Country or region and language

    Choose your country or region and your language.

    Suggested for you

      All countries and regions

        Choose your character

        It shows in the top bar and on your profile. Change it any time.

        Interviews in Statistics for data roles

        Work in the tool while you explain your decisions.

        Choose a role and level. Each interview has a new task, follow-up questions and a report that checks your work and your explanation.

        Speak or type. Sign in to save your results across devices.

        Timed assessment

        • 12 practical questions
        • 45 minutes
        • Mid level

        Questions are drawn from a bank of 36. Your report includes a skill breakdown and lessons to practice next.

        Practice interviews

        Junior tasks cover the fundamentals. Mid and senior tasks add debugging, changing requirements and review decisions.

        10 interviews

        • Data analyst
        • Junior
        • 30 min

        A sample spread report

        Plover Dispatch recorded sample as a sample of delivery minutes, with one missing observation. Create numeric result containing the sample standard deviation, excluding missing observations. The denominator must use the number of observed values minus one.

        Start interview
        Skills and rubricA sample spread report
        • I exclude the missing observation and calculate sample standard deviation with ddof=1 because these deliveries are a sample of the process. The result is in minutes. I report the observed count separately so the missing record does not masquerade as a zero-minute delivery.
        • The missing observation reduces the observed sample size. I neither fill it with zero nor include it in the denominator because either choice changes the measured spread. Squaring standard deviation yields variance, which has different units.
        • I retain the source records, the saved work and the checked output because another analyst must be able to reproduce this result. I record the observation window and the unresolved limitation. The decision owner approves any exception before this result changes a business action.
        • Data scientist
        • Junior
        • 30 min

        A pooled signup conversion rate

        Lark Subscriptions supplies successes and trials for three disjoint signup cohorts. Create numeric result equal to the overall conversion rate as a share. Every trial belongs to exactly one cohort.

        Start interview
        Skills and rubricA pooled signup conversion rate
        • I divide total successes by total eligible trials because each trial is one opportunity to convert. Cohorts have unequal sizes, so their unweighted mean is a different quantity. I confirm disjoint membership and that each success belongs to the denominator.
        • The smallest cohort contributes fewer trials than the largest. I weight by the trial counts because a cohort label does not create another equally sized population. Overlapping cohorts would prevent this simple pooling.
        • I retain the source records, the saved work and the checked output because another analyst must be able to reproduce this result. I record the observation window and the unresolved limitation. The decision owner approves any exception before this result changes a business action.
        • Data analyst
        • Junior
        • 30 min

        A small-sample confidence interval

        Cove Fulfillment treats the observed sample values as independent measurements from an approximately normal process. Create result as a two-item NumPy array [lower, upper] for a 95% t confidence interval for the mean, omitting the missing observation.

        Start interview
        Skills and rubricA small-sample confidence interval
        • I omit the missing measurement, estimate standard error from the observed sample and multiply by the t critical value with n minus one degrees of freedom because population spread is unknown. The interval estimates the process mean under the stated independence and shape assumptions.
        • There are only six observed measurements. I use t rather than a normal critical value because spread was estimated from this small sample. The interval concerns the mean, so its uncertainty uses standard error rather than individual-delivery spread.
        • I retain the source records, the saved work and the checked output because another analyst must be able to reproduce this result. I record the observation window and the unresolved limitation. The decision owner approves any exception before this result changes a business action.
        • Data scientist
        • Mid
        • 30 min

        An unequal-variance supplier comparison

        Tern Purchasing recorded independent delivery times in first and second for two suppliers. Create numeric result equal to the two-sided Welch t-test p-value for equal population means. Keep the samples independent and allow unequal variances.

        Start interview
        Skills and rubricAn unequal-variance supplier comparison
        • The groups have unequal sample sizes and visibly different spread. I do not pair rows by their array position because those positions do not identify matched deliveries. A one-sided alternative would need a direction chosen before observing this result.
        • I define the two-minute equivalence margin and use an equivalence analysis because failure to reject equal means does not establish equivalence. I retain the difference interval and check whether the data are precise enough for the stated practical margin.
        • I retain the source records, the saved work and the checked output because another analyst must be able to reproduce this result. I record the observation window and the unresolved limitation. The decision owner approves any exception before this result changes a business action.
        • Data analyst
        • Mid
        • 30 min

        A matched process improvement interval

        Reed Operations measures the same six workstations before and after a process change. Create result as [lower, upper] for the 95% t confidence interval of mean before minus after. Keep workstation pairing.

        Start interview
        Skills and rubricA matched process improvement interval
        • The rows identify the same workstations before and after. I preserve that match because mixing workstation order changes the differences and their uncertainty. The sign convention must stay before minus after so the recommendation is interpretable.
        • I retain the missing pair as an exception and explain any complete-pair analysis because average imputation would fabricate an observed change and understate uncertainty. I investigate why the measurement is missing before extending the result to all workstations.
        • I retain the source records, the saved work and the checked output because another analyst must be able to reproduce this result. I record the observation window and the unresolved limitation. The decision owner approves any exception before this result changes a business action.
        • Data scientist
        • Mid
        • 30 min

        An exact defect-rate alarm

        Quarry Assembly observes observed defective units out of total independent units. Create numeric result equal to the exact binomial-test p-value against null_rate, with the pre-specified alternative that the defect rate is greater.

