Short on time? This page is the recap. For explanations and worked examples, use the full study guide.
Recap
- Data are values; information is the facts and patterns extracted from data.
- Metadata are data about data (date, size, location). They help find, organise and manage data. Changing or deleting metadata does not change the primary data.
- Cleaning makes data uniform without changing their meaning.
- Challenges at any size: incomplete data, invalid data, combining sources, non-uniform entries.
- Bias comes from the type or source of data; more data does not remove it.
- Correlation does not mean causation; more research is needed.
- Processes: transform every element, filter, combine or compare, visualise.
- Very large datasets need parallel systems; scalability matters.
Key relationships
| Idea | How to recognise it |
|---|---|
| Primary data | The thing itself (photo, song, number list) |
| Metadata | Facts about it (date, author, size) |
| Cleaning | Same meaning, one consistent format |
| Filtering | Keep only items that meet a condition |
| Transforming | Apply the same change to every item |
| Combining | Add, compare or merge data sources |
| Correlation | Two variables move together |
| Bias | Result distorted by who or what was included |
Assumptions and conventions
- When asked about a conclusion, separate what the data show from what they do not show.
- Name the specific third factor or missing data when discussing causation.
- Bias and correlation questions often need a concrete next step: more varied data, or a combined source.
Mistakes to avoid
- Writing “causes” when the data show only a correlation.
- Thinking metadata edits change the primary data.
- Believing a larger sample from the same biased source fixes bias.
- Cleaning in a way that changes meaning (merging different answers).
- Forgetting to account for incomplete records.
- Confusing filtering (keeping some items) with transforming (changing every item).
Two quick worked examples
1. Songs have metadata fields artist, year and length. To list songs by one artist from 2020 onward, filter on two metadata fields; the audio data are never changed or opened.
2. Sales of umbrellas and sales of raincoats rise together in autumn. This is a correlation. It does not show that umbrellas cause raincoat sales; rainy weather is a likely third factor, and weather data would help test that.
A revision routine
Make a small table of six records with a free-text field typed in different ways. Clean it, recount, and note whether the answer changed. Then write one sentence that says what a correlation in the table does show and one that says what it does not show. Finish by listing three challenges that exist at any data size.
Quick self-check
- A song file has the artist name stored with it. Primary data or metadata? (Metadata; the song audio is the primary data.)
- Replacing “NY”, “ny” and “New York” with one form is an example of what? (Cleaning.)
- A survey of 5,000 people all from one club shows a strong preference. Does the large size remove bias? (No; bias from the source remains.)
Next: practice questions.