In our previous lesson, we constructed a robust treatment sequence using a balanced Latin Square design. This ensures that the order in which our vee triad experiences the different living arrangements (Triadic Cohabitation, Anchor-Satellite, Dispersed Living) is statistically sound, allowing us to distinguish treatment effects from confounding by period or first-order carryover. We now move from the "what order" to the "how long" and "what else" of our experimental protocol.
This lesson addresses two critical components of timing and control. Our goal is to specify blocking factors (e.g., external events) and determine an adequate period length for outcome stabilization. First, we will identify predictable external variables that could introduce noise into our satisfaction measurements and discuss how to neutralize their influence through blocking. Second, we will define a rigorous, data-driven method for deciding how long each treatment and washout period should last, moving beyond arbitrary fixed durations to a more dynamic, responsive strategy.
1. Controlling for Extraneous Variables: Blocking Factors
Your background in experimental design has equipped you with the principle of blocking to control for known sources of variation. In a classic agricultural trial, this might be a gradient in soil quality across a field. In our N-of-1 social experiment, the "field" is the life of the triad, and the "soil gradients" are the predictable ebbs and flows of life that can influence satisfaction independent of our intervention. In the language of single-case designs, these are potential threats to internal validity.
The following resource provides a formal introduction to these time-related threats.
[PDF] SINGLE-CASE EXPERIMENTAL DESIGNS - UNCW
This chapter provides the vocabulary for the challenges we face. While you are familiar with these concepts, this section frames them specifically for single-case studies.
Focus on the definitions of the four time-related threats to internal validity: history, maturation, testing, and instrumentation. Consider how each of these could manifest in our vee triad experiment. For instance, "history" could be an external work project causing stress, while "maturation" could include physiological cycles or simply the natural evolution of the relationships over the course of the study.
Rather than letting these nuisance variables add random noise (or worse, systematic bias) to our data, we can proactively account for them. This is where blocking becomes essential.
Independent variables | NC3Rs EDA
This guide from the NC3Rs EDA provides a clear, practical framework for dealing with nuisance variables, which is directly applicable here.
First, read the brief definitions of nuisance variables and the rationale for managing them. Then, focus on the section Blocking a nuisance variable. This section explains how breaking an experiment into "mini-experiments" or blocks effectively isolates the variability from a known source.
Let's apply this to our triad. What are some predictable, recurring nuisance variables we should consider blocking?
- Work/Life Stress Cycles: Many professions have predictable high-stress and low-stress periods (e.g., end-of-quarter deadlines, academic semesters). If one living arrangement is tested entirely during a high-stress period and another during a low-stress period, we might falsely attribute the stress-induced satisfaction changes to the treatment.
- Menstrual Cycles: For the female pivot, hormonal fluctuations can significantly impact mood, energy, and social dynamics. This is a classic "maturation" variable.
- Week Structure: The social and logistical realities of weekends are often vastly different from weekdays.
We can use blocking to control for these. For example, instead of a simple randomization, our analysis could block by "Work Status," comparing satisfaction scores within high-stress weeks and within low-stress weeks separately before aggregating. For the week structure, we can ensure each treatment period is long enough to include several full weekend/weekday cycles, allowing us to compare, for example, "Satisfaction on Weekends under Treatment A" versus "Satisfaction on Weekends under Treatment B."
For some factors, formal blocking might be impractical. We may not be able to perfectly balance the number of "high-stress days" in each treatment period. In such cases, we don't ignore the variable; we measure it and can use it as a covariate in our final statistical model. For instance, we could have all three members rate their daily external stress on a simple 1-5 scale. This data allows us to statistically control for the effect of stress when evaluating the effect of the living arrangement.
2. Period Duration: The Steady-State Strategy
A common flaw in crossover designs is the use of arbitrarily fixed period lengths (e.g., "four weeks for each treatment"). Behavior, especially complex social satisfaction, does not stabilize on a fixed schedule. A more rigorous approach, common in behavioral analysis and highly suited to N-of-1 trials, is the steady-state strategy.
The core idea is that a treatment or washout period continues until the dependent variable (in our case, satisfaction) reaches a state of equilibrium. This ensures we are comparing stable states, not transitional phases. Your background in stochastic processes and simulation, where systems are run until they reach equilibrium, provides a strong parallel.
The following reading details this powerful strategy.
[PDF] SINGLE-CASE EXPERIMENTAL DESIGNS - UNCW
This part of the chapter is the methodological core of our lesson. It outlines the logic and practice of using behavioral stability to guide experimental progression.
First, read the section on the steady-state strategy to understand its fundamental logic. Pay close attention to the three requirements: control, time, and recognition. Next, focus on the "Stability Criteria" subsection. This explains how to recognize a steady state. Read through the different approaches, from visual inspection to quantitative rules. This is what we will use to create a stopping rule for each period.
Let's translate this into a concrete protocol for our vee triad.
A. Determining Treatment Period Length
Each treatment period (e.g., the period of Triadic Cohabitation) does not end after a pre-set number of weeks. Instead, it ends when the triad's satisfaction scores meet a pre-defined stability criterion.
This requires two steps:
-
Measurement: As we will discuss in a future module, the triad will need to complete daily satisfaction diaries. This gives us a time-series of satisfaction scores for each individual and for the triad as a whole.
-
Criterion: We must define "stable" quantitatively. A reasonable stability criterion could be:
"A treatment period will conclude when, for two consecutive weeks, the weekly average triad satisfaction score shows no monotonic trend, and the mean of the second week is within ±5% of the mean of the first week."
This criterion prevents us from ending a period prematurely while satisfaction is still trending upwards (or downwards) or while it is highly volatile. We are waiting for the system to settle into its new equilibrium under the given living arrangement.
B. Determining Washout Period Length
The same logic applies to washout periods. The purpose of a washout is to allow the effects of the previous condition to dissipate, returning the triad to a neutral baseline. A fixed two-week washout is a guess. A criterion-based washout is an empirical verification.
The stability criterion for a washout would be different:
"A washout period will conclude when the triad's satisfaction scores have returned to the levels observed in the initial pre-experiment baseline phase and have met the stability criterion (no trend, low variance) for one full week."
This ensures that we don't start the next treatment while the members are still feeling the "afterglow" or "rebound" from the previous one, thus minimizing carryover effects in a practical, not just statistical, sense.
Conclusion
By moving from a fixed-schedule design to a dynamic, criterion-based approach, we significantly increase the rigor and internal validity of our N-of-1 experiment. We are letting the data from the triad itself tell us when it is appropriate to move to the next phase.
Our key takeaways are:
- Predictable external events and internal cycles (e.g., work stress, physiological cycles) are nuisance variables that threaten internal validity. They should be managed through blocking in the analysis or by measuring them for use as covariates.
- The duration of treatment and washout periods should not be fixed arbitrarily. Instead, we should use a steady-state strategy, where each period continues until the outcome measures (satisfaction scores) achieve a pre-defined, quantitative stability criterion.
- This approach ensures we are comparing the effects of different living arrangements on the triad's stable state of satisfaction, rather than on transitional, unstable periods.
In our next lesson, we will confront another major challenge in this behavioral experiment: bias from participant and researcher expectations. We will focus on the learning outcome: Design partial blinding strategies (e.g., blinded outcome assessment) to mitigate bias when participant blinding is impossible.