What Data Analysis Is, and How to Ask a Good Question
Follow a small workshop from a vague hunch to a clear question, and learn the seven steps that lead from data to a decision.
In this lesson you will
- Explain what data analysis is for
- List the seven steps from question to decision
- Turn a vague question into a measurable one
- Tell describing from comparing, trends and cause
About 22 minutes
“Should we open a second evening?”
Leila runs Spoke and Co, a volunteer bike workshop in a church hall. Last month a neighbour asked her, “Why not open on a second evening?” Leila has a feeling that the workshop is busy, but a feeling is not an answer, and a second evening means finding more volunteers. She needs evidence. That is what data analysis is: turning facts you have collected into an answer you can explain, so that a decision rests on more than a hunch.
Throughout this course you will follow Leila’s workshop. All the visits, names and money are made up for teaching, and the money is sample money, so nothing here describes a real business.
The path from question to decision
- Ask a question that a decision depends on.
- Measure: decide which facts would answer it.
- Collect those facts into a table.
- Clean the table so that the facts can be trusted.
- Analyse: count, compare and look for patterns.
- Explain what you found, and how sure you are.
- Decide, and later check whether the decision worked.
Beginners tend to rush to step five, because that is the part with the numbers. Most of the quality of an analysis is decided in steps one to four. A brilliant calculation on the wrong question, or on dirty data, is a confident wrong answer.
What makes a question good
- It names something you can measure, such as visits, minutes, cost or rating. “How is the workshop doing?” names nothing.
- It has a number or a time period, such as “a week” or “more than twelve”. That keeps you from arguing about what the answer means afterwards.
- It leads to a decision. If every possible answer would leave you doing the same thing, the question is only curiosity.
- It is small enough to answer with the data you can really get.
Vague questions are not wrong, they are unfinished. “Do people like us?” is a fine start. The work is to turn it into something you can measure, for example “What share of visits get a rating of four or five?”. Here is a small program that scores four questions against the first three rules. It is a crude teaching tool made of simple word checks, and it will miss good questions that use other words, but it shows the idea.
def check(question):
q = question.lower()
return {
"names a thing to measure": any(w in q for w in ("visits", "minutes", "cost", "rating")),
"has a number or a time": any(c.isdigit() for c in q) or "week" in q or "month" in q,
"leads to a decision": any(w in q for w in ("should", "open", "hire", "raise")),
}
questions = [
"How is the workshop doing?",
"Do people like us?",
"What is the average cost of a repair?",
"Should we open a second evening if there are more than 12 visits a week?",
]
for number, text in enumerate(questions, 1):
result = check(text)
print("question " + str(number) + ": " + str(sum(result.values())) + " of 3")
weekly = [4, 5, 6, 5, 6, 5]
average = sum(weekly) / len(weekly)
print("average visits a week:", round(average, 1))
print("open a second evening:", "yes" if average > 12 else "no")
question 1: 0 of 3
question 2: 0 of 3
question 3: 1 of 3
question 4: 3 of 3
average visits a week: 5.2
open a second evening: no
The first two questions score 0 and 0 out of 3, so they cannot yet guide anything. The third scores 1 out of 3: it names a measure, but it has no time period and no decision attached. The fourth scores 3 out of 3, so it is ready to work with. The program then uses the six weekly counts to answer it. The workshop averages 5.2 visits a week, far below the made-up bar of twelve, so the answer today is no, with an open door to ask again when the numbers change.
Description, comparison, trend and cause
Questions come in four levels of difficulty. Describing asks “how many?” and “how much?”. Comparing asks which group is higher or lower. Trends ask what is changing over time. Cause asks why, and it is by far the hardest, because data on its own rarely proves it. This course teaches the first three well and warns you about the fourth in lesson 8.
Collecting data first and hoping a question will appear. You end up with a pile of numbers that nobody knows what to do with. Ask the question first, then decide what to measure.
The workshop, the questions and the threshold of twelve visits are made up, and the checker is a simple word matcher. A real decision about opening hours would also weigh volunteers, cost and what neighbours want.
Practice for this lesson
Check what you learned
Five quick questions. Pick an answer and you will see straight away whether it is right and why.
What is data analysis?
- Collecting as many numbers as possible
- Turning collected facts into an answer that supports a decision
- Drawing charts
- Using a spreadsheet
Leila needed evidence for a decision about a second evening.
Which step usually decides most of the quality of an analysis?
- The final calculation only
- Choosing a colourful chart
- Writing the title
- Asking the question and preparing the data
A brilliant calculation on the wrong question is a confident wrong answer.
Which question is the most ready to work with?
- Should we open a second evening if there are more than 12 visits a week?
- How is the workshop doing?
- Do people like us?
- What is going on?
It scored three of three on measure, number and decision.
Why should a question lead to a decision?
- Decisions are required by law
- Questions must be short
- If every answer leaves you doing the same thing, it is only curiosity
- Data cannot be collected otherwise
The decision gives the analysis a purpose.
Which kind of question is the hardest to answer from data alone?
- How many there are
- Which group is higher
- What changed over time
- Why something happens
Cause is the hardest because data rarely proves it by itself.
Summary
- Data analysis turns facts into an answer that supports a decision.
- Most of the quality is decided before the calculation: the question and the data.
- A good question names a measure, has a number or time, and leads to a decision.
- Describing, comparing and trends are easier than proving cause.
Questions
Do I need to be good at maths?
Not at the start. This course uses counting, averages and percentages. The computer does the arithmetic, and your job is to ask sensible questions and check the answers.
Do I need a lot of data?
No. Small tables can answer small questions. What matters more is that the data is clean and suits the question, and that you are honest about how little it may show.
Which tool should I learn first?
The ideas are the same in a spreadsheet, in Python or in a database. This course uses Python with pandas so that every step is visible and repeatable, and it describes spreadsheet equivalents where they help.