        Start interview
        Skills and rubricAn exact defect-rate alarm
        • The observed count itself belongs in the upper tail. I use the binomial test rather than subtracting a cumulative probability at the same count because that shortcut can omit the boundary mass. Independence and the fixed inspection plan remain assumptions.
        • I replace the single fixed-look interpretation with a pre-specified sequential monitoring rule because repeated optional stopping changes the false-alarm rate. I retain the inspection times and counts, then calibrate the alarm to that actual monitoring plan.
        • I retain the source records, the saved work and the checked output because another analyst must be able to reproduce this result. I record the observation window and the unresolved limitation. The decision owner approves any exception before this result changes a business action.
        • Data analyst
        • Mid
        • 30 min

        A sparse campaign outcome table

        Morrow Marketing supplies contingency as a 2 by 2 table: rows are independent campaign groups and columns are converted and not converted. Create numeric result equal to the two-sided Fisher exact-test p-value for association.

        Start interview
        Skills and rubricA sparse campaign outcome table
        • A small cell makes an asymptotic approximation a different procedure. I keep the observed table intact because replacing counts with percentages loses the sample-size information the test needs. Group independence still has to be justified by collection.
        • I need paired transition counts because the independent-group table discards which customer changed outcome. I would use a matched binary analysis on the discordant pairs and state the observation window before interpreting campaign effects.
        • I retain the source records, the saved work and the checked output because another analyst must be able to reproduce this result. I record the observation window and the unresolved limitation. The decision owner approves any exception before this result changes a business action.
        • Data scientist
        • Senior
        • 30 min

        A family of launch-metric tests

        Alder Product pre-specified four launch metrics in one confirmatory family. Create result with the Holm-adjusted p-values for raw_p in their original metric order. Control familywise error at alpha.

        Start interview
        Skills and rubricA family of launch-metric tests
        • I label the fifth result exploratory and retain the original confirmatory family because post-launch selection must not silently redefine the original claim. If the owner wants one expanded confirmatory family, I recompute and document that revised analysis before making a joint claim.
        • I retain the source records, the saved work and the checked output because another analyst must be able to reproduce this result. I record the observation window and the unresolved limitation. The decision owner approves any exception before this result changes a business action.
        • The strongest metric does not make the remaining tests disappear. I keep all four original p-values in the family because dropping inconvenient outcomes after inspection changes the error-control claim. Holm applies decreasing multipliers to the ordered p-values.
        • Data analyst
        • Senior
        • 30 min

        A two-arm sample-size plan

        Brookline Research plans equal independent arms for a two-sided mean comparison. Using the stated normal approximation, create numeric result as the ceiling of 2*(z_(1-alpha/2)+z_power)^2/effect^2, the required observations per arm. effect is the standardized mean difference.

        Start interview
        Skills and rubricA two-arm sample-size plan
        • I recalculate for clustered assignment because customers within a store may be correlated and the independent-person formula overstates information. I need a defensible intracluster correlation, cluster sizes and the number of stores before approving recruitment.
        • I retain the source records, the saved work and the checked output because another analyst must be able to reproduce this result. I record the observation window and the unresolved limitation. The decision owner approves any exception before this result changes a business action.
        • The formula returns observations per arm, so total recruitment is twice that ceiling before any attrition allowance. I retain both the factor of two and alpha divided by two because removing either changes the planned power.
        • Data scientist
        • Senior
        • 30 min

        A fitted workload slope

        Cove Services records predictor x as completed tasks and response y as elapsed minutes. Create numeric result equal to the ordinary least-squares slope of y on x with an intercept. Preserve its units as minutes per task.

        Start interview
        Skills and rubricA fitted workload slope
        • I can report the fitted association within these observations because workload may correlate with staffing or task complexity. A causal intervention claim needs a credible experiment or explicit identification assumptions and checks; this regression alone does not supply them.
        • I retain the source records, the saved work and the checked output because another analyst must be able to reproduce this result. I record the observation window and the unresolved limitation. The decision owner approves any exception before this result changes a business action.
        • The slope depends on which variable is the response. I keep y as minutes and x as tasks because reversing and inverting the line does not minimize the requested squared minute residuals. A positive slope alone does not establish that changing task load causes the time change.

        Interview Hub access

        The Interview Hub is included in All Access. Choose a plan or sign in to your account to continue.

        View plansSign in or join

        Your Work Gym account

        Save your progress and pick up on another device. Your browser's progress comes with you.

        or

        You can keep practicing without an account.

        Change your plan

        Checking the price

        Help and feedback

        Ask a question, tell us what is not working, or share an idea. We reply by email.

        What is it about?

        We use it only to reply to you.