How to assess · For hiring teams

How to Assess Pandas Skills When Hiring

The test formats that actually work for Pandas, what a strong answer looks like, sample questions and a scoring rubric you can use as-is.

The short answer

Assess Pandas with a task, not a conversation: clean and analyse a messy file, find the bugs, ai-scored assessment (e.g. cohesyve) or performance rewrite. Score it against written criteria you fix before you see any submissions, and weight the criteria that the role actually depends on.

  • Checks shapes and counts after every merge, filter and groupby
  • Understands index alignment and the difference between views and copies
  • Handles missing data explicitly and knows how each operation treats it
  • Uses vectorised operations and knows when apply is the wrong tool

Paste a job description; Cohesyve generates a role-specific assessment and rubric. Ten candidates free, no card.

Pandas is where most Python data work happens and where most of it goes subtly wrong: a chained assignment that did not assign, a merge that silently multiplied rows, a groupby that dropped the rows with missing keys, an apply that took forty minutes because the vectorised version was not obvious. The library is easy to use and hard to use correctly. This page covers how to assess pandas for data analyst, scientist and engineering roles: data manipulation correctness, performance, handling of missing and messy data, and the habits that make analysis trustworthy.

Why Pandas is worth testing

Pandas errors do not throw. A merge that fans out, a filter that missed nulls, an index that misaligned during arithmetic — all produce a number, and the number is wrong. Testing with realistic messy data exposes whether a candidate checks shapes, understands alignment and validates results, and that separates analysts whose numbers can be trusted from those whose cannot.

What strong Pandas looks like

  • Checks shapes and counts after every merge, filter and groupby
  • Understands index alignment and the difference between views and copies
  • Handles missing data explicitly and knows how each operation treats it
  • Uses vectorised operations and knows when apply is the wrong tool
  • Reshapes with melt, pivot and stack correctly
  • Writes analyses as readable pipelines, not accumulated notebook cells
  • Validates results against a known total before reporting

Ways to assess Pandas

Clean and analyse a messy file

Provide a CSV with mixed date formats, duplicate rows, a key with inconsistent casing and a few impossible values, plus a business question. Ask for the cleaned data, the answer, and the checks along the way. Sixty minutes.

Pros

The daily job; exposes validation habits.

Cons

Needs a crafted dataset.

Best for Any level.

Find the bugs

Provide a notebook with a fan-out merge, a chained-assignment that did nothing, and a groupby that dropped null keys. Ask what is wrong and what the correct numbers are.

Pros

The real silent errors; scoreable.

Cons

Needs a crafted notebook.

Best for Mid and senior roles.

AI-scored assessment (e.g. Cohesyve)

Generate a pandas task from the job description — a cleaning task, a bug hunt, a performance question — with a rubric. Each candidate receives a different variant; the reasoning is scored alongside the work.

Pros

Asynchronous and consistent across a large pool; a different task per candidate removes shared answers; scores the explanation, which is where judgement shows.

Cons

Cannot run tools on the candidate's behalf; keep a human review for shortlisted finalists.

Best for Screening an applicant pool fairly before interview time is spent.

Performance rewrite

Give a slow row-wise apply and ask for a vectorised version.

Pros

Tests idiomatic use.

Cons

Narrow.

Best for Roles with large data.

Cohesyve

Run a Pandas assessment on your next opening

Cohesyve generates a unique Pandas task per candidate from your job description, with the scoring rubric attached. Questions are different for every applicant, so they cannot be shared or looked up.

What to test

Correctness

Whether operations do what was intended.

Merge two tables and verify row countsFix a chained assignmentExplain why a groupby result is missing rows

Missing and messy data

Whether edge cases are handled.

Parse mixed date formatsDecide how to treat nulls in a meanDeduplicate with a defensible rule

Reshaping

Whether data can be shaped for the question.

Pivot long to wideMelt a wide tableCompute a rolling metric per group

Performance and structure

Whether code is efficient and readable.

Vectorise a slow applyRefactor a notebook into a pipelineExplain when to leave pandas for a database

Sample Pandas questions

After a merge, your row count doubled. What happened?

Entry

Look for Duplicate keys on one side; check with value counts; decide on aggregation or dedup.

What is the difference between a view and a copy, and why does it matter?

Entry

Look for Chained assignment may not modify the original; use loc; the warning is a real problem.

A groupby mean is missing a category. Why?

Mid

Look for Null keys are dropped by default; dropna parameter; check counts.

This apply takes forty minutes on a million rows. Fix it.

Mid

Look for Vectorised operations, numpy, or groupby transforms; measures the result.

How do you make sure a reported total is right?

Senior

Look for Reconcile against a known value, check shapes at each step, assert invariants, keep a validation cell.

Red flags

  • Never checks shapes after a merge
  • Ignores the chained-assignment warning
  • Row-wise apply for everything
  • Silent null handling
  • Notebook state that cannot be rerun

Scoring rubric

CriterionWeightWhat strong looks like
Correctness and validation35%Checks counts and invariants; catches silent errors.
Messy-data handling25%Nulls, duplicates and formats are handled explicitly.
Reshaping fluency20%Shapes data for the question without hacks.
Performance and structure20%Vectorised, readable, rerunnable.

Mistakes hiring teams make

  • Testing API recall instead of validation habits
  • Using a clean dataset
  • Not planting a fan-out merge — the most common silent error
  • Ignoring performance
  • Accepting a correct answer reached without any checks

Roles that need Pandas

Data AnalystData ScientistData EngineerAnalytics EngineerMachine Learning EngineerQuantitative Analyst

Common questions

Is pandas skill the same as Python skill?

No. Pandas has its own model — alignment, views, null semantics — and strong Python developers make pandas errors. Test it with data.

What is the best single pandas question?

Describe a merge that doubled the row count and ask why. Key understanding and validation habit show together.

Should I test polars or other alternatives?

If the role uses them. The validation habits transfer; the API does not matter much.

How long should a pandas assessment take?

Sixty minutes for a clean-and-analyse task; thirty for a bug hunt.

Cohesyve · Skill assessments for hiring

Test Pandas before the first interview

Generate a role-specific Pandas assessment from your job description and see who can do the work before you spend interview time on them.

1,500+

assessments completed

50%

faster time-to-hire

90%

completion rate

5 min

from JD to assessment

No credit card · 10 free candidates · Plans sized to your hiring volume

See Cohesyve in action

Free 30-min walkthrough

See it on your role