Crafting a Pre-Registration Document for a Triadic Experiment

1. The What, Why, and How of Pre-registration

Before we begin drafting, let's solidify our understanding of what a pre-registration is and why it's so critical for distinguishing planned, confirmatory hypothesis testing from unplanned, exploratory discovery.

As the video outlines, a pre-registration locks in your study plan, hypotheses, data collection procedures (including stopping rules), variable definitions, and analysis plan. This process protects against our own natural cognitive biases and the temptation to chase statistical significance. It transforms your research plan into a formal commitment.

2. A Systematic Approach: The Researcher Degrees of Freedom Checklist

An effective pre-registration must be specific, precise, and exhaustive. It needs to anticipate and constrain the myriad choices a researcher can make throughout a study's lifecycle. To ensure we are thorough, we will use a comprehensive checklist of so-called "researcher degrees of freedom" (DFs) as our guide.

This checklist provides the "why" for each section of the pre-registration form. As we fill out the form, we are systematically closing the loopholes identified in this table.

3. Drafting the Pre-registration Document

We will now walk through the standard "OSF Prereg" template, section by section. For each item, we'll formulate a precise entry for our vee-triad experiment, noting which DFs from the Wicherts et al. checklist we are addressing.

3.1. Study Information

This section sets the stage and locks in your primary research questions.

• Title: "The Causal Effect of Living Arrangements on Relationship Satisfaction in a Polyandrous Vee Triad: An N-of-1 Crossover Experiment."
• Description: "This study uses a within-subject crossover design to investigate the causal impact of two different living arrangements ('Nesting' vs. 'Separate') on the relationship satisfaction of a single, established polyandrous vee triad. The primary goal is to identify the arrangement that maximizes overall triadic satisfaction. Secondary analyses will explore potential mediating psychological pathways, such as perceived autonomy and security."
• Hypotheses: This is where we constrain T1 (Conducting explorative research) and T2 (Studying a vague hypothesis). Hypotheses must be directional and testable.
  1. H1 (Primary): The 'Nesting' living arrangement will cause a higher mean Triadic Satisfaction Index score compared to the 'Separate A' arrangement. This is a directional hypothesis.
  2. H2 (Secondary): The 'Nesting' arrangement will cause a higher mean Security Scale score for all three members compared to the 'Separate A' arrangement. This is a directional hypothesis.
  3. H3 (Secondary): The 'Separate A' arrangement will cause a higher mean Autonomy Scale score for all three members compared to the 'Nesting' arrangement. This is a directional hypothesis.

3.2. Design Plan

Here we specify the architecture of the experiment.

• Study type: Experiment.
• Blinding: This addresses C2 (Insufficient blinding).

  "No participant blinding is involved, as the participants are aware of their living arrangement. However, the data analyst will be blinded to the condition labels (i.e., 'Condition A' vs. 'Condition B') during the primary analysis of the main outcome. The key linking labels to the actual arrangements will only be revealed after the primary confirmatory analysis is complete."

• Study design: This addresses D1 (Creating multiple manipulated IVs). We are precise about our design.

  "A single-case (N-of-1) AB/BA crossover design with two sequences and two periods. The triad is the single unit of study. The design is within-subject, with all members experiencing both conditions."

• Randomization: This addresses C1 (Failing to randomly assign).

  "The sequence of conditions will be randomized. The order (AB vs. BA, where A='Nesting' and B='Separate A') will be determined by a single, fair coin toss conducted prior to the start of Period 1. Heads will correspond to the AB sequence; tails to the BA sequence."

3.3. Sampling Plan

This section is critical for preventing optional stopping and data dredging.

• Data collection procedures:

  "Participants will be the three members of the pre-existing vee triad. Data will be collected daily via a structured electronic diary administered each evening for a total of 24 weeks. Each experimental period will last 12 weeks."

• Sample size:

  "The sample size is fixed by the study design. For each of the three participants, we will collect data for 84 days per period (12 weeks x 7 days), for a total of 168 observations per participant across the study."

• Sample size rationale: This addresses D6 (Failing to conduct a power analysis).

  "As this is an N-of-1 study, a traditional power analysis for determining the number of subjects is not applicable. The period length of 12 weeks was chosen to be sufficiently long to allow for adaptation to the new living arrangement and for the outcome measures to stabilize, while remaining practical for the participants. This duration is longer than typical mood-stabilization periods noted in behavioral literature (e.g., 4-6 weeks)."

• Stopping rule: This is our primary defense against C4 (Intermediate significance testing) and D7 (Failing to specify the sampling plan). We will insert the rule developed in the last lesson.

  "Data collection within each 12-week experimental period will follow a group sequential design with an O'Brien-Fleming-like alpha-spending function for efficacy and a Pocock-like beta-spending function for futility (overall two-sided $\alpha =0.05$, $\beta =0.20$). We will perform two interim analyses at 4 and 8 weeks, plus a final analysis at 12 weeks. The exact p-value boundaries for stopping for efficacy or futility at each look will be calculated using the rpact package in R prior to the experiment and registered as part of this document. The decision to perform an interim analysis is fixed by this schedule and is not contingent on the data observed."

