← Meta Interview Insights

Meta·Data Scientist·Onsite - Multi Round·Senior

SeniorPrefer not to say
Jul 2026

Summary

Brutal Meta DS onsite focused entirely on experiment design for a Group Call feature with network interference. One long, multi-part question that spiraled into graph theory, sample size math, and spillover bias. Walked out not totally sure how I did.

Questions Asked (6)

Q1

How would you define the exposure unit and the estimand for an A/B test of a Group Call feature where network interference is expected? For example, should you target intent-to-treat at the cluster level or a per-user average treatment effect?

A/B Testing & ExperimentationProduct Analytics & Metrics
Author's notes

This is where I started to feel the ground shift.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the causal question: define the estimand as the average effect of assigning the Group Call feature to clusters (e.g., groups or social neighborhoods) to capture intent-to-treat (ITT) under interference. Then choose the exposure unit as the cluster (group) to align with the randomization unit and avoid contamination, and discuss how to estimate per-user effects via cluster-level ITT divided by cluster-level engagement (complier average causal effect) if needed.

Pro tip: Emphasize that under interference, the estimand must be defined at the randomization unit level, and per-user effects are only interpretable as a weighted average of cluster-level effects; avoid claiming a pure per-user ATE without strong assumptions.

1. Clarify the causal question and interference mechanism

Identify how interference occurs (e.g., calls within groups affect non-treated members) and what decision the experiment informs (launch, targeting). This determines whether the estimand should be at cluster or user level.

2. Define the exposure unit and randomization unit

Choose the exposure unit as the cluster (e.g., group, social neighborhood) where interference is contained, and randomize at that same level to ensure valid inference. Explain why user-level randomization would violate SUTVA.

3. Specify the estimand: cluster-level ITT vs. per-user ATE

Define the primary estimand as the average treatment effect of assignment at the cluster level (ITT). If a per-user effect is desired, define it as a complier average causal effect (CACE) or a weighted average of cluster-level effects, acknowledging assumptions.

4. Choose estimation and inference methods

Use cluster-level analysis (e.g., t-test on cluster means) or mixed-effects models with cluster random effects. For per-user CACE, use instrumental variables or two-stage least squares with cluster-level assignment as instrument.

5. Validate assumptions and interpret results

Check for spillovers across clusters, partial interference, and cluster size variability. Report both ITT and CACE with confidence intervals, and discuss generalizability to the target population.

Key Points to Mention

  • Stable Unit Treatment Value Assumption (SUTVA) and how interference violates it
  • Cluster-level randomization as a solution to interference
  • Intent-to-treat (ITT) effect at the cluster level as the primary estimand
  • Complier Average Causal Effect (CACE) for per-user effects under non-compliance
  • Design-based vs. model-based inference for cluster-randomized experiments
  • Trade-offs between cluster-level ITT and per-user ATE in terms of interpretability and policy relevance

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

Q2

How would you construct clusters from a social graph for this experiment? Walk through how you'd weight edges for strong ties, choose a clustering rule like connected components after thresholding, and prevent cluster overlap.

A/B Testing & ExperimentationSystem Design
Author's notes

I actually liked this part.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining the social graph and the criteria for strong ties, then explain how you would weight edges based on interaction frequency, recency, and reciprocity. Next, describe a thresholding approach to filter weak edges, apply connected components to form clusters, and outline methods to handle overlaps such as assigning nodes to the cluster with the strongest connection or using a resolution parameter in community detection algorithms.

Pro tip: Emphasize that cluster construction should align with the experiment's goals—e.g., minimizing interference in A/B tests—and mention that you'd validate clusters by checking metrics like modularity or conductance, and consider using overlapping community detection if the experiment allows.

1. Define strong ties and edge weights

Identify signals of strong ties (e.g., frequent interactions, mutual friends, recent communication) and assign weights to edges accordingly, possibly using a composite score.

2. Choose a clustering rule

Select a method like thresholding edge weights followed by connected components, or use community detection algorithms (e.g., Louvain, Leiden) that naturally handle weighted graphs.

3. Prevent cluster overlap

Decide on a strategy: either enforce disjoint clusters by assigning each node to one cluster (e.g., via label propagation or greedy assignment) or allow overlap if the experiment permits, using methods like clique percolation.

4. Validate and iterate

Evaluate cluster quality using metrics like modularity, conductance, or size distribution, and adjust thresholds or algorithms to meet experiment requirements (e.g., balanced cluster sizes).

