Estimate Population Mean and Conversion Rate Accurately

Quick Overview

Evaluates core statistical inference skills across p-values, confidence intervals, standard error reduction, tail probability estimation, efficient conversion-rate sampling, and truncated-normal estimation. Strong answers state assumptions, use appropriate intervals and weights, and avoid common interpretation errors.

Estimate Population Mean and Conversion Rate Accurately

Company: Google

Role: Data Scientist

Category: Statistics & Math

Difficulty: hard

Interview Round: Onsite

##### Scenario General statistical inference tasks: testing a population mean, controlling standard error, estimating tail probabilities, computing conversion rate, and parameter estimation under truncated normal sampling. ##### Question You collected a sample and want to test whether the population mean differs from 0. What does a p-value of x% mean in this hypothesis-testing context? Given sample mean x̄ = 1 and standard error 0.1, construct the 95% confidence interval for the population mean. What sample size would you need to achieve a standard error of 0.01 instead of 0.1? What actions can you take if the sample size cannot be increased? Given independent observations X₁,…,Xₙ from distribution X, propose an estimator for p = P(X > 10). Construct a 95% confidence interval for P(X > 10) and interpret a resulting interval [a, b] in terms of the true probability p. You have 1,000 binary features and want to estimate the overall conversion rate. Describe how you would design the estimation or sampling strategy. Assume X ∼ N(μ, σ²) but you only observe Y = X conditioned on X > 3 (a truncated normal). How would you estimate μ and σ²? How would you construct 95% confidence intervals for μ and σ² under this truncation setting? ##### Hints Use definitions of p-value, z/t intervals, se = s/√n, plug-in estimator for probability, normal or Wilson CI, sample size formula, MLE for truncated normal, bootstrap/delta method for CI.

Quick Answer: Evaluates core statistical inference skills across p-values, confidence intervals, standard error reduction, tail probability estimation, efficient conversion-rate sampling, and truncated-normal estimation. Strong answers state assumptions, use appropriate intervals and weights, and avoid common interpretation errors.

|Home/Statistics & Math/Google
Google logo
Google
Jul 12, 2025, 6:59 PM
hardData ScientistOnsiteStatistics & Math
76
0

Estimate Population Mean and Conversion Rate Accurately

You are asked a series of statistical inference questions covering hypothesis testing, confidence intervals, sampling design, tail probability estimation, conversion-rate estimation, and estimation under truncation.

Constraints & Assumptions

  • Assume independent sampling unless a question states otherwise.
  • Use standard large-sample approximations when appropriate, and state when a small-sample or exact method would be better.
  • Explain interpretations in plain language, not only formulas.
  • For the truncated normal question, account for truncation explicitly instead of treating the observed sample as an ordinary normal sample.

Clarifying Questions to Ask Guidance

  • Are sample sizes large enough for normal approximations?
  • Is the standard deviation known or estimated?
  • Are the samples simple random samples, stratified samples, or case-control samples?
  • Is the truncation threshold known and applied consistently?

Part 1 - Interpret a P-Value

You test whether a population mean differs from 0 using a two-sided test. What does a p-value of x% mean?

What This Part Should Cover Guidance

  • Probability of observing a result at least as extreme as the sample result under the null hypothesis.
  • Clear rejection of common misinterpretations, such as treating the p-value as the probability the null is true.

Part 2 - Construct a Confidence Interval

Given sample mean x_bar = 1 and standard error SE = 0.1, construct a 95% confidence interval for the population mean and state assumptions.

What This Part Should Cover Guidance

  • Normal or t critical value depending on assumptions.
  • The interval x_bar +/- critical_value * SE .
  • Interpretation of the confidence procedure.

Part 3 - Reduce Standard Error

What sample-size factor is needed to reduce standard error from 0.1 to 0.01?

What This Part Should Cover Guidance

  • Relationship SE proportional to 1 / sqrt(n) when variance is stable.
  • Squared ratio of old to new standard error.
  • Implication that sample size must increase by a factor of 100.

Part 4 - Improve Inference Without More Sample

If you cannot increase sample size, what actions can improve inference, precision, or power?

What This Part Should Cover Guidance

  • Variance reduction, stratification, blocking, covariate adjustment, better measurement, robust methods, or justified one-sided/non-inferiority framing.
  • Trade-offs and assumptions for each option.

Part 5 - Estimate a Tail Probability

Given independent observations from a distribution X, propose an estimator for P(X > 10), construct a 95% confidence interval, and interpret an interval [a, b].

What This Part Should Cover Guidance

  • Indicator estimator for exceedance probability.
  • Standard error and interval options such as normal approximation, Wilson, or exact interval.
  • Correct frequentist interpretation of confidence intervals.

Part 6 - Estimate Overall Conversion with Many Binary Features

You want to estimate the overall conversion rate in a population where each unit has 1,000 binary features. Describe an efficient, unbiased or approximately unbiased estimation and sampling strategy.

What This Part Should Cover Guidance

  • Simple random sampling as a baseline.
  • Stratified sampling, post-stratification, inverse-probability weighting, calibration, or model-assisted estimation.
  • Importance of known inclusion probabilities and representative population margins.
  • Design-effect and variance estimation.

Part 7 - Estimate a Truncated Normal

Assume X is normally distributed with mean mu and variance sigma^2, but you only observe Y = X | X > 3. How would you estimate mu and sigma^2, and how would you construct 95% confidence intervals?

What This Part Should Cover Guidance

  • Truncated-normal likelihood rather than naive sample mean and variance.
  • Maximum likelihood estimation for mu and sigma , solved numerically if needed.
  • Confidence intervals using observed Fisher information, delta method, likelihood profiling, or parametric bootstrap.

What a Strong Answer Covers Guidance

A strong answer explains each inference task accurately, states assumptions, chooses appropriate approximations, and avoids common interpretation errors around p-values, confidence intervals, sampling weights, and truncation.

Follow-up Questions Guidance

  • When would you use a Wilson interval instead of a normal interval for a proportion?
  • How would you report uncertainty under a complex sampling design?
  • What bias occurs if you ignore truncation in the normal-estimation problem?
Loading comments...