HOW SINGAPORE EDUCATION WORKS · University reproducible figures
Find your next learning step
A figure is reproducible when another authorised person can start from the declared source data, run the documented transformation and plotting steps, and obtain the same plotted values, labels and export—not merely a similar-looking image.
Full chapter index · Practice and checking · Sources and next reading · How Singapore Education Works
A polished chart can still be impossible to audit if the source spreadsheet, filtering decisions or export settings are missing. Treat the image as the last product in a chain: data, transformation, plotting specification, validation and caption.
This is a practical learning workflow. It does not replace a university, laboratory, journal, funder, ethics board, copyright or data-protection requirement. Use approved storage and follow the responsible institution’s rules.
The choices at a glance
| What you notice | Likely issue | Useful next move |
|---|---|---|
| The PNG exists but no source table can be found | The final image has been separated from provenance | Declare the source data and rebuild path |
| Rerunning code changes labels or ordering | Hidden defaults or unsorted categories control the plot | Fix ordering, units, versions and export settings |
| The caption repeats the title only | The reader lacks conditions and interpretation limits | State variables, units, sample, uncertainty and scope |
Full chapter index
Work and diagnose
Repair and practise
Recall, transfer and questions
The visible image is not the whole research object. It depends on source data, exclusions, transformations, plotting choices and a caption that tells the reader what is being shown.
Keep raw or approved source data separate from derived plotting data. A small table prepared for a figure should have a documented route back to the source.
Reproducibility does not mean every dataset can be public. Restricted data can still have a controlled, documented workflow inside authorised systems.
Record the variables, units, filters, grouping, statistical summary, uncertainty display, axis limits, category order, colour meaning and output dimensions.
Prefer a script or saved analysis notebook when the tool supports it. If a graphical interface is required, keep an exportable recipe or step log with software and version.
Avoid manual pixel edits that change plotted evidence. If layout work is necessary, keep the editable source and document what changed.
Assumptions: an approved CSV contains week, cohort, attended and enrolled counts. The figure shows attendance percentage by week for two cohorts. Missing enrolment is excluded; percentage equals attended divided by enrolled times 100.
The script reads the declared CSV version, checks attended is not greater than enrolled, applies the exclusion rule, sorts weeks chronologically, calculates percentages and exports a 1600-by-900 image plus vector PDF.
Checked answer: a second authorised user reruns the script and matches every plotted percentage to the derived table. The caption states cohorts, weeks, percentage definition, exclusions and that the figure is descriptive rather than causal.
Start from the caption and trace backwards. Can you identify the dataset version, exact columns, transformation, plotting command and export? Mark the first missing link.
If values match but the picture differs, inspect fonts, software versions, random seeds, category order and dimensions. Cosmetic differences may be harmless, but altered labels or geometry can change interpretation.
If the source changed after publication, preserve the cited version. Rebuilding from a newer file without noting the change is not verification of the original figure.
Create one figure folder or manifest that points to source data, derived table, code or step log, environment information, output files and caption.
Name the figure consistently across manuscript, files and code. Store checks such as row counts, ranges or hashes where appropriate so silent input changes are visible.
Do not delete prior published outputs merely because a new version exists. Retention and supersession should follow project and institutional rules.
For guided practice, provide a tiny dataset and an intentionally incomplete bar chart. The learner adds units, category order, calculation notes and a caption, then exports through one documented command.
For independent practice, give a different dataset plus a figure but remove the plotting script. The learner reconstructs a transparent workflow, states assumptions and compares values—not pixels alone—with the supplied figure.
Check by handing the manifest to another person. If they need undocumented oral instructions, the workflow is not yet handover-ready.
After a delay, ask the learner to list the minimum rebuild chain from memory: source, transformation, plot specification, validation, export and caption.
Transfer the method from a bar chart to a map, microscopy panel or model-performance figure. Extra tools may be needed, but provenance and declared decisions remain.
Global connection: research teams work across institutions, languages, software and time zones. A rebuildable figure reduces dependence on one person’s memory while local ethics, access and licensing rules still govern the material.
These answers describe robust habits, not a universal journal specification.
Is saving the spreadsheet enough?
Not if filtering, calculations, order or export settings are unclear. Preserve the transformation and plotting specification too.
Must every figure use code?
No. The method can use a documented GUI workflow, but another authorised person must be able to repeat the material steps.
Should the image itself contain every detail?
No. Use a clear visual plus a complete caption and linked methods. Avoid overcrowding the plot.
What should be validated?
Check source version, row counts, calculations, units, labels, uncertainty, category order and agreement between plotted values and the derived table.
Sources and next reading
NUS Libraries: Research Data Management—Data Organisation — official guidance on consistent organisation, filenames and versioning.
NUS Libraries: Metadata and Documentation — official guidance on README information that supports reproducibility and reuse.
NUS Libraries: Open Science — official context for making research processes available for reuse and reproduction where conditions allow.
NUS Libraries: Data Formats — official guidance on sustainable formats for tabular and digital image data.
University data-cleaning decision logs — a related guide to recording transformations and validation decisions.
University research file naming and version control — a related guide to masters, versions and handover-ready READMEs.