← Google Interview Insights

Google·Data Scientist·Technical Phone Screen·Intermediate

Intermediate
Jun 2026

Summary

Statistics-heavy technical phone screen for a Data Scientist role at Google. Four questions back to back, all focused on probability and estimation fundamentals. Not a coding round at all, which I wasn't fully expecting.

Questions Asked (4)

Q1

Given an i.i.d. sample from a Normal distribution, derive the maximum-likelihood estimators for the mean and variance.

Technical Trade-offsData Modeling
Author's notes

This is the kind of question that looks easy on paper and then you second-guess yourself mid-derivation.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by writing the likelihood function for an i.i.d. Normal sample, then take the log and derive the partial derivatives with respect to μ and σ². Set the derivatives to zero and solve to obtain the MLEs, noting that the variance estimator is biased and uses n in the denominator.

Pro tip: Mention that the MLE for variance is biased and that the unbiased estimator uses n-1, showing awareness of practical implications. Also, connect the MLEs to the method of moments and discuss their asymptotic properties like consistency and efficiency.

1. Write the likelihood function

For an i.i.d. sample x₁, ..., xₙ from N(μ, σ²), the joint density is the product of individual normal densities. Write L(μ, σ²) = ∏ (1/√(2πσ²)) exp(-(xᵢ - μ)²/(2σ²)).

2. Take the log-likelihood

Simplify by taking the natural logarithm: ℓ(μ, σ²) = -n/2 log(2π) - n/2 log(σ²) - (1/(2σ²)) Σ (xᵢ - μ)².

3. Derive partial derivatives

Compute ∂ℓ/∂μ and ∂ℓ/∂σ². For μ: ∂ℓ/∂μ = (1/σ²) Σ (xᵢ - μ). For σ²: ∂ℓ/∂σ² = -n/(2σ²) + (1/(2σ⁴)) Σ (xᵢ - μ)².

4. Set derivatives to zero and solve

Set ∂ℓ/∂μ = 0 to get μ̂ = (1/n) Σ xᵢ = x̄. Set ∂ℓ/∂σ² = 0 and substitute μ̂ to get σ̂² = (1/n) Σ (xᵢ - x̄)².

5. Verify and discuss properties

Check second derivatives for maximum. Note that μ̂ is unbiased, but σ̂² is biased; the unbiased estimator is s² = (1/(n-1)) Σ (xᵢ - x̄)². Mention consistency and asymptotic normality.

Key Points to Mention

  • Likelihood function for i.i.d. normal sample
  • Log-likelihood simplification and partial derivatives
  • MLE for mean is the sample mean (unbiased)
  • MLE for variance uses n in denominator (biased)
  • Unbiased variance estimator uses n-1
  • Asymptotic properties: consistency, efficiency, normality

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q2

For two random variables X and Y with a known joint pdf, derive the conditional distribution of X given Y.

Data Modeling
Author's notes

Straightforward if you remember the formula.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clearly defining the conditional distribution formula for continuous random variables: f_{X|Y}(x|y) = f_{X,Y}(x,y) / f_Y(y), where f_Y(y) is the marginal pdf of Y. Then explain how to obtain the marginal pdf by integrating the joint pdf over all possible values of X, and finally discuss the importance of the support and any assumptions (e.g., f_Y(y) > 0).

Pro tip: Emphasize that the conditional distribution is only defined when the marginal pdf of Y is positive at the given value, and mention that in practice, you might need to handle cases where the joint pdf is zero or the marginal is zero. This shows attention to mathematical rigor and practical edge cases.

1. State the definition

Write the formal definition of the conditional pdf: f_{X|Y}(x|y) = f_{X,Y}(x,y) / f_Y(y), for f_Y(y) > 0.

2. Compute the marginal pdf of Y

Integrate the joint pdf over the entire range of X: f_Y(y) = ∫ f_{X,Y}(x,y) dx. Ensure the integration limits are correct based on the support of X.

3. Substitute and simplify

Plug the marginal pdf into the conditional formula and simplify the expression if possible. Specify the support of the conditional distribution.