3.4. Variables

Precision here is key to prevent D3, D4, A5, A6, and A7.

• Manipulated variables:

  "The single manipulated variable is Living Arrangement. It has two levels:

  1. Nesting: All three individuals reside in the same household full-time.
  2. Separate A: The pivot and one partner (Partner A) reside in the primary household; the second partner (Partner B) resides in a separate dwelling."

• Measured variables:

  Primary Outcome:

  • Triadic Satisfaction Index.
    Secondary Outcomes / Mediators:
  • Individual Satisfaction Score (for each of the 3 members).
  • Autonomy Scale Score (for each of the 3 members, using the [Specify validated scale, e.g., 'Relationship Autonomy Scale']).
  • Security Scale Score (for each of the 3 members, using the [Specify validated scale, e.g., 'Relationship Security Scale']).
    Covariates:
  • Household Financial Stress (measured weekly via a single 1-10 item).
  • External Social Commitments (measured weekly as total hours)."

• Indices:

  "The Triadic Satisfaction Index is the primary outcome and will be calculated as the unweighted arithmetic mean of the three daily Individual Satisfaction Scores. The Individual Satisfaction Score will itself be the mean of three 7-point Likert items assessing global, emotional, and logistical satisfaction with the relationship on that day."

3.5. Analysis Plan

This section is the analytical core of the pre-registration, constraining A1 through A15 and preventing R6 (HARKing).

• Statistical models:

  "To test H1, we will use a linear mixed-effects model.

  • Outcome: Triadic Satisfaction Index.
  • Fixed Effects: condition (Separate A vs. Nesting), period (1 vs. 2 to model temporal trends), sequence (AB vs. BA to test for carryover effects).
  • Random Effects: A random intercept for day to account for non-independence of observations within a given day.
    The main test for H1 is the statistical significance of the condition coefficient. The same model structure will be used for secondary hypotheses H2 and H3, with Autonomy and Security scores as the outcomes."

• Transformations:

  "No data transformations are planned. If visual inspection of model residuals reveals severe non-normality or heteroscedasticity, we will report results from both the original model and a model run on rank-transformed data. The original model will remain the primary confirmatory test."

• Inference criteria: This constrains A15 (Choosing inference criteria).

  "We will use a p-value criterion of $\alpha =0.05$ (two-tailed) for the primary hypothesis. The sequential analysis boundaries will be used as described in the stopping rule. For the two secondary hypotheses (H2, H3), we will apply a Bonferroni correction, using an alpha of 0.025 for each."

• Data exclusion: This constrains D5 and A12 (participant exclusion).

  "Data from any day where fewer than 50% of the daily diary items are completed by all three members will be excluded from the analysis. Participants are the investigators and are considered fully engaged; no other exclusion criteria will be applied."

• Missing data: This constrains A1 (handling missing data).

  "We will use Full Information Maximum Likelihood (FIML) as part of the lme4 package in R to handle missing data, which is robust to data missing at random. No other imputation methods will be used."

• Exploratory analysis: This is the designated "sandbox" to prevent HARKing.

  "Any analysis not explicitly defined above will be considered exploratory. We plan to explore potential moderation of the condition effect by the weekly 'Household Financial Stress' covariate. These exploratory findings will be clearly labeled as such in any resulting manuscript and will be used to generate hypotheses for future studies."

4. Finalizing and Submitting

Once drafted, the document is submitted to a public registry like the Open Science Framework. The platform time-stamps it and makes it immutable (read-only). This provides verifiable proof that your plan was in place before the experiment concluded.

An important feature highlighted is the ability to embargo your pre-registration. This keeps it private for a set period (up to four years on OSF), protecting you from being "scooped," while still maintaining the integrity of the time-stamp. You can also create anonymous links for blinded peer review.

Conclusion

In this lesson, you have learned how to synthesize the principles of robust research into a single, powerful document: a pre-registration. By meticulously defining every step of your research plan—from hypotheses to analysis—before observing the outcomes, you create a firewall against the biases that fuel the replication crisis.

Key Takeaways:

• A pre-registration is a time-stamped, immutable research plan that distinguishes confirmatory from exploratory analyses.
• A robust pre-registration should be specific, precise, and exhaustive, leaving no room for opportunistic choices during the study.
• Using a structured template (like the OSF Prereg form) and a checklist of researcher degrees of freedom ensures all aspects of the study are prospectively defined.
• The analysis plan is the core of the pre-registration, specifying the exact statistical models, inference criteria, and rules for handling data to prevent p-hacking and HARKing.

This lesson concludes our module on pre-registration. You are now fully equipped to design and document an experiment with the highest standards of transparency and rigor. In our next module, "Structural Causal Modeling of Triadic Dynamics," we will move from planning the experiment to modeling its underlying causal structure. We will construct a Directed Acyclic Graph (DAG) for the vee system, allowing us to formally represent and analyze the complex interference pathways that we've accounted for in our experimental design.