6.1 Quantitative Research Designs, Metrics & Statistical Significance

Quantitative Research Designs, Metrics & Statistical Significance

Core Principle: Quantitative inquiry in Evidence-Based Design (EBD) operates through a deductive scientific paradigm. Researchers state testable hypotheses, isolate environmental and operational variables, gather structured numerical data, and apply inferential biostatistics to establish whether physical design interventions produce statistically verifiable and clinically meaningful improvements.

Healthcare environments represent intricate, dynamic socio-technical ecosystems. Unlike laboratory bench science, where external influences can be sealed in a vacuum, architectural modifications occur in active clinical spaces populated by vulnerable patients and shifting multidisciplinary teams. Consequently, conducting quantitative research in healthcare design requires a sophisticated understanding of experimental design structures, variable classifications, threats to validity, and biostatistical metrics.

---

Quantitative Research Designs in Healthcare Architecture

Quantitative study designs fall along a continuum of scientific control and internal validity. In physical healthcare architecture, selecting a design requires balancing rigorous causal inference with physical feasibility, institutional ethics, and capital constraints.

1. True Experimental Designs (Randomized Controlled Trials)

In a true experimental design, the investigator exercises full control over the administration of the independent variable and uses random assignment (randomization) to allocate participants or operational clusters into experimental and control arms.

• Application in EBD: True experiments are extraordinarily difficult to execute at the full architectural scale because researchers cannot randomly assign patients to live in a new hospital tower versus a dilapidated 1970s wing without introducing grave ethical and operational concerns. However, true experimental designs are frequently and effectively deployed in environmental simulation laboratories, high-fidelity mock-up rooms, and micro-environmental clinical trials.
• Representative Example: An EBD team investigating the impact of dynamic circadian lighting (varying correlated color temperature from 2700K to 6500K and melanopic lux across a 24-hour cycle) randomly assigns 60 healthy ICU night-shift nurses into two identical simulation suites: one with circadian-attuned tunable LED systems and the other with standard static 4000K fluorescent troffers. Salivary melatonin assays, pupillometry, and computerized vigilance tasks are recorded.

2. Quasi-Experimental Designs: The Workhorses of EBD

Because physical architecture deals with permanent buildings and intact clinical departments, quasi-experimental designs represent the primary methodological standard for published healthcare design literature. Quasi-experiments lack random assignment but incorporate deliberate structural controls—such as baseline pre-tests, matched comparison units, or repeated longitudinal observations—to rule out rival explanations.

A. Pretest-Posttest Non-Equivalent Control Group Design

This design evaluates an intact clinical unit receiving a physical renovation (the experimental unit) alongside a demographically and operationally similar unit in the same institution or health system that remains unrenovated (the non-equivalent control unit).

Experimental Unit:  O1  ──>  [Environmental Intervention: X]  ──>  O2
Comparison Unit:    O1  ───────────────────────────────────────────>  O2
(Where O1 = Pretest Baseline, X = Architectural Modification, O2 = Posttest Evaluation)

• Clinical Application: A hospital replaces a 24-bed centralized medical-surgical ward with a decentralized nursing alcove layout featuring point-of-care charting and direct bed sightlines. Simultaneously, an identical 24-bed medical-surgical unit in an adjacent tower retains its centralized station.
• Why It Matters: If patient fall rates drop by 40% on the renovated unit over 12 months, but also drop by 38% on the unrenovated unit, the improvement cannot be attributed to the decentralized architecture. Instead, it reflects an unmeasured hospital-wide confounder, such as a newly mandated nursing fall-risk assessment protocol or the rollout of non-skid socks.

B. Interrupted Time Series (ITS) Design

The interrupted time series design involves taking multiple, evenly spaced quantitative measurements over an extended baseline period before an environmental intervention occurs, followed by an extended series of repeated post-occupancy measurements.

• Overcoming the "Novelty Effect" and "Move-In Dip": When clinical teams move into a brand-new, state-of-the-art facility, operational metrics frequently deteriorate during the first 30 to 90 days. Staff suffer orientation fatigue, struggle to locate supplies in new decentralized storage rooms, and adjust to unfamiliar technology. A simple two-point before-and-after study conducted during this "move-in dip" would falsely conclude that the new design harmed workflow.
• Secular Trends & Seasonality: An ITS design collecting monthly medication error rates for 24 months pre-occupancy and 24 months post-occupancy allows biostatisticians to establish underlying seasonal spikes (e.g., winter respiratory surges) and long-term trajectory changes in both slope and intercept immediately following occupancy.

