Data Analyst Interview Questions

Quantitative data is numeric and measurable (sales, age, counts). Qualitative data is descriptive/categorical (colour, feedback text, department). Many analyses encode qualitative categories so they can be counted or filtered.

Data cleaning fixes problems that would distort analysis: duplicates, missing values, inconsistent formats, typos and outliers. Clean data is the foundation of trustworthy results — "garbage in, garbage out".

• Mean is the average; sensitive to outliers.
• Median is the middle value; robust to outliers, better for skewed data like income.
• Mode is the most frequent value.

Structured data fits neatly into rows and columns (databases, spreadsheets). Unstructured data has no fixed format (emails, images, free text, social posts). Unstructured data needs extra processing before analysis.

A primary key uniquely identifies each row in a table (e.g. CustomerID). A foreign key is a column in one table that references the primary key of another, creating a relationship between tables.

First understand why it is missing. Options: remove rows/columns if few and random; impute with mean/median/mode or a model; or flag with an indicator column. The choice depends on how much is missing and whether the gap itself carries meaning.

Correlation measures how two variables move together (−1 to +1). It does not imply causation — a third factor may drive both, or the link may be coincidental. Proving cause usually needs a controlled experiment or careful study design.

• Bar: compare categories.
• Line: show a trend over time.
• Scatter: show the relationship between two numeric variables.

Pick the chart that answers the question with the least clutter.

A PivotTable aggregates and reshapes data instantly — grouping by dimensions and summarising measures (sum, count, average). It is the fastest way to explore patterns like "revenue by region and month" without writing formulas.

Data visualisation represents data graphically so patterns are easy to see. A good chart answers one clear question, uses the right type for the data, has honest axes, minimal clutter, clear labels, and highlights the key insight rather than decorating.

An inner join keeps only rows that match in both datasets. A left join keeps all rows from the main dataset and adds matches where available (nulls otherwise). Use left join when you must not lose rows from your base table — e.g. all customers, even those with no orders.

A confidence interval is a range that likely contains the true value, with a stated confidence (e.g. 95%). It tells you how precise an estimate is — a narrow interval means more certainty, a wide one means more uncertainty.

Normalisation rescales values to a fixed range (often 0–1). Standardisation rescales to mean 0 and standard deviation 1 (z-score). Standardisation suits data that is roughly normal and algorithms sensitive to variance; normalisation suits bounded features and distance-based methods.

An outlier is a value far from the rest. Common detection: the IQR rule (below Q1−1.5·IQR or above Q3+1.5·IQR) or z-score beyond ±3. Decide case by case whether to keep, cap, or remove it — some outliers are errors, others are the most important signal.

Run an A/B test: split users randomly into control and treatment, keep everything else equal, and compare the key metric. Check that the difference is statistically significant (not just random noise) and that the sample is large enough before concluding.

Start with the headline insight and its business impact, use one clear visual per point, avoid jargon, show the "so what" and a recommended action, and keep the detailed methodology in an appendix. Tailor depth to the audience.

Supervised learning trains on labelled data to predict an outcome (e.g. predict churn from past churn labels). Unsupervised learning finds structure in unlabelled data (e.g. clustering customers into segments). The presence of a target label is the key difference.

When one class is rare, accuracy is misleading. Techniques: resample (oversample the minority / undersample the majority), use synthetic methods like SMOTE, apply class weights, and evaluate with precision, recall and F1 or AUC instead of plain accuracy.

Data storytelling combines data, visuals and narrative to drive a decision. It matters because stakeholders act on understanding, not raw numbers — a clear story with context, a key insight and a recommendation turns analysis into business impact.