What the spreadsheet does well
Starting here is not politeness: a comparison that fails to acknowledge the strengths of the incumbent tool convinces nobody who uses it.
- It is immediate. No installation, no account, no training. An analytical idea can be tested in ten minutes.
- It is universal. Every colleague can open it, and every statistical package can read a CSV.
- It excels at exploration. Sorting, filtering, building a pivot table to see the shape of the data: nothing is faster.
- It depends on nobody. The file is there, on disk, with no vendor to ask for access.
The six properties a spreadsheet cannot carry
These are not absent features that a plug-in would add. They follow from what a spreadsheet is: a grid of free cells, where structure is a convention between people rather than a constraint of the file.
| Property | In a spreadsheet | What it costs |
|---|---|---|
| Typing | A cell accepts anything. An "age" column can hold 54, ~50 and unknown | Cleaning happens at analysis, by retrospective interpretation |
| Declared unit | It lives in the header at best, and is not carried by the value | Two laboratories, two units, one wrong aggregation |
| Origin of a value | No place is provided to say where a figure came from | The value is unverifiable once the record is closed |
| Reasoned log | Version history says *that* a value changed, never *why* | A correction cannot be distinguished from a new measurement |
| Database freeze | Nothing separates capture time from analysis time | Nothing technically prevents picking an endpoint after seeing results |
| Qualified absence | One empty cell, for three different situations | The missingness mechanism is unknown, so imputation is arbitrary |
The characteristic incidents
These are known, documented, and reflect no particular carelessness — they follow from the tool's model.
- Automatic format recognition. A spreadsheet interprets anything resembling a date, a fraction or scientific notation. The phenomenon is widespread enough in genomics to have led a nomenclature body to rename certain gene symbols.
- Partial sorting. Sorting a range without including every column detaches rows from each other. The error is silent and often permanent.
- Formula drift. One inserted row, and a formula that referenced
B12now references something else. The result stays plausible. - Proliferating copies.
thesis_v2_final_corrected.xlsxcirculates by email, and nobody knows which one is authoritative. - Paste that bypasses validation. A spreadsheet's validation rules apply to typing; a paste goes straight through them.
The tipping point
The spreadsheet stops being adequate when at least one of these conditions holds. None depends on the number of rows.
- More than one person enters data into the same dataset.
- The study is destined for a publication that will require describing the capture method.
- Capture spans more than a few weeks, with interruptions.
- Some values are extracted from free text rather than structured fields.
- The dataset must serve a later study as well.
- A third party — committee, reviewer, methodologist — will have to verify what is claimed.
Point-by-point comparison
| Criterion | Spreadsheet | AnzarSeha |
|---|---|---|
| Getting started | Immediate | Create a study, then adjust the dictionary |
| Column typing | None by default | Declared before the first value |
| Units | Informal convention | Declared, normalised to UCUM |
| Value origin | Not provided for | Registry · rule · quoted model · human decision |
| Log | Version history, no reason | Chained log, with reason — see the audit trail |
| Freeze before analysis | None | A study state, dated and signed |
| Missing values | Empty cell | Declared reason, per variable |
| Statistical export | CSV needing rework | Workbook with dictionary, metadata and missing-value table |
| Working as a team | File copies | Authenticated access, per-study isolation |
| Vendor dependence | None | Real — mitigated by export in open formats |
The last row is a genuine disadvantage and it would be dishonest to omit it. It is why the full export — data, dictionary, metadata, missing values, decisions — is available at any time in Excel, CSV, PDF and Word.
Frequent questions
Should the spreadsheet be abandoned entirely?
No. Exploratory analysis, results tables and intermediate calculations remain perfectly at home there. What changes is that the spreadsheet stops being the reference store for the data: it becomes a working tool downstream of a dataset managed elsewhere.
Can an Excel form already in progress be imported?
This is the most common situation, and it is not handled by a simple import: the columns of an existing file must first be typed and documented, which is precisely the work the file never did. Import is possible, but it always comes with a dictionary-building pass.
Is a spreadsheet less secure?
Not intrinsically — an encrypted file on a managed machine can be very secure. The real risk is organisational: files get copied, emailed and end up on personal devices, which is hard to prevent and impossible to establish after the fact.
How many rows before a spreadsheet stops being enough?
Row count is the wrong criterion. A forty-record dataset destined for publication already raises every methodological problem; a ten-thousand-row purely exploratory dataset raises none.