Learning data analysis: questions before tools

Data analysis is often taught as a list of tools: spreadsheets, SQL, Python, dashboards. Tools matter, but they are not the skill. The skill is asking a clear question, preparing data that can answer it, examining that data honestly and communicating what it shows. This article sets out that process and the habits that make an analyst trustworthy.

Begin with a question

Every analysis should start with a question specific enough to answer: not "what does our data say?" but "which month had the highest enrolment, and did it follow a campaign?" A clear question determines which data is needed, which comparisons matter and what a useful answer looks like.

Most of the work is preparation

Real data is incomplete, inconsistent and duplicated. Dates are entered in different formats, names are spelled differently and values are missing. Cleaning and organising data usually takes more time than any other step. Hadley Wickham's principle of "tidy data" (2014) offers a useful standard: each variable in its own column, each observation in its own row, and each type of observational unit in its own table. Data in this shape is far easier to analyse and less prone to error.

Look before you summarise

In 1977 John Tukey argued for exploratory data analysis: looking at data visually and openly before testing formal hypotheses. Francis Anscombe had made the point vividly in 1973 with four small datasets that share almost identical means, variances and correlations yet look completely different when plotted. Summary statistics alone can conceal what matters; a simple chart often reveals it.

Compare like with like

• Use rates and percentages, not only totals, when groups differ in size.
• Compare equivalent periods, such as month with month or term with term.
• Check whether a change is larger than normal variation before calling it meaningful.
• Remember that correlation does not establish cause.

A worked example

Suppose a school asks whether its online advertising increased enrolment. A careful analysis would proceed in steps:

• Define the question precisely: did monthly enrolments in the three months after the campaign exceed the same months last year?
• Gather enrolment records with dates, and the campaign dates.
• Clean the data: standardise dates, remove duplicate registrations, and note any months with missing records.
• Plot enrolments by month for both years on one chart before calculating anything.
• Compare the relevant months, and consider other explanations, such as a new programme opening or the school calendar.

The conclusion should state what the data shows, how confident one can be, and what it cannot show. A rise after a campaign is consistent with an effect, but it does not prove one on its own.

Communicate the answer

An analysis is useful only if others understand it. Lead with the answer to the question, support it with one or two clear charts, and state the limitations of the data honestly. A good chart has a title that states the finding, labelled axes and no unnecessary decoration.

Common chart mistakes

• Bar charts whose vertical axis does not start at zero, which exaggerate differences.
• Pie charts with many slices, which are hard to compare.
• Three-dimensional effects, which distort size.
• Too many colours, so that nothing stands out.
• Missing units, labels or sources.

Integrity in analysis

Analysts influence decisions, so honesty is part of the craft. Choosing a convenient time range, omitting inconvenient data or using a misleading axis can make a chart say something the data does not support. The Qur'an commands justice even against one's own interest:

The principle applies directly to how evidence is presented.

A practical learning path

• Become fluent in a spreadsheet: sorting, filtering, formulas and pivot tables.
• Learn to clean and reshape data reliably.
• Learn to build clear charts and a simple dashboard.
• Add a query or programming language once the fundamentals are secure.
• Complete a project on a dataset you care about, from question to presentation.