HOW SINGAPORE EDUCATION WORKS · Counting people, visits and records
Find your next learning step
Before counting rows in a research dataset, write what one row represents and identify what the research question asks you to count. Five visit records may describe only three people. A correct calculation can still answer the wrong question if its counting unit changes unnoticed.
Full chapter index · Practice and checking · Sources and next reading · How Singapore Education Works
This guide helps university students inspect a small dataset before reporting descriptive results. It covers people, events and repeated records, rather than selecting a statistical model or estimating an effective sample size. For assessed work, use the definitions and analysis plan agreed with your lecturer or supervisor.
All student IDs, visits and measurements below are fictional. Use suitable invented or authorised practice data while learning; real participant records require the permissions and handling arrangements of the project.
The choices at a glance
| What you notice | Possible issue | Useful next move |
|---|---|---|
| Rows exceed participants | People have repeated visits | Count distinct documented participant IDs |
| Same value occurs twice | Different events may share a value | Inspect the record key before deleting |
| A join increases totals | Matching rows may multiply | Check one relationship by hand |
Full chapter index
Count and inspect repeats
Check joins and practise
Transfer and report
Open the dataset documentation before the spreadsheet formula. A row might represent a person, a household, a visit, a questionnaire answer or a person observed on one day. Column headings such as “ID” and “time” do not settle that question by themselves.
Write a concrete sentence: “One row is one visit by one participant to the study booth.” Then identify the fields that should distinguish two legitimate records, such as participant ID and visit ID.
Do the same for the result you want: “I want the number of distinct participants,” or “I want average minutes per visit.” Those are different requests. This is the practical starting point for distinguishing the observations collected from the entities or events your analysis describes.
A participant can make several visits. Each visit may contain several answers. That creates levels: participant, visit and answer. Counting the lowest-level records does not automatically count the higher-level entities.
Use identifiers whose meaning is documented. If two people can share a display name, the name is not a reliable participant key. If visit numbers restart for each person, “Visit 1” alone is not a unique visit key. The combination of participant ID and visit number may be needed.
Keep a plain-language record of which identifiers link the levels. This helps a reader distinguish repeated participation from accidental duplication. It also reveals when the available data cannot answer the intended question: a table of visits without a usable participant identifier cannot reliably tell you how many different people attended.
Assume every fictional visit below is a separate completed visit. Participant IDs are reliable and no records are missing. Minutes measure the same activity in the same way at every visit.
| Participant | Visit number within participant | Minutes |
|---|---|---|
| A | 1 | 3 |
| A | 2 | 4 |
| B | 1 | 8 |
| C | 1 | 6 |
| C | 2 | 6 |
There are five visit rows and three distinct participants. Total recorded time is 3 + 4 + 8 + 6 + 6 = 27 minutes. Mean time per visit is 27 ÷ 5 = 5.4 minutes.
For mean total time per participant, first collect each person's visits: A has 7 minutes, B has 8, and C has 12. Then calculate (7 + 8 + 12) ÷ 3 = 9 minutes per participant across the recorded visits.
Checking answer: both 5.4 and 9 are correct for their stated questions. Reporting “the average participant spent 5.4 minutes across all their visits” changes the denominator and misdescribes the result. Include the unit and observation period in the sentence, not just in a private calculation.
CHAPTER 4 OF 8 · Count and inspect repeats
4. Distinguish legitimate repeats from duplicate records
The two rows with 6 minutes are not duplicates: they describe different visits by C. Deleting repeated numerical values would discard real events. Conversely, a row accidentally pasted twice might repeat the same participant and the same visit identifier.
Check a suspected duplicate against the source record and the documented key. If an identifier repeats, investigate rather than automatically deleting one row. Some data formats legitimately contain multiple answers or measurements under one visit identifier.
Keep the original data intact and record any authorised correction in a separate working copy or documented process. A useful correction note states the affected key, what was wrong, the evidence, the action and the resulting row count. “Removed duplicates” is too vague to reconstruct which observations survived.
CHAPTER 5 OF 8 · Check joins and practise
5. Check what happens when tables are joined
Suppose participant A has two visit rows in one table and three questionnaire-answer rows in another. A join using only participant ID can pair every visit with every answer. That produces 2 × 3 = 6 rows for A.
Those six joined rows do not establish six visits, six people or six independently collected questionnaires. They are combinations produced by the join. A total of visit minutes could be repeated three times if calculated carelessly after that operation.
Before joining, state the intended relationship: one participant to many visits, or one visit to several answers. Check the keys, count distinct entities before and after, and inspect one small example by hand. If the output changes the meaning of a row, write a new “one row is …” sentence before computing the next result.
CHAPTER 6 OF 8 · Check joins and practise
6. Practise a count and a reporting sentence
For guided practice, add a sixth row to the worked case: B, visit 2, 3 minutes. Ask what changes before calculating anything. The number of visits rises; the number of distinct participants does not.
Checking answer: six visits, three participants and 30 recorded minutes. Mean minutes per visit is 30 ÷ 6 = 5. Mean total minutes per participant is 30 ÷ 3 = 10. B's total becomes 11 minutes, while A and C remain at 7 and 12.
For independent practice, a fictional attendance file contains two sessions for P1, one for P2 and four for P3. Every participant-session pair is distinct. Report the number of participants, attendance events and mean attendance events per participant.
Checking answer: three participants, seven attendance events and 7 ÷ 3, approximately 2.33 events per participant. A clear sentence is: “Across the recorded period, three participants contributed seven attendance events.” It preserves both levels rather than presenting seven as the participant count.
CHAPTER 7 OF 8 · Transfer and report
7. Revisit the question before transferring the method
Later, start with a fresh question rather than the memorised answer. If a table contains one row per household, counting rows does not count household members. If it contains one row per country-year, counting rows does not count distinct countries. First inspect the documentation and identifiers.
The UK's Data Service explicitly identifies the analysis unit in its student dataset guides; its British Social Attitudes guide describes interviewing one adult per sampled household. That is a useful global reminder to read how a dataset is organised rather than infer its meaning from a familiar survey name.
For a Singapore university project, carry the same check from questionnaire design into the final report. Correct entity counts do not, by themselves, establish representative sampling, independent observations or causation. Those are additional research-design questions. Preserve the distinction instead of claiming that a tidy spreadsheet proves a strong study.
Is the row count ever the participant count?
Yes, if the documented structure is exactly one row per participant and the records satisfy that structure. Verify it rather than assuming it.
Can I average the averages for each person?
That gives equal weight to each person's average. Pooling all visits weights people according to their number of visits. Either may serve a defined question, but they are different summaries. State the target before choosing.
Should I always reduce repeated observations to one number?
No. The research question and approved analysis plan determine how repeated observations should be used. Aggregation can remove relevant timing or variation. Seek methodological guidance before inferential analysis.
What if the identifier is missing?
Mark the uncertainty and consult the documentation or authorised data owner. Do not invent an identity from similar names or assume every unidentified row is a different person. Explain how unresolved records affect the reported count.
Sources and next reading
Primary resources checked on 23 October 2026: UK Data Service: British Social Attitudes quick start guide for a documented analysis unit; World Bank: Gathering and Preparing the Data Set for microdata preparation and analysis-unit context. The worked tables, joins and arithmetic in this guide are original fictional examples, not analyses of either source's data.
Continue with laboratory notebooks and research records or writing research limitations, and return to How Singapore Education Works.
Learning at other stages
Continue with a different learning task in this series. Each guide has its own worked cases, practice and limits.