---

Study Types Named in EDAC Study Materials

EDAC Study Guide 2 describes the following study types. Several can overlap (for example, a prospective before-after quasi-experiment):

Study materials also define a positive correlation: when one variable increases, the other variable also tends to increase.

---

Variable Classification in Built Environment Research

In quantitative research, variables must be explicitly operationalized—defined in concrete, measurable terms. EDAC study materials put it simply: independent variables are causes and dependent variables are effects, and in EBD research environmental variables are independent and outcomes are dependent. Be able to distinguish these classifications:

                   ┌─────────────────────────────┐
                   │     MODERATING VARIABLE     │
                   │   (e.g., Patient Acuity)    │
                   └──────────────┬──────────────┘
                                  │ (Modifies strength/direction)
                                  ▼
┌──────────────────────┐     ┌─────────────────────┐     ┌─────────────────────┐
│ INDEPENDENT VARIABLE │ ──> │ MEDIATING VARIABLE  │ ──> │ DEPENDENT VARIABLE  │
│ (Environmental X)    │     │ (Psychophysiological│     │ (Clinical / ROI Y)  │
│ e.g., Sound Panels   │     │  e.g., Sleep Arch.) │     │ e.g., Delirium Rate │
└──────────────────────┘     └─────────────────────┘     └─────────────────────┘
                                  ▲
                                  │ (Distorts true causal relationship)
                   ┌──────────────┴──────────────┐
                   │    CONFOUNDING VARIABLE     │
                   │ (e.g., Sedative Medication) │
                   └─────────────────────────────┘

1. Independent Variables (IV) — The Environmental Interventions

The independent variable is the causal factor, attribute, or physical condition manipulated, introduced, or categorized by the researcher. In EBD, independent variables are physical elements of the built environment:

• Acoustic Interventions: High-performance sound-absorbing ceiling tiles with a Noise Reduction Coefficient (NRC) of 0.90 versus standard NRC 0.50 tiles; installation of sound-masking systems producing pink noise spectrums at 45–48 dBA.
• Lighting Spectra & Photometrics: Ambient horizontal illuminance (lux), correlated color temperature (CCT in Kelvin), Equivalent Melanopic Lux (EML), or direct access to daylight via exterior fenestration.
• Spatial Layout & Typology: Distance from nursing workstations to patient beds (meters); centralized versus decentralized nursing stations; single-bed private rooms versus multi-occupancy semi-private rooms.
• Material Finishes & Hygiene Touchpoints: Hand-hygiene sink placement directly in the visual field upon room entry versus concealed behind an entrance vestibule; copper-impregnated or antimicrobial composite touch surfaces versus standard stainless steel.

2. Dependent Variables (DV) — The Outcome Endpoints

The dependent variable represents the measurable outcome, effect, or response hypothesized to change when the independent variable is altered. In healthcare EBD, dependent variables span three primary domains:

• Clinical Safety Outcomes: Hospital-Acquired Infections (HAIs) such as Clostridioides difficile or Central Line-Associated Bloodstream Infections (CLABSIs) per 1,000 patient days; unassisted patient fall rates; post-operative analgesic consumption (morphine milligram equivalents [MME]); post-operative delirium incidence.
• Staff Performance & Operational Outcomes: Registered nurse travel distance (kilometers per 12-hour shift measured via RFID badges); medication dispensing turnaround times; medication administration error rates; nurse recruitment and retention rates; physical fatigue scores.
• Perceptual & Organizational Outcomes: Patient Satisfaction scores (such as HCAHPS scores for "quietness of the hospital environment" or "cleanliness"); Average Length of Stay (ALOS) adjusted for Case Mix Index (CMI); operating margins and capital return on investment (ROI).

3. Mediating Variables — The Causal Pathway

A mediating variable explains the mechanism, pathway, or "why" through which the independent variable influences the dependent variable. It functions as an intermediate psychological, physiological, or behavioral stepping stone.

