BENV0139 Chap.7 Sampling, Study Design and Robustness
Sampling, Study Design and Robustness
Sampling, Study Design and Robustness builds an analysis plan from a measurable outcome to a decision. Longitudinal environmental records do not create longitudinal evidence about deterioration when condition is rated only once.
Repeated months are grouped within museums, missing sensor periods can follow extreme events, and building type, material mix, visitor numbers or conservation resources can confound observed associations. A useful plan audits units and missingness, visualises time and distribution structure, chooses a model matching the outcome and dependence, checks assumptions, and stress-tests defensible alternatives.
Robustness is not a search for the preferred answer; each alternative tests a named uncertainty such as exposure window, outlier handling or outcome coding. Generalisability also needs a target population. The chapter converts these design choices into a transparent sequence that another analyst can audit and repeat.
What this chapter covers
- 01
Outcome definition
- 02
Cross-sectional and longitudinal designs
- 03
Missing-data mechanisms
- 04
Confounding and interactions
- 05
Grouped observations
- 06
Sensitivity analysis
- 07
Generalisability and stakeholder action
Worked example · free
Plan a multi-museum environmental study
- 2Define and harmonise a condition outcome with dates and material context. The marks are original practice weighting.
- 2Audit missing exposure periods, repeated museum structure and predictor correlations.
- 2Identify confounders and choose a model representing museum grouping, time and outcome type.
- 2Check residuals, influence and nonlinearity, then repeat under alternative exposure windows and data-quality rules.
- 1Translate estimates into material-specific action and state limits on generalisation.
Key terms
- Longitudinal design
- A design that follows the same units across multiple relevant time points.
- Missingness mechanism
- The process determining why some intended observations are unavailable.
- Sensitivity analysis
- A comparison designed to show whether a conclusion depends on a plausible analytical choice.
Sampling, Study Design and Robustness FAQ
Why is one condition survey not longitudinal?
Repeated environmental predictors do not make the outcome repeated. One final condition rating supports mainly between-museum comparison and remains vulnerable to baseline differences. Repeated standardised condition assessments are needed to estimate within-museum change directly.
How should missing sensor data be handled?
First map when and where values are missing and why. Complete-case analysis is risky when outages relate to extremes or maintenance. Any imputation should reflect time and site structure, and the conclusion should be checked under alternative defensible assumptions.
What makes a robustness check useful?
It tests a named uncertainty that could change the conclusion, such as excluding low-quality periods, changing exposure windows or using a nonlinear form. Report all planned alternatives, including those that weaken the preferred result, rather than searching until one specification passes a threshold.
Exam move
Memorise a nine-part planning sequence: decision, outcome, observational unit, sampling, data audit, exploratory display, confounders, model and robustness. Apply it to different heritage settings until the same sequence produces different specific choices. Practise identifying what a dataset cannot answer.
Environmental time series without repeated condition cannot estimate condition change; many objects from one site may not generalise to all museums; missing extreme periods may bias exposure summaries. Draw simple causal diagrams to decide whether a variable is a confounder, mediator or consequence. Build a robustness table with the uncertainty tested, alternative analysis and possible decision change.
For generalisability, name the target population and compare it with the observed museums, materials and operating conditions. End every plan with the data collection step that would most improve the decision. This makes limitations constructive and keeps the analysis tied to real heritage management rather than a catalogue of software techniques.
Expand the sequence into an analysis-plan table with columns for choice, evidence, assumption, diagnostic and decision consequence. Put dates on outcomes, exposures, moves and treatments so causes precede the changes they might explain. Make a missingness map by museum, object, variable and time, and translate each blank into sensor failure, access restriction, detection limit, not applicable or unknown.
Compare complete and incomplete records using variables observed for both groups. Draw a causal diagram for visitor numbers, resources, climate control and condition, then explain why adjusting for a mediator can create a different bias. Design robustness checks before fitting: alternative exposure windows, outcome codings, nonlinear forms, outlier treatment and leave-one-site-out analyses.
For each, predict what result would weaken the recommendation. Match validation to the target population and explain why precision within fifty selected museums cannot erase selection bias. Finish with the single next measurement that most improves the decision, its timing and sampling route. Distinguish limitations that widen intervals from those likely to shift the estimate.
Working through Sampling, Study Design and Robustness in BENV0139? Sia is AskSia’s AI Statistics tutor — ask any BENV0139 Sampling, Study Design and Robustness question and get a clear, step-by-step explanation grounded in how BENV0139 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.