Browse
Log in / Register
Share
Understanding data skewness is critical for HR professionals to avoid misinterpreting recruitment metrics, salary data, and performance analytics. Skewness, a measure of the asymmetry of a probability distribution, can significantly distort your analysis, leading to flawed decisions in hiring, compensation, and talent management. By learning to identify and account for it, you can base your strategies on accurate, reliable data.
In human resources, you constantly work with data sets—from applicant test scores to employee salary figures. Skewness describes how much a data set deviates from a symmetrical bell curve. In a perfectly symmetrical set, the mean (average), median (middle value), and mode (most frequent value) are all the same. Skewness occurs when outliers—extremely high or low values—pull the distribution to one side. Recognizing this is the first step in ensuring your data-driven decisions are sound.
Understanding the direction of skew helps diagnose what's happening within your data. There are two primary types:
Positive Skew (Right-Skewed) This occurs when a data set has a long tail on the right side. The mean is greater than the median, which is greater than the mode. In an HR context, this is common with salary data for a role where a few high-level executives earn significantly more than the broader team. The "average" salary appears inflated because of these outliers.
| Average Type | Position in a Positively Skewed Distribution |
|---|---|
| Mode | Highest point (peak) of the curve |
| Median | Right of the mode |
| Mean | Furthest to the right, pulled by high outliers |
Negative Skew (Left-Skewed) Here, the long tail is on the left side. The mean is less than the median, which is less than the mode. This could happen when analyzing time-to-fill a position where most roles are filled quickly, but a few hard-to-fill positions take months. The "average" time would be misleadingly low.
You don't need to be a statistician to gauge skewness. A simple rule of thumb or a basic formula can provide quick insights.
1. The Quartile Check A quick way to check for symmetry is to see if the data is evenly distributed around the median. If the distance from the first quartile (Q1) to the median (Q2) is roughly equal to the distance from the median to the third quartile (Q3), your data is likely symmetrical.
2. Pearson's Median Skewness Coefficient
For a more precise measure, you can use a standard formula:
Skewness ≈ 3 * (Mean – Median) / Standard Deviation
Ignoring skewness can lead to costly mistakes. Here’s why it’s a vital concept for evidence-based human resources.
Improve Recruitment and Selection Accuracy If you’re using pre-employment assessment scores to shortlist candidates, a negatively skewed result (where most candidates scored very high) might make a good candidate look average. By identifying the skew, you can adjust your screening process to better differentiate top talent.
Set Accurate and Fair Salary Bands Salary data is often skewed. Relying solely on the mean salary for a role can lead to setting non-competitive offers (if negatively skewed) or overspending on payroll (if positively skewed). The median is often a more reliable measure for establishing salary bands because it is less influenced by extreme outliers.
Conduct Valid Performance Appraisals When analyzing performance ratings, skewness can reveal rating biases. For instance, a manager who gives mostly high ratings with one or two very low scores creates a negatively skewed distribution. Recognizing this allows HR to address potential rating errors and ensure fairer employee evaluations.
Enhance Employer Branding Surveys Employee engagement survey data can be skewed. A positively skewed result (mostly low scores) might indicate widespread dissatisfaction, while a negatively skewed result (mostly high scores) might mask specific, critical issues. Understanding the skew helps you pinpoint the real areas for improvement.
Based on our assessment experience, the most practical advice is to always compare the mean and median. If they are significantly different, your data is skewed, and you should investigate the cause before making decisions.