• Causal Chain: Independent Variable (IV) ──> Mediating Variable (Mediator) ──> Dependent Variable (DV)
• Clinical Example: An EBD team installs decentralized supply and medication alcoves outside patient rooms (IV). This reduces the total physical walking distance and steps taken by nurses per shift (Mediator), which in turn reduces physical exhaustion and cognitive fatigue (Second Mediator), thereby significantly reducing clinical medication administration errors (DV).

4. Moderating Variables — The Contextual Boundary Conditions

A moderating variable is an existing condition, characteristic, or factor that alters the strength, direction, or presence of the relationship between the independent and dependent variables. It answers the question: "For whom or under what operational conditions does this environmental intervention work?"

• Clinical Example: An architectural team introduces large nature views and circadian daylighting (IV) to reduce patient depressive symptoms and post-operative length of stay (DV). The clinical acuity of the patient (e.g., comatose trauma patients versus alert elective orthopedic surgical patients) acts as a moderating variable. Hypothetically, nature views might shorten stays for alert elective patients yet show no measurable effect for heavily sedated patients who cannot see or process the view.

5. Confounding and Extraneous Variables

A confounding variable is an unmeasured or uncontrolled factor that is correlated with both the independent variable and the dependent variable, potentially creating a false (spurious) association or masking a real causal link.

• Clinical Example: A hospital upgrades an oncology ward's HVAC system with HEPA filtration and higher air change rates (IV) to reduce invasive aspergillosis infection rates (DV). Concurrently, the hospital mandates a new clinical contact-isolation cleaning protocol using ultraviolet-C (UV-C) disinfection robots. The UV-C protocol is a severe confounder; without rigorous statistical adjustment, the research team cannot determine whether reduced fungal infections stemmed from the HVAC air changes or the UV-C cleaning robots.

---

Establishing Causality in Built Environment Studies

Demonstrating a mathematical correlation between a design feature and an operational metric does not prove causality. Three widely used criteria for causal inference (rooted in John Stuart Mill's reasoning) are:

1. Temporal Precedence (Cause Precedes Effect): The independent environmental intervention must occur prior in time to the observed change in the dependent clinical outcome. Cross-sectional studies that measure spatial layout and staff burnout at the exact same instant cannot determine whether poor layout caused burnout or whether burned-out teams organized their physical spaces chaotically.
2. Covariation of Cause and Effect (Systematic Relationship): There must be an empirical, verifiable statistical association between the physical environment and the outcome. When the environmental factor is present or increased (e.g., higher sound absorption), the outcome changes predictably (e.g., nocturnal awakenings decrease); when the environmental factor is absent or reduced, the outcome reverts.
3. Non-Spuriousness / Elimination of Plausible Alternative Explanations: The researcher must prove that the observed relationship is not the result of a third, unmeasured confounding variable (e.g., changes in nurse staffing ratios, seasonal disease incidence, clinical care protocols, patient acuity shifts, or Hawthorne effects). This criterion is the most difficult to satisfy in healthcare research and requires robust quasi-experimental controls and multivariate statistical adjustments.

---

Biostatistics: Significance, Effect Size & Metrics

Evidence-based designers must possess statistical literacy to critically appraise published literature in journals like HERD (Health Environments Research & Design Journal) and separate scientific evidence from marketing hyperbole.

Descriptive vs. Inferential Statistics

CHD's current exam content outline asks teams to collect, analyze, and evaluate data and the effects or outcomes of the completed project using both descriptive statistics and inferential statistics. Know the difference:

Measures of central tendency describe the typical value: the mean (average, pulled by outliers), the median (middle value, useful for skewed data such as length of stay), and the mode (most frequent value). Measures of variability describe spread: the range, interquartile range, and standard deviation.

A simplified guide to common inferential tests:

1. The Null Hypothesis and the p-Value

• Null Hypothesis (H0): Assumes that there is no true difference, effect, or relationship between the physical environmental intervention and the clinical outcome (i.e., any observed difference is purely the result of random sampling variation).
• Alternative / Research Hypothesis (H1): Posits that the environmental intervention produces a real, measurable change in the outcome.
• The p-Value (Probability Value): The probability of obtaining an empirical result equal to, or more extreme than, what was actually observed, assuming that the null hypothesis is completely true.
• Significance Threshold (alpha = 0.05): By convention, a result with p < 0.05 is called statistically significant, and researchers reject the null hypothesis. It means results this extreme would be unlikely (under 5%) if there were truly no effect—not that there is a 95% chance the hypothesis is true.

