QUIZ
Statistics
11 questions on Statistics, from 12 lessons. Pick an answer in your head, then open it.
On the roadmap for: ML Engineering, Data Science.
Question 1 of 11
When would you report the median instead of the mean?
- A. When the data is skewed or has outliers, such as incomes
- B. When you need an unbiased estimate of the population mean
- C. Only for categorical data, where the mean doesn't exist
- D. When the data is symmetric, so both give the same answer
View answer
✓ A. When the data is skewed or has outliers, such as incomes
A few billionaires move the mean a lot but barely move the median.
Question 2 of 11
You study only startups that are still running to find what makes startups succeed. What bias is this?
- A. Measurement bias
- B. Recall bias
- C. Survivorship bias
- D. Confirmation bias
View answer
✓ C. Survivorship bias
The failed startups, which might share the same traits, are missing from your data.
Lesson: Sampling and Bias
Question 3 of 11
What does a 95% confidence interval promise?
- A. The point estimate is within 5% of the true value
- B. Across repeated studies, 95% of intervals built this way contain the true value
- C. There is a 95% chance the true value lies inside this particular interval
- D. 95% of the observed data points fall inside the interval
View answer
✓ B. Across repeated studies, 95% of intervals built this way contain the true value
In frequentist statistics the 95% describes the procedure, not this one interval. The Bayesian credible interval is the one with the first meaning.
Question 4 of 11
What is a p-value?
- A. The probability that the null hypothesis is true, given the data
- B. The probability that the observed effect is practically important
- C. The probability of data at least this extreme if the null were true
- D. The probability of making a mistake if you go ahead and reject the null
View answer
✓ C. The probability of data at least this extreme if the null were true
It measures surprise under the null. It says nothing directly about how likely the null is or how big the effect is.
Question 5 of 11
You check an A/B test every day and stop as soon as p < 0.05. What's the problem?
- A. The test loses all statistical power after the first look
- B. Each look lowers the p-value, so results look worse
- C. Repeated looks push the false-positive rate well above 5%
- D. Nothing; checking often just finds the answer sooner
View answer
✓ C. Repeated looks push the false-positive rate well above 5%
Each look is another chance to cross the line by luck. Fix the sample size first, or use a sequential test designed for peeking.
Question 6 of 11
You measure the same users before and after a change. Which test fits best?
- A. A paired t-test on each user's change
- B. A two-sample z-test comparing before and after
- C. A chi-squared test on the before and after counts
- D. A one-way ANOVA across the two time points
View answer
✓ A. A paired t-test on each user's change
Pairing compares each user with themselves, which removes between-user variation and gives more power.
Lesson: T-Tests and Z-Tests
Question 7 of 11
You want to compare the means of five groups. Why use ANOVA instead of t-tests between every pair?
- A. ANOVA is more accurate because it uses medians
- B. t-tests can only compare two groups, and only when they have equal sizes
- C. Ten t-tests inflate the false-positive chance; ANOVA tests all at once
- D. ANOVA needs less data per group than a t-test
View answer
✓ C. Ten t-tests inflate the false-positive chance; ANOVA tests all at once
Five groups make 10 pairs. At 5% each, the chance of at least one false alarm is about 40%.
Lesson: Chi-Squared Tests and ANOVA
Question 8 of 11
How does a permutation test build its null distribution?
- A. From a normal table, using the sample mean and variance
- B. By assuming equal variances and using a t-distribution
- C. By bootstrapping each group separately and comparing the two resampled means
- D. By shuffling the group labels many times and recomputing the statistic
View answer
✓ D. By shuffling the group labels many times and recomputing the statistic
If the groups don't differ, labels are exchangeable, so shuffled labels show what differences look like by chance.
Question 9 of 11
In a multiple regression, the coefficient on 'years of experience' is 2,000. What does it mean?
- A. The average salary in the data is 2,000
- B. Each extra year goes with 2,000 more salary, holding other features fixed
- C. Salary rises by 2,000 per year only for people with no other features
- D. Each extra year of experience causes 2,000 more salary for every employee
View answer
✓ B. Each extra year goes with 2,000 more salary, holding other features fixed
Coefficients are partial associations. Calling them causal needs a design that rules out confounders.
Lesson: Inference for Regression
Question 10 of 11
A treatment looks better in each hospital separately but worse when you pool the hospitals. What is this?
- A. Regression to the mean between hospitals
- B. A sampling error that more data would fix
- C. Survivorship bias, because some patients left the study before it ended
- D. Simpson's paradox, caused by a confounder such as case severity
View answer
✓ D. Simpson's paradox, caused by a confounder such as case severity
If one hospital gets sicker patients and uses the treatment more, pooling mixes the groups and flips the trend.
Lesson: Correlation vs Causation
Question 11 of 11
You can't run an A/B test, but one city got a feature and a similar city didn't. Which method compares the before-to-after change between them?
- A. Difference-in-differences
- B. A paired t-test between the two cities
- C. Bootstrapping the treated city's data
- D. Principal component analysis
View answer
✓ A. Difference-in-differences
It subtracts the control city's change from the treated city's change, assuming the two would have moved in parallel without the feature.