4. Verify properties

Check that the resulting conditional pdf integrates to 1 over x for any fixed y, and discuss any conditions or edge cases (e.g., when f_Y(y)=0).

Key Points to Mention

  • Conditional probability density function formula for continuous variables
  • Marginal probability density function and how to compute it via integration
  • Support of the random variables and domain of integration
  • Condition f_Y(y) > 0 for the conditional distribution to be well-defined
  • Relationship to Bayes' theorem for continuous distributions
  • Potential need for numerical integration if the marginal is not analytically tractable

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q3

Write out the pdf and cdf of the standard Normal distribution.

Data Modeling
Author's notes

I blanked on the exact constant in the pdf for a second.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clearly stating the PDF and CDF formulas for the standard normal distribution, using proper notation. Then briefly explain the parameters (mean 0, variance 1) and mention that the CDF is the integral of the PDF. Finally, connect these to practical applications in data science, such as hypothesis testing and confidence intervals.

Pro tip: Emphasize that the standard normal is a special case of the normal distribution with μ=0 and σ=1, and mention that the CDF is often denoted by Φ(z). This shows you understand the notation and can communicate precisely.

1. Define the standard normal distribution

State that it's a normal distribution with mean 0 and variance 1, denoted as Z ~ N(0,1).

2. Write the PDF

Present the probability density function: f(z) = (1/√(2π)) * e^(-z²/2). Explain each component briefly.

3. Write the CDF

Present the cumulative distribution function: Φ(z) = ∫_{-∞}^{z} (1/√(2π)) * e^(-t²/2) dt. Note that it has no closed-form expression in elementary functions.

4. Discuss properties and applications

Mention symmetry, the 68-95-99.7 rule, and how the CDF is used to compute p-values and critical values in statistical tests.

Key Points to Mention

  • Standard normal has mean 0 and variance 1.
  • PDF formula: f(z) = (1/√(2π)) e^(-z²/2).
  • CDF formula: Φ(z) = ∫_{-∞}^{z} f(t) dt.
  • CDF is often denoted by Φ(z) and is related to the error function.
  • The distribution is symmetric around 0.
  • Used in z-tests, confidence intervals, and the central limit theorem.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q4

Starting from a concrete estimation problem, what assumptions are needed for an estimator to be unbiased, and how do you justify each one?

A/B Testing & ExperimentationTechnical Trade-offs
Author's notes

This one surprised me a bit because it's more conceptual than mechanical.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start with a concrete estimation problem, such as estimating the average treatment effect in an A/B test. Then, systematically outline the assumptions required for an estimator to be unbiased, explaining each assumption's role and how to justify it in practice. Emphasize the importance of validating assumptions through data diagnostics and domain knowledge.

Pro tip: Connect each assumption to potential real-world violations and their impact on bias, showing you can anticipate and mitigate issues like selection bias or measurement error. This demonstrates practical maturity beyond textbook definitions.

1. Choose a concrete estimation problem

Select a relevant problem, e.g., estimating the mean of a population or the average treatment effect in an A/B test. Clearly define the estimand and the estimator.

2. List key assumptions for unbiasedness

Identify assumptions such as random sampling, independence, correct model specification, no measurement error, and exogeneity. Explain why each is needed for the estimator to be unbiased.

3. Justify each assumption

For each assumption, discuss how it can be justified using domain knowledge, study design (e.g., randomization), or statistical tests. Mention consequences if violated.

4. Discuss practical implications and trade-offs

Explain how violations lead to bias and what trade-offs exist (e.g., bias vs. variance). Highlight methods to detect and correct violations, such as robust estimators or sensitivity analysis.

Key Points to Mention

  • Random sampling or random assignment ensures representativeness and independence.
  • Independence of observations prevents correlated errors that bias estimates.
  • Correct model specification: the functional form and included variables must match the true data-generating process.
  • No measurement error in variables, or use of errors-in-variables models if present.
  • Exogeneity: no omitted variable bias or simultaneity; can be justified via randomization or instrumental variables.
  • Practical justification: use of randomization in A/B tests, diagnostic checks (e.g., residual plots, overidentification tests), and sensitivity analyses.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.