Data Science interview questions and answers

Use a t-test when the variance is estimated from the sample, especially with modest n, because the t distribution has heavier tails.

A two-tailed test flags large deviations in either direction

A chi-square test of independence checks whether two categorical variables are associated in a contingency table.

One-way ANOVA tests whether three or more group means differ, under normality and similar variance assumptions.

Power is the probability of rejecting the null when a specified alternative is true, equal to one minus the Type II error rate.

A tiny effect can be highly significant in a huge sample, and a large effect can miss significance in a small sample.

People treat p as the probability the null is true, or as the probability the result is a fluke in an informal sense.

For a symmetric null distribution, the two-sided p-value is typically twice the one-sided p-value from the observed direction, capped at 1.

P-hacking is trying many analyses, metrics, segments, or stopping times until a significant p-value appears, then reporting that path as if it were planned.

Sample size depends on baseline rate, minimum effect you care about, variance, alpha, and desired power.

The novelty effect is a short-run change in behavior because the UI is new, not because the design is durably better.

A primary metric is the one decision that determines ship or no-ship, which protects you from fishing across twenty dashboards.

MDE is the smallest true effect the test is designed to detect with the planned power, alpha, and sample size.

Random assignment balances both observed and unobserved confounders in expectation, which observational matching cannot guarantee.

groupby.agg reduces each group to summary values and returns a smaller frame keyed by group.

merge joins on keys like SQL, aligning rows by column values or index.

apply with a Python function per row is convenient but often loops in the interpreter and is slow on large frames.

A pivot table reshapes long data into a grid of aggregations, with one field as rows, one as columns, and a value aggregated in cells.

Object columns of strings use much more memory than categorical or string dtypes, and float64 IDs can corrupt large integers.

Broadcasting expands arrays of compatible shapes so elementwise operations work without explicit copies of the data.