King's College London · FACULTY OF STATISTICS

7SSGN110 Chap.1 Scientific Questions and Reproducible Data

- one subject, every graph, every model, every mark
6 Chapters3-page Bible
Our own words - no uploaded lecturer files
Updated for this semester
Chapter 1 of 8 · 7SSGN110

Scientific Questions and Reproducible Data

Design before software

An environmental analysis begins by identifying the population, observational unit, response, explanatory variables and measurement scale. Those decisions determine what a row means and which comparisons are legitimate. A table with several readings per site may contain many records but few independent units.

Before opening a statistical function, write one sentence that names the process of interest, the entities sampled and the scope of the intended conclusion.

Build a data dictionary

Translate cryptic column names into environmental quantities with units, valid ranges and collection methods. Record whether a variable is categorical, ordered, continuous or a count, and distinguish explanatory roles from responses.

Missing-value codes deserve special attention because a zero, blank and minus 999 may have different meanings. Stable site and occasion identifiers must survive sorting, filtering and joining so that repeated observations cannot be mistaken for unrelated samples.

Preserve a reproducible preparation trail

Keep the acquired file unchanged and generate the analytical dataset through a saved script.

Each recode, exclusion, unit conversion and derived variable should be visible in sequence. Add simple checks after important operations: row counts, duplicate identifiers, plausible ranges and the number of missing values. A final table should reconcile how many records entered, how many were removed and why.

This audit is more informative than a cleaned spreadsheet whose history has disappeared.

Separate replication from independence

Repeated readings can improve knowledge of one sensor, plot or day without expanding spatial or biological coverage. Dependence enters when observations share a site, subject, instrument or time process.

The repair depends on the question: summarise repeated readings at the independent-unit level, represent clustering in the model, or narrow the claim.

Simply calling every row a replicate understates uncertainty and gives the appearance of evidence that the design did not collect.

Connect question, method and claim

A difference question, a relationship question and a time-ordered question require different structures even when they use the same columns.

State the desired claim before selecting a technique, then identify an alternative explanation that the design cannot eliminate. The report should let a reader trace each figure, table and coefficient back to a named prepared dataset.

A defensible conclusion states the observed pattern, the sampled scale and the conditions that limit generalisation rather than using software output as proof.

Audit the complete chain

Before reporting, select one numerical result and reconstruct it from the raw source. Verify the identifier, unit conversion, missing-data rule, transformation, method call and final rounding.

Then compare the sample described in the methods with the records actually used by the model. This reverse audit catches silent row loss and stale figures. Store code, input versions and output together, while keeping original data read-only.

Another analyst should be able to reproduce the same table and understand why each choice was made without relying on memory.

For revision, take a small unfamiliar dataset and narrate the full chain aloud: unit, variables, checks, method and claim. Then change one feature of the design and identify which conclusion must narrow.

Recording that first broken link builds a reusable audit habit rather than familiarity with one polished example.

In this chapter

What this chapter covers

  • 01

    Research questions

  • 02

    Observational units

  • 03

    Variable roles

  • 04

    Metadata

  • 05

    Reproducible preparation

  • 06

    Claim boundaries

Worked example · free

Audit repeated wetland measurements

Q [7 marks]. AskSia-authored practice. The marks shown support revision and are not an official university marking scheme. Four plots are measured on three mornings. Explain why twelve rows are not twelve independent plots and repair the analysis plan.
Analysis workflow.Analysis workflow. Each box supplies a check for the next, so interpretation cannot be detached from design and data quality.QuestionVariablesInspectAssumptionsMethodInterpret
Analysis workflow. Each box supplies a check for the next, so interpretation cannot be detached from design and data quality.
  • 2Name the plot as the independent environmental unit.
  • 2Identify morning as repeated within plot.
  • 3Narrow or restructure the comparison.
The twelve rows are plot-morning measurements, but observations within a plot share place and conditions. Summarise within plot for a narrow comparison or use a model that represents repeated measurements, then restrict the conclusion to the sampled plots unless the design supports a wider population.
Sia tip — Write the unit represented by one independent piece of evidence before counting the sample size.
Glossary

Key terms

Observational Unit
An Observational Unit is the independent entity or occasion represented by an analytical record.
Metadata
Metadata describe variables, units, provenance and measurement protocols.
Reproducibility
Reproducibility is the ability to rerun a documented procedure and obtain the same result.
FAQ

Scientific Questions and Reproducible Data FAQ

Why does the observational unit matter?

The observational unit determines what counts as independent evidence and which population the result can describe. Repeated readings may improve measurement precision without adding independent sites, so counting rows alone can exaggerate sample size and understate uncertainty.

What belongs in a data audit?

Record identifiers, variable roles, units, valid ranges, missing-value codes, collection frequency and any transformations. Preserve the untouched input and write cleaning steps in a script so every figure, table and model can be traced back to the same prepared data.

How broad can the conclusion be?

Keep the claim within the sampling frame, period and measurement process. A week of hourly readings from one sensor can describe that sensor and period precisely, but it does not automatically represent every location in a city or catchment.

Study strategy

Assessment move

Take an unfamiliar dataset and build a one-page audit before calculating anything. Name the research question, row identity, independent unit, variable roles, units, missing codes and target population. Then describe one plausible dependency or bias and the analysis change it requires.

Working through Scientific Questions and Reproducible Data in 7SSGN110? Sia is AskSia’s AI Statistics tutor — ask any 7SSGN110 Scientific Questions and Reproducible Data question and get a clear, step-by-step explanation grounded in how 7SSGN110 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.

A+Everything unlocked
Unlocks this Bible + all 18 of your subjects at King's College London, plus 1,000+ more Bibles across the AskSia library.
Sia - your 7SSGN110 tutor, unlimited, worked the way the exam marks it
The full 3-page Bible + practice bank with worked solutions
Chrome extension - sync your LMS so Sia knows your deadlines
Bilingual EN / Chinese on every Bible and every Sia answer
$0.99 Trial
30-day money-back · cancel in one tap · how it works