Final Project · Global Data & AI Workforce Compensation Analytics

Client brief: Professor McGovern
Client role: Executive-search steering committee
Core data: Global data, AI, and ML salary survey (CC0), 2020-2025, hosted as a frozen course snapshot
Due date: Friday, October 9, 2026 at 11:59 p.m. Eastern

Premise

Your team is the analytics group for an executive-search firm deciding where to build a data-and-AI recruiting practice. The steering committee needs more than salary charts. It needs a defensible target employer, a set of priority talent lanes, directional pay benchmarks, and an honest account of what the supplied archive cannot establish.

Use the compensation data to answer a client decision. Choose a large employer with a documented need for data and AI talent, connect the labor-market evidence to that employer's strategy, and recommend what the search firm should approve, pilot, reject, or investigate next.

This is not a plot collection or a claim that the Kaggle archive represents the entire labor market. It is a client-facing visual analysis built around evidence boundaries.

Core Dataset

The supplied archive contains 151,445 rows and 11 fields:

• work year;
• experience level;
• employment type;
• job title;
• reported salary and currency;
• salary converted to USD;
• employee country of residence;
• remote-work ratio;
• employer country; and
• employer size.

Read the course copy directly — no download or account needed:

salaries <- readr::read_csv(paste0(
  "https://stat.cmu.edu/~sgallagh/courses/mads-fall-2026/",
  "datavis-36613/data/salaries.csv.gz"
))

This is a frozen snapshot taken August 3, 2025. The upstream survey is republished weekly, so do not substitute a fresh download — it would change your row counts, add 2026 records, and make your audit incomparable to everyone else's.

Use the compensation dataset guide before beginning analysis.

The archive requires an explicit audit. It contains many exact duplicate rows, sharply uneven year coverage, no employer name or posting identifier, and no row-level provenance. It ends in 2025. Repeated rows therefore cannot automatically be treated as independent job openings, and record counts cannot be interpreted as employer demand.

Required Analytical Work

1. Data scope and credibility

• Explain what one supplied row appears to represent and what cannot be verified.
• Audit exact duplicates and justify whether and how they are handled.
• Show the coverage that matters for your claims: year, geography, company size, role, or remote status.
• State how the audit limits the analysis.

2. Compensation landscape

• Compare salary distributions across relevant job titles or transparent role families.
• Address skew and outliers deliberately; a log scale, quantiles, or robust summaries may be more informative than an unqualified mean.
• Explain why your comparison groups serve the client's decision.

3. Career and market signals

Address at least two of the following where the data are sufficiently supported:

• changes over time for stable job titles;
• compensation differences across experience tiers;
• company-size patterns;
• remote, hybrid, and on-site comparisons;
• employer/employee geography; or
• another clearly justified workforce question.

Descriptive differences are not causal effects. Time comparisons use nominal USD unless you explicitly add and document an inflation adjustment.

4. Recruiting playbook

• Select and justify a target employer using credible external evidence.
• Identify priority roles and experience tiers.
• Provide directional compensation benchmarks with sample sizes and scope.
• Explain which searches should be core, selective, experimental, or deferred.
• Name the next data source the client should commission or acquire.

Deliverables

1. One self-contained HTML report: produced from Quarto/R Markdown with all assets embedded by setting embed-resources: true.
2. Embedded reproducibility appendix: data links, access dates, duplicate-handling and cleaning decisions, relevant code, role-grouping rules, package/session information, and limitations.

Open the Gradescope assignment from Canvas and submit the single .html file. Do not submit a ZIP archive, separate asset folder, or separate source file.

Only these report materials are graded in 36-613, using the 36-613 Data Vis Rubric. The Professional Skills Rubric is a separate working draft for the companion-course presentation and is not yet final; it is not part of the Data Visualization grade.

Report Requirements

• Three clearly stated, decision-relevant client questions.
• Six to ten visualizations.
• At least three foundational forms from the first half of the course.
• At least three later-course or advanced forms where they genuinely help.
• No more than two one-variable plots.
• No more than three plots of the same type.
• At least one formal statistical analysis, model summary, sensitivity analysis, or uncertainty display that complements a graph.
• Visible units, subsets, denominators, sample sizes, and sources where they matter.
• A conclusion stating what the client should approve, pilot, reject, and investigate next.

Claims This Dataset Does Not Support by Itself

Do not use the supplied archive alone to claim:

• that record frequency measures employer demand or job openings;
• that a salary difference is caused by remote work, company size, geography, or title;
• that exact duplicates are certainly erroneous or certainly independent observations;
• that salary represents total compensation, purchasing power, or cost-of-living-adjusted value;
• that the data contain 2026 compensation observations; or
• that cross-border hiring or geographic labor arbitrage is well measured when the relevant pairs are sparse.

Recommended Report Shape

1. Client decision and one-sentence recommendation.
2. Data scope and credibility audit.
3. Compensation landscape.
4. Career or market signal 1.
5. Career or market signal 2.
6. Target employer and recruiting playbook.
7. Uncertainty, caveats, and rejected claims.
8. Approval request and next evidence to acquire.

Lead with the decision rather than notebook order. Cleaning detail belongs in the report when it changes the conclusion and in the reproducibility appendix otherwise.