Good analysis starts with a trustworthy source

You find a dataset that looks perfect for your project. It has the right columns, covers the right period and downloads in seconds. It is tempting to open it and start building charts straight away.

Before you do, stop and ask a simpler question: should I trust this data?

Careful cleaning and clever analysis cannot rescue a dataset that is inaccurate, incomplete or out of date. If the source is weak, every conclusion built on it is weak too.

A helpful way to check a source is the ROCCC framework, a checklist popularised by the Google Data Analytics Certificate. ROCCC stands for Reliable, Original, Comprehensive, Current and Cited. Each letter is a question to ask before you rely on a dataset.

R: Is it reliable?

Reliable data is accurate, complete enough for its purpose and collected in a way that does not unfairly favour one result.

Questions to ask:

  • Who collected the data, and how?
  • Is the method described, including sample size and how respondents or records were selected?
  • Are there known errors, corrections or limitations?
  • Do totals and subtotals add up?

A warning sign is a dataset with no explanation of how it was produced. If you cannot find out how the data was collected, you cannot judge how much to trust it.

O: Is it original?

Original means you are using the data from the organisation that produced it, or can trace it back there.

Data often passes through several hands before it reaches you. A survey is published by a statistics agency, summarised in a news article, copied into a blog post, then uploaded to a data-sharing site by someone else. At each step, figures can be rounded, filtered, mislabelled or misquoted.

Whenever you can, go back to the primary source. If you found a figure in a news report, look for the original survey report or database it came from. If you downloaded a dataset from a sharing platform, check whether the uploader explains where it came from and whether that source still publishes it.

C: Is it comprehensive?

Comprehensive data contains what you need to answer your question without important gaps.

Check the coverage carefully:

  • Time: Does it cover the full period you need, or are some months missing?
  • Place: Does it include every location you want to compare?
  • Population: Does it include everyone you are making claims about? Facility data that only covers public facilities cannot describe the whole health system.
  • Variables: Does it include the fields needed for your analysis, such as dates, categories or identifiers to join with other data?

A dataset can be accurate and still be the wrong one for your question if it leaves out the part you most need.

C: Is it current?

Current data is recent enough to be relevant for the decision or question at hand.

Look for two dates: when the data was collected and when it was published. They can be a year or more apart. Then consider how quickly the subject changes. Prices, exchange rates and disease outbreak figures can change in weeks. Population structure or road networks change much more slowly.

Older data is not automatically useless. It may be the best available, or it may be exactly what you need for a trend analysis. The important thing is to state the date clearly and avoid presenting old figures as a picture of today.

C: Is it cited?

Cited data is referenced by credible organisations and comes with enough documentation for you to reference it yourself.

Good signs include a clear publisher, a publication date, a methodology document or data dictionary, a version number and a licence explaining how the data may be used. If government agencies, universities or established research organisations cite the same source, that adds confidence.

If you cannot work out how you would cite a dataset in your own report, that is a sign you do not yet know enough about it.

Putting ROCCC to work: an example

Suppose you are building a portfolio project on food prices and find a spreadsheet titled "Nigeria Food Prices 2019 to 2024" on a public data-sharing website. Here is how a quick ROCCC check might look:

  • Reliable: The description does not explain how prices were collected or from which markets. Some months show identical prices across every state, which looks unlikely.
  • Original: The uploader mentions that the figures came from an official price survey but gives no link. You search for that survey and find the original reports on the publisher's website.
  • Comprehensive: The uploaded file covers 20 states. The original reports cover all 36 states and the Federal Capital Territory.
  • Current: The file stops in mid-2023, despite its title. The original publisher has more recent releases.
  • Cited: The original reports have a clear publisher, dates and methodology notes. The uploaded file has none of these.

The decision is clear. Use the original reports rather than the uploaded copy. Five minutes of checking has given you a more complete, more current and more defensible dataset.

What ROCCC does not do

ROCCC helps you judge whether a source deserves your trust. It does not replace data cleaning or validation. A source can pass every check and still contain duplicates, typing errors or inconsistent categories. You still need to examine the data itself, as we discussed in Five Things That Happen When You Finally Clean Your Data.

It is also a judgement, not a pass or fail test. Sometimes the only available data has weaknesses. In that case, use it carefully and tell your audience what those weaknesses are.

Write a short source note

Make the check visible. For every dataset you use, write a short source note covering:

  • Where the data came from, with a link
  • Who produced it and when it was collected
  • What it covers and what it leaves out
  • Any known limitations that affect your conclusions

This note takes a few minutes and makes your work more credible. In a portfolio project, it shows an employer that you think about evidence, not just tools. In the workplace, it helps colleagues understand how far your findings can be relied on.

Trustworthy analysis starts before the first formula. It starts with knowing where your data came from.