What a Data Analyst Does

A manager at a small organization asks a simple question: "Why do some of our locations return more late items than others?" Nobody can answer it by looking at a pile of records. Somebody has to gather the records, check that they are trustworthy, compare the locations fairly and explain what the comparison does and does not show. That person is doing data analysis.

A data analyst works with recorded facts to help people make decisions. This entry describes the typical tasks, the skills and tools involved, a small worked example you can follow by hand, and the ways people learn the work. Job titles vary a great deal here, and the same work may appear under names such as business analyst, reporting analyst or operations analyst. For the other areas, see the map of common IT roles.

The core job: from a question to an answer

"Data analyst" is not a tightly defined title. The closest official description comes from the Bureau of Labor Statistics profile for data scientists, which lists tasks such as deciding which data are available and useful for a project, collecting, categorizing and analyzing data, using visualization software to present findings, and making recommendations to stakeholders. Analyst roles often cover the same ground with less emphasis on building statistical models. Because titles are not standardized, read the task list of any posting.

A typical piece of analysis passes through the same stages.

Ask a clear question

"How are we doing?" is too vague to answer. "Which branch had the highest share of late returns last quarter?" is a question with an answer. Much of an analyst's value is in turning a loose request into a precise one, by asking what decision the answer will support.

Find and collect the data

The numbers may sit in a spreadsheet, a database, an exported file or a reporting tool. The analyst finds where they live and who owns them. Formats such as CSV and JSON files are common, and a database lets you pull rows with a SELECT query.

Clean the data

Real data is messy. Dates appear in several formats, names are spelled differently, and some cells are empty. Cleaning means finding and fixing such problems, or at least noting them, before any conclusion is drawn. The entry on data cleaning basics covers it. Skipping this stage is one of the surest ways to produce a wrong answer.

Analyze

Analysis can be as simple as counting, grouping and comparing, or it can use statistics. The skill lies in choosing a comparison that is fair.

Present the findings

A result nobody understands changes nothing. Analysts build charts, tables, short reports or dashboards, and they say plainly what the numbers do not show.

A worked example you can check by hand

Take the library question from the start. Two branches gave this list of loans from one week. "Late" means returned after the due date.

To compare the branches fairly, the analyst looks at late loans out of all loans, not at raw counts alone. North had 1 late loan out of 4. South had 3 late loans out of 4. South looks worse.

Now the analyst asks the questions that protect against a wrong answer.

1. Is the sample large enough? Four loans per branch is tiny. One extra late return at North would change the picture. A week of data cannot support a firm statement.
2. Are the definitions the same? If South counts a return on the due date after closing time as late and North does not, the comparison is unfair.
3. Is something else different? South may lend items with shorter loan periods, such as popular new books.
4. What does the manager need? If the real goal is to reduce late returns, the next step may be a longer period of data, or a conversation with South's staff, instead of a conclusion.

The finished message might read: "In the one week we checked, South had more late returns than North, but the sample is very small and the rules may differ. We recommend collecting a full quarter before acting." That sentence is honest, short and useful.

Skills and tools

The skills split into three groups.

Tools are examples. A spreadsheet program is a normal starting point and still used by working analysts. SQL is the standard language for asking questions of relational databases, covered in the SQL joins entry once you pass the basics. Dashboard software turns a regularly updated query into a live page. Languages such as Python or R are common for larger or repeated work. No single tool is required, and the questions matter more than the program.

How people learn this work

There is no single route. People reach data work from study in statistics, business, science or computing, from jobs in which they already worked with reports, and from self-directed practice. Employers describe different requirements for the same title, so check each posting.

A solid learning path starts with spreadsheets: sorting, filtering, grouping and charting a small data set you understand. Next comes SQL, practiced on a free database tool. Then comes the habit of explaining a result to someone who was not there. Public data sets from government statistics agencies give you real, messy material, and every project teaches the same lesson: check the data before trusting it.

Keep a small collection of your own analyses with the question, the data source, the steps you took and the limits you found. A reader can then follow your reasoning. That is more convincing than a list of tool names.

Limits and trade-offs

Data work can be slow and detailed. Questions arrive vague, data arrives late, and people sometimes want a specific answer to support a decision they have already made. Analysts need the confidence to say "the data does not show that." Some of the job is repetitive reporting, and some roles are narrow, with one tool and one business area. Others are broad. If you prefer building products, development may suit you better; if you prefer making people's problems go away immediately, support work might.

Official outlook figures for related occupations are in the Occupational Outlook Handbook, which gives the date of each profile. This entry quotes none, because pay and hiring depend on place and time.

Common mistakes

• Comparing raw counts. A big branch will have more late returns than a small one even if it behaves better. Compare shares or rates, and state the totals.
• Trusting data that was never checked. Duplicates and blanks quietly distort results. Look at the data before analyzing it.
• Drawing strong conclusions from tiny samples. Four loans prove nothing. Say how much data you have.
• Mistaking correlation for cause. Two things that move together may both be caused by a third. Describe what you saw, not why, unless you can show why.
• Hiding the limits. A chart without caveats looks more certain than it is. Add a line on what the data cannot say.
• Overdesigning the report. A clear table often beats a complicated chart.

FAQ

Is a data analyst the same as a data scientist?

The titles overlap and employers use them differently. Data scientist roles often involve building models and more programming. Analyst roles often center on reporting and answering business questions. Compare the task lists.

Do I need to know programming?

Not always. Many analysts start with spreadsheets and SQL. Scripting languages become useful for larger or repeated work.

Do I need advanced math?

Basic arithmetic, percentages as concepts, averages and careful comparison cover a great deal. More advanced statistics matter in some roles. Read the requirements of the posting.

What is the difference between a spreadsheet and a database?

A spreadsheet suits small, flexible work done by one person. A database suits larger shared data with rules. The entry on spreadsheet vs database compares them.

Where can I find practice data?

Government statistics agencies publish open data sets, and many cities publish public records. Pick a topic you care about so the questions come naturally. The Careers and Skills section suggests how to plan your practice.