[!CAUTION]

EXAM TRAP: What a p-Value Does NOT Mean

A common error is mistaking statistical significance for magnitude or practical importance:

1. A p-value does not measure the size, strength, or clinical importance of an architectural design effect.
2. With an enormous sample size (e.g., N = 250,000 patient records across a national healthcare chain), a trivial, clinically meaningless difference—such as a 4-second reduction in nurse call-light response time—can easily yield a statistically significant p < 0.001.
3. Conversely, in a small pilot study (N = 18), a massive clinical improvement may fail to achieve p < 0.05 solely due to low statistical power (a Type II error).

2. Effect Size: Quantifying Practical and Clinical Significance

While the p-value tells an architect whether an effect exists, the effect size measures how large the effect is in real-world operational and biological terms.

• Cohen's d (Standardized Mean Difference): Used when comparing continuous means between two groups (e.g., average decibel levels in carpeted versus resilient vinyl corridors):
  • d = 0.20 : Small effect
  • d = 0.50 : Moderate effect
  • d = 0.80 : Large, highly impactful effect
• Pearson's Correlation Coefficient (r): Measures the linear relationship between two continuous variables, ranging from −1.00 (perfect inverse correlation) to +1.00 (perfect direct correlation), with 0.00 representing no linear association.
  • r = 0.10 to 0.29 : Weak correlation
  • r = 0.30 to 0.49 : Moderate correlation
  • r ≥ 0.50 : Strong correlation
• Coefficient of Determination (R²): Represents the proportion of variance in the dependent variable that is predictable from the independent environmental variable (e.g., an R² of 0.36 indicates that 36% of the variance in nurse fatigue is explained by walking distance from central supply).

3. Confidence Intervals (CIs)

A confidence interval (typically reported as a 95% CI) is a range calculated from the data; if a study were repeated many times, about 95% of intervals calculated this way would contain the true population value.

• Precision Indicator: A narrow confidence interval indicates high measurement precision and sample reliability; a wide confidence interval signals small sample size, noisy data, or high measurement variability.
• Determining Statistical Significance via CIs: If a 95% confidence interval for a difference between two means spans across zero (e.g., 95% CI [-1.2, +3.8]), the finding is not statistically significant (p ≥ 0.05). For risk ratios or odds ratios, if the 95% CI spans across 1.00, the finding is not statistically significant.

---

Threats to Internal and External Validity

When evaluating research on built environments, design researchers must identify and mitigate systematic threats to validity.

Threats to Internal Validity (Is the Environmental Design Truly the Cause?)

1. History: External events occurring between the pretest and posttest that alter the outcome independently of the physical design (e.g., a viral pandemic or hospital-wide staffing layoffs during a renovation).
2. Maturation: Biological or psychological processes within participants that change naturally over time (e.g., novice nurses naturally becoming faster at medication delivery as they gain clinical experience).
3. Hawthorne Effect (Observation Bias): Participants alter their behavior simply because they know they are being observed and studied as part of an architectural pilot project, rather than in response to the physical space itself.
4. John Henry Effect (Compensatory Rivalry): Staff in the unrenovated control unit realize they are being compared to colleagues in a glamorous new facility and exert extraordinary extra effort to prove their old unit is equally effective.
5. Regression to the Mean: If an experimental unit is selected for an architectural redesign specifically because its clinical metrics were exceptionally terrible during the prior quarter, its metrics will statistically tend to improve toward the institutional average regardless of the renovation.

Threats to External Validity (Can Findings Be Generalized?)

1. Context Dependency: A spatial prototype optimized for an academic tertiary trauma center in an urban center may fail completely in a rural critical access hospital with different patient acuity, staffing ratios, and cultural demographics.
2. Interaction of Setting and Treatment: Unique architectural characteristics of a host building (such as expansive historical window openings) cannot be generalized to standard low-ceiling suburban healthcare facilities.

---

High-Yield Summary Table for Quantitative Research