Data Analytics Statistics Interview Questions and Answers

Statistics is the layer of a data analyst interview where "I can write SQL" turns into "I can be trusted with a decision". Whether you are interviewing for an analyst, business-intelligence or junior data-science role, expect a block of questions that check whether you understand what your numbers actually mean. The good news: the topic list is short and predictable. This page walks the questions that come up in nearly every analyst loop, with the interpretation-first framing interviewers reward.

How to answer statistics questions

Answer from the definition outward, and always land on a plain-English interpretation. A hiring manager is testing whether you could explain the same result to a product owner who has never taken a statistics class. "The standard deviation is 12, meaning most customers fall within about 12 units of the average" beats a formula every time.

Q1. What is the difference between mean, median and mode, and when do you use each?

The mean is the arithmetic average, the median is the middle value when sorted, and the mode is the most frequent value. Use the median when the data is skewed or has outliers, because the mean gets dragged toward extreme values.

The classic example is salary or house-price data: a few very high values pull the mean upward, so the median is the honest "typical" figure. This is why income statistics are almost always reported as medians. Interviewers love to follow up with "your average order value jumped last month — is that good news?" The correct instinct is to ask whether a handful of large orders skewed the mean, and to check the median.

Interview note: Trap: "when are mean and median equal?" In a perfectly symmetric distribution, such as the normal distribution. A gap between them is a quick diagnostic for skew.

Q2. Explain variance and standard deviation.

Variance is the average of the squared differences from the mean; standard deviation is its square root. Standard deviation is preferred for reporting because it is in the same units as the data, so it is directly interpretable.

We square the deviations so that positive and negative differences do not cancel out, and squaring also penalizes larger deviations more heavily. The trade-off is that variance ends up in squared units (squared rupees, squared seconds), which is why the square root brings it back to something a stakeholder can read.

import numpy as np

values = np.array([10, 12, 23, 23, 16, 23, 21, 16])
print(np.mean(values))   # 18.0
print(np.var(values, ddof=1))   # sample variance (divides by n-1)
print(np.std(values, ddof=1))   # sample standard deviation

Interview note: Follow-up: "why ddof=1?" Using n−1 (Bessel's correction) gives an unbiased estimate of the population variance from a sample. Dividing by n underestimates it.

Q3. What is the Central Limit Theorem and why does it matter?

The Central Limit Theorem states that the distribution of sample means approaches a normal distribution as the sample size grows, regardless of the shape of the underlying population — usually for samples of about 30 or more.

This is the reason inferential statistics works at all. Even if customer spend is heavily skewed, the average spend of repeated samples is approximately normal, which lets us build confidence intervals and run hypothesis tests using normal-based methods. Without the CLT, most A/B testing math would not hold.

Interview note: Trap: "does the CLT make the raw data normal?" No — it is about the distribution of the sample mean, not the individual observations.

Q4. Explain a p-value in plain English.

A p-value is the probability of observing a result at least as extreme as the one you got, assuming the null hypothesis is true. A small p-value (commonly below 0.05) suggests the observed effect is unlikely under the null, so you reject it.

The single most common mistake is saying a p-value is "the probability the null hypothesis is true" or "the probability the result happened by chance". It is neither. It is a conditional probability that assumes the null is true from the start. A p-value of 0.03 does not mean there is a 3% chance the null is correct.

Interview note: Follow-up: "is 0.05 a law of nature?" No — it is a convention. The right threshold depends on the cost of a false positive; medical trials use stricter thresholds than marketing tests.

Q5. What are Type I and Type II errors?

A Type I error is a false positive: rejecting a true null hypothesis (concluding an effect exists when it does not). A Type II error is a false negative: failing to reject a false null (missing a real effect). The significance level alpha controls Type I; statistical power (1 − beta) controls the ability to avoid Type II.

The business framing wins points: a Type I error might mean shipping a feature that does nothing, while a Type II error might mean scrapping a feature that actually worked. Lowering alpha reduces false positives but, holding sample size fixed, raises the false-negative rate — so you increase sample size to protect both.

Interview note: Trap: "how do you reduce both errors at once?" Collect more data. A larger sample tightens the sampling distribution and improves power without inflating alpha.

Q6. What is a confidence interval, and how do you interpret 95%?

A confidence interval is a range of plausible values for a population parameter, computed from sample data. A 95% confidence interval means that if you repeated the sampling process many times, about 95% of the intervals constructed this way would contain the true parameter.

The subtle correctness check: it is not correct to say "there is a 95% probability the true mean lies in this specific interval". Once the interval is computed, the parameter is either in it or not. The 95% describes the long-run reliability of the procedure, not a probability about one interval.

Interview note: Follow-up: "what makes an interval narrower?" A larger sample size or lower variability. Higher confidence (99%) makes it wider.

Q7. Correlation versus causation — how do you handle it?

Correlation measures how two variables move together; causation means one drives the other. Correlation never proves causation because of confounding variables, reverse causation or coincidence. Establishing causation typically requires a controlled experiment such as an A/B test.

The interviewer wants to hear a concrete confounder. Ice-cream sales correlate with drowning deaths — the confounder is hot weather, which drives both. A strong analyst flags the confounder rather than declaring a causal story from a correlation coefficient.

Interview note: Trap: "correlation of 0 means no relationship, right?" Only no linear relationship. A U-shaped relationship can have near-zero correlation while being strongly dependent.

Q8. How would you design and evaluate an A/B test?

Define a single primary metric, state the null hypothesis (no difference between control and variant) and alternative, randomly assign users, pick a significance level and power to size the sample in advance, then run to the planned sample before checking significance.

The failure modes are what separate levels. Peeking at results and stopping as soon as p dips below 0.05 inflates the false-positive rate dramatically. Testing many variants without correction (the multiple-comparisons problem) does the same. And a statistically significant 0.1% lift may not be worth shipping — practical significance matters as much as statistical significance.

from scipy import stats

# control: 200 conversions of 2000; variant: 240 of 2000
count = [240, 200]
nobs = [2000, 2000]
from statsmodels.stats.proportion import proportions_ztest
stat, pvalue = proportions_ztest(count, nobs)
print(round(pvalue, 4))   # compare against alpha (e.g. 0.05)

Interview note: Follow-up: "the p-value is 0.04 but the lift is tiny — do you ship?" Weigh effect size, cost and risk. Significance is necessary, not sufficient.

Q9. What is skewness, and how does it change your analysis?

Skewness measures asymmetry in a distribution. Right (positive) skew has a long right tail and mean greater than median; left (negative) skew is the reverse. Skew tells you to prefer the median over the mean and may call for a transformation such as a log before modeling.

Recognizing skew from a quick mean-versus-median comparison is a practical, testable skill. Revenue, session duration and wait times are almost always right-skewed, and reporting their mean without noting the skew is a common junior mistake.

Interview note: Trap: "which is bigger in right skew, mean or median?" The mean, because the long right tail pulls it up.

What interviewers really test

Underneath every question above is one skill: can you turn a number into a decision and explain it without jargon? Practice saying each answer to an imaginary product manager. Pair this set with the probability questions, which share the same reasoning muscles, and if you are early in your journey, the freshers question set frames these fundamentals at entry level. A structured Data Analytics path and a focused mock interview are the fastest way to move from reciting definitions to interpreting results under pressure.