Key Points to Mention

  • Edge weighting based on interaction frequency, recency, and reciprocity
  • Thresholding to filter weak ties and reduce noise
  • Connected components as a simple clustering rule after thresholding
  • Handling overlap via node assignment to strongest cluster or using overlapping community detection
  • Validation metrics like modularity, conductance, and cluster size distribution
  • Alignment with experiment goals, such as minimizing interference in A/B tests

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

Q3

If you randomize at the cluster level, how do you handle users who appear on edges between treatment and control clusters? What role do holdout buffers or partial saturation designs play?

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

Frontier users.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by acknowledging the core problem: cluster randomization creates spillover and contamination at cluster boundaries. Then explain how holdout buffers and partial saturation designs mitigate these issues by creating geographic or temporal separation between treatment and control units, and discuss the trade-offs in terms of statistical power and bias.

Pro tip: Emphasize that the choice depends on the expected spillover mechanism and the cost of contamination; in practice, a combination of buffer zones and partial saturation often balances bias and power better than either alone.

1. Define the contamination problem

Explain how users on edges between clusters can be exposed to both treatment and control conditions, leading to spillover effects that bias estimates.

2. Introduce holdout buffers

Describe how creating buffer zones (geographic or network-based) between treatment and control clusters reduces spillover by excluding edge users from the analysis or by separating clusters physically.

3. Explain partial saturation designs

Discuss how partial saturation (e.g., treating only a fraction of users within a cluster) can create a gradient of exposure, allowing estimation of spillover effects and reducing contamination.

4. Compare trade-offs

Analyze the trade-offs: buffers reduce effective sample size and power, while partial saturation may introduce dilution of treatment effect; both require careful design to balance bias and precision.

5. Recommend a combined approach

Suggest that in practice, combining buffers with partial saturation or using network-based cluster detection can mitigate edge effects while maintaining sufficient power.

Key Points to Mention

  • Spillover and contamination in cluster-randomized experiments
  • Holdout buffers: geographic or network-based separation to prevent interference
  • Partial saturation: treating a subset within clusters to measure spillover
  • Trade-offs: statistical power vs. bias reduction
  • Network analysis to identify edge users and cluster boundaries
  • Meta's specific context: social networks, marketplace, or content experiments

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

Q4

What primary metric and two guardrail metrics would you choose for this feature? Critically assess 'time spent per user per day' as a success metric, including its pros, cons, and manipulation risks, and propose at least one alternative primary metric.

Product Analytics & MetricsA/B Testing & ExperimentationProduct Sense & Ideation
Author's notes

Time spent is such a seductive metric and I've seen it get gamed badly.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the feature's goal and the company's north star, then propose a primary metric that directly measures user value delivered, along with two guardrail metrics that protect against negative side effects. Critically evaluate 'time spent per user per day' by discussing its alignment with goals, potential for manipulation, and propose a better alternative like 'meaningful interactions per user' or 'retention rate'.

Pro tip: Always tie metrics to the feature's specific objective and the company's overall mission; avoid generic metrics without context. Show awareness that time spent can be a vanity metric if it doesn't reflect genuine user value.

1. Clarify feature and goals

Ask clarifying questions to understand what the feature does, its intended user impact, and how it aligns with Meta's mission and business objectives.

2. Propose primary metric

Select a primary metric that directly measures the feature's success in delivering value to users, such as meaningful interactions or retention.

3. Propose guardrail metrics

Choose two guardrail metrics that monitor potential negative consequences, such as user well-being, content quality, or system performance.

4. Critically assess 'time spent per user per day'

Discuss pros (e.g., easy to measure, correlates with engagement), cons (e.g., may not indicate value, can incentivize addictive behavior), and manipulation risks (e.g., autoplay, clickbait).

5. Propose alternative primary metric

Suggest at least one alternative primary metric that better captures user value, such as 'meaningful social interactions' or 'daily active users completing core action'.

Key Points to Mention

  • Alignment with Meta's mission and business goals (e.g., meaningful connections, revenue).
  • Definition of 'meaningful interaction' and how to measure it.
  • Guardrail metrics: user well-being (e.g., sentiment surveys), content quality (e.g., report rate), or system health (e.g., latency).
  • Pros of time spent: easy to measure, correlates with engagement, familiar metric.
  • Cons of time spent: can be gamed, may not reflect value, can promote addictive design.
  • Manipulation risks: autoplay, infinite scroll, clickbait, notifications.
  • Alternative primary metric: retention rate, daily active users, or meaningful interactions per user.

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

Q5

Compute the design effect for cluster randomization given an average cluster size of 20 and an ICC of 0.05. How does the design effect change if the average cluster size doubles but the ICC halves?

