replicate the analysis.
• The authors explicitly state that the macros used for the example (for other statistical packages) are available online, providing the necessary resources to apply these methods to your own data. The example details the variables used (helpful, jobsat, politics, people, perform) and the number of bootstrap resamples (5,000).