Employee salaries at a fictional company

What it is good for

• Describe a right-skewed salary distribution with percentiles and quartiles.
• Compare salaries between groups and see how a third variable (job level) changes the comparison.
• Fit a regression of log salary on level, experience and department and interpret coefficients as approximate percentages.
• Recognise zero-inflated data (bonus) and summarise it in two parts.

Methods: percentiles, group comparison, two-sample t-test, multiple regression, log transformation, stratification, adjustment and mediation.

Data quality notes

• Women earn about 15% less on average overall, but comparing within job levels the gap averages only about 1% (individual levels range from about −2% to +3%): most of the raw gap goes with fewer women at senior levels. Job level cannot cause gender, so it is not a confounder here; it is a possible mediator — if promotion itself is unequal, adjusting for level hides that part of the disparity. Neither the raw nor the adjusted gap proves or disproves discrimination. The dataset was built this way to teach the difference; it says nothing about any real employer.
• annual_bonus_sar is zero for about 28% of employees; its distribution is a spike at zero plus a right-skewed part.
• performance_rating and job_level are ordinal; performance_rating is stored as integers 1–5.
• No missing values.

Provenance and licence

Gender (42% female), department, education and experience (gamma) are drawn first. Job level comes from a latent promotion score that rises with experience and education and is lower for women (a simulated promotion barrier), cut at fixed thresholds. Base salary = level base pay × department factor × experience factor × log-normal noise (sd 0.08), with a 3% lower base for women of equal level, department and experience (partly offset because women who reach a level have more experience on average, so simple within-level gaps are only about 1%); rounded to 10 SAR. Performance ratings (1–5) are independent of gender. Bonus is zero-inflated: the chance of any bonus and its size in months of base pay increase with the rating.

Synthetic data · CC0-1.0 (synthetic data generated by Statistics Quest) · generated deterministically (seed 20241002). It does not describe real people or organisations; any resemblance is coincidental.