Data science
| English | Chinese | Pinyin |
|---|---|---|
| data science | 数据科学 | shù jù kē xué |
| descriptive statistics | 描述性统计 | miáo shù xìng tǒng jì |
| visualisation | 可视化 | kě shì huà |
| correlation is not causation | 相关不等于因果 | xiāng guān bù děng yú yīn guǒ |
The analysis is the short part
- Data science 数据科学 turns data into a decision. The analysis is the short part.
- Cleaning, joining and checking what the data can actually support is where the time goes, and where the credibility comes from.
- The project is marked on whether it answers a question, not on how sophisticated the method looks.
Cleaning
- Real data has missing values, duplicates, inconsistent spellings and impossible entries.
- Decide what to do with each, and write down the decision. Dropping rows silently changes what your result describes.
- ⚠ A dataset with 200 rows and 40 blanks is a dataset of 160 rows plus a question about who the 40 were.
Dropping rows with missing values is a neutral choice that needs no explanation.
It changes who your result describes. Write down the decision and who was removed.
Describing and showing
- Descriptive statistics 描述性统计 summarise: centre, spread, and how many.
- A visualisation 可视化 shows shape, which is what a table of numbers hides.
- Choose the chart for the question: a bar chart compares categories, a line shows change over time, a scatter shows a relationship.
- Label the axes and give the units. An unlabelled chart is decoration and is marked as such.
Match each question to the chart that answers it.
Choose by the question, not by which chart looks best in the report.
A chart without axis labels or units is marked as decoration.
It is the cheapest mark available in a data project and the one most often lost.
Two conclusions from the same finding.
Students who use the library more get better grades.
Overclaim: "Using the library improves grades."
Supported: "Library use and grades are positively associated in this sample. This does not establish cause: students who study more may do both, and the data cannot separate them."
The second is shorter than it looks and scores higher, because it says exactly what the evidence supports and no more.
Library use and grades rise together in your sample. What can you claim?
Something else may cause both. Saying exactly what the evidence supports is what scores.
Correlation is not causation 相关不等于因果 — the sentence every data project needs and most omit. Two things moving together can mean one causes the other, that something else causes both, or that the sample was small.
Write one limitation sentence for a survey of 60 students from your own class.
Example: "All 60 respondents were from one class, so the results may not describe other year groups."
State the limitations of your dataset explicitly: how it was collected, who is missing from it, what period it covers. A project that names what its data cannot show scores above one that quietly overclaims — and an examiner will find the gap you did not mention.