A/B Testing & ExperimentationProduct Analytics & Metrics
Author's notes

Design effect formula is 1 + (m-1)*ICC.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by stating the design effect formula for cluster randomization: DE = 1 + (m - 1) * ICC, where m is the average cluster size. Then plug in the given values to compute DE for the first scenario, and for the second scenario recalculate with the new m and ICC, comparing the results to explain how DE changes.

Pro tip: Emphasize that the design effect quantifies the variance inflation due to clustering, and that even if ICC halves, doubling cluster size can still increase DE because the (m-1) term grows. This shows you understand the trade-off between cluster size and intra-cluster correlation.

1. State the design effect formula

Recall and write the formula for design effect in cluster randomized trials: DE = 1 + (m - 1) * ICC, where m is the average cluster size and ICC is the intraclass correlation coefficient.

2. Compute DE for initial scenario

Plug in m = 20 and ICC = 0.05 into the formula: DE = 1 + (20 - 1) * 0.05 = 1 + 19 * 0.05 = 1.95.

3. Compute DE for modified scenario

For doubled cluster size (m = 40) and halved ICC (ICC = 0.025), compute DE = 1 + (40 - 1) * 0.025 = 1 + 39 * 0.025 = 1.975.

4. Compare and interpret

Compare the two design effects: 1.95 vs. 1.975. The design effect slightly increases, meaning that despite the ICC halving, the larger cluster size leads to a net increase in variance inflation. Discuss implications for sample size and power.

Key Points to Mention

  • Definition of design effect: variance inflation factor due to clustering.
  • Formula: DE = 1 + (m - 1) * ICC.
  • Calculation for m=20, ICC=0.05 yields DE=1.95.
  • Calculation for m=40, ICC=0.025 yields DE=1.975.
  • Interpretation: DE increases slightly, showing that cluster size has a stronger effect than ICC reduction in this case.
  • Implication for experimental design: need to account for design effect when computing sample size to maintain power.

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

Q6

If you ignored network effects and randomized by individual user instead of cluster, under what conditions would the naive treatment effect estimate be biased downward versus upward? Give intuition for positive and negative spillovers and a real-world example of each.

A/B Testing & ExperimentationProduct Analytics & Metrics
Author's notes

Positive spillovers bias the naive estimate downward because control users get some benefit from being connected to treated users, shrinking the apparent gap.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

First, clarify that the naive estimate is the difference in means between treated and control individuals, which ignores interference. Then, analyze how spillovers from treated to control units and within treatment arms bias this estimate, considering the sign and magnitude of spillovers. Finally, provide real-world examples of positive and negative spillovers to illustrate the intuition.

Pro tip: Emphasize that the direction of bias depends on the relative strength of spillovers to control versus treatment units, and that even with positive spillovers, the naive estimate can be biased downward if control units benefit more.

1. Define the naive estimator and assumptions

State that the naive estimator compares outcomes of treated and control individuals, assuming no interference (SUTVA). Explain that with spillovers, this assumption is violated.

2. Analyze bias from spillovers to control units

Consider how treated units affect control units. If spillovers are positive, control outcomes increase, reducing the observed difference (bias downward). If negative, control outcomes decrease, increasing the difference (bias upward).

3. Analyze bias from spillovers within treatment units

Consider how treated units affect each other. Positive spillovers among treated units increase treated outcomes, increasing the difference (bias upward). Negative spillovers decrease treated outcomes, reducing the difference (bias downward).

4. Combine effects to determine net bias

The net bias depends on the sum of spillover effects on control and treatment units. If positive spillovers to control outweigh those to treatment, bias is downward; if the reverse, bias is upward. Similarly for negative spillovers.

5. Provide real-world examples

Give examples: Positive spillover: vaccination (treated individuals protect controls). Negative spillover: ride-sharing (treated drivers reduce demand for control drivers).

Key Points to Mention

  • SUTVA (Stable Unit Treatment Value Assumption) and its violation under interference
  • Positive spillovers: treated units benefit control units, leading to downward bias if control benefits more than treated
  • Negative spillovers: treated units harm control units, leading to upward bias if control is harmed more than treated
  • Spillovers within treatment arm: positive spillovers increase treated outcomes, biasing upward; negative spillovers decrease treated outcomes, biasing downward
  • Net bias depends on the difference in spillover effects between treatment and control groups
  • Real-world examples: vaccination (positive), ride-sharing (negative), social media features (positive), ad auctions (negative)

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