Descriptive Statistics for Safety Data

Why Descriptive Statistics Matter for the Safety Professional

Safety management is a data-driven discipline. Every incident report, near-miss log, audit score, exposure measurement, and training record generates data. Raw data alone, however, tells a story that nobody can read. Descriptive statistics are the tools that condense thousands of individual observations into a handful of numbers and pictures that reveal patterns, highlight problems, and support decisions.

The BCSP Certified Safety Professional (CSP) exam tests descriptive statistics under the Program Management and Risk Management domains because a CSP is expected to:
• Summarize injury and illness experience for management and regulators
• Compare performance across departments, sites, or time periods
• Identify trends and outliers that signal emerging hazards
• Interpret industrial hygiene sampling results
• Build a factual foundation for inferential statistics and hypothesis testing
• Communicate risk clearly to audiences who are not statisticians

If you cannot correctly calculate and interpret a mean, standard deviation, or incidence rate, you cannot credibly defend a safety program, justify a budget, or evaluate whether an intervention actually worked. That is why these questions appear on the exam and why they are usually answerable with careful reading and basic arithmetic.

What Descriptive Statistics Are

Descriptive statistics organize, summarize, and present data that has already been collected. They describe what the data shows without attempting to generalize beyond it. This is the key contrast with inferential statistics, which use sample data to draw conclusions, estimate parameters, or test hypotheses about a larger population.

Simple test: if a question asks you to calculate, summarize, chart, or describe a data set, it is descriptive. If it asks you to predict, estimate a population value, test a hypothesis, or state a confidence level, it is inferential.

Descriptive statistics fall into four families:
1. Measures of central tendency – where the center of the data lies (mean, median, mode)
2. Measures of dispersion (variability) – how spread out the data is (range, variance, standard deviation, interquartile range, coefficient of variation)
3. Measures of position – where an individual value sits relative to the rest (percentiles, quartiles, z-scores)
4. Measures of shape and graphical summaries – the form of the distribution (skewness, kurtosis, normal curve) and visual tools (histograms, Pareto charts, box plots, run charts)

Levels of Measurement: The Foundation

Before choosing a statistic you must know what kind of data you have. The exam frequently tests this.

• Nominal – categories with no order (injury type, department, body part). Only the mode and frequency counts are meaningful.
• Ordinal – ordered categories with unequal or unknown spacing (risk ranking of low/medium/high, Likert-scale survey responses, OSHA severity classifications). Median and mode are appropriate; the mean is technically improper.
• Interval – equal spacing but no true zero (temperature in °F or °C). Mean, median, mode, and standard deviation are valid; ratios are not.
• Ratio – equal spacing with a true zero (hours worked, lost workdays, noise exposure in dBA converted to pressure, chemical concentrations in ppm). All statistics are valid and ratios make sense (twice as many lost days).

Remember the acronym NOIR (Nominal, Ordinal, Interval, Ratio) moving from least to most informative.

Measures of Central Tendency: How They Work

Mean (arithmetic average)
Formula: Mean = (Sum of all values) ÷ (Number of values). Population mean is symbolized μ (mu); sample mean is x̄ (x-bar).
Strength: uses every data point. Weakness: heavily influenced by outliers.
Example: Lost workdays for six incidents: 2, 3, 3, 5, 7, 40. Mean = 60 ÷ 6 = 10 days. Notice one severe case drags the average far above what most incidents look like.

Median
The middle value when data are arranged in order. With an even count, average the two middle values.
Same example sorted: 2, 3, 3, 5, 7, 40. Middle two are 3 and 5, so median = 4 days.
Strength: resistant to outliers; preferred for skewed data such as lost-time days, claim costs, or exposure concentrations.

Mode
The most frequently occurring value. The example has a mode of 3. A data set can be bimodal or multimodal, or have no mode. It is the only measure available for nominal data (most common injury type = strains and sprains, for instance).

Relationship in skewed data:
• Symmetric (normal) distribution: mean = median = mode
• Positively (right) skewed, with a long tail to the right: mean > median > mode. Typical of injury costs and lost days where most are small and a few are catastrophic.
• Negatively (left) skewed: mean < median < mode
Memory aid: the mean is pulled toward the tail.

Measures of Dispersion: How They Work

Two departments can have the same average injury rate but very different consistency. Dispersion tells you about predictability and control.

Range = Maximum − Minimum. Quick but uses only two values. Example above: 40 − 2 = 38 days.

Variance = average of the squared deviations from the mean.
• Population variance σ² = Σ(x − μ)² ÷ N
• Sample variance s² = Σ(x − x̄)² ÷ (n − 1)
The (n − 1) is called degrees of freedom and corrects for the tendency of samples to underestimate population variability. On the exam, if the problem describes a sample (almost always the case in safety work), use n − 1.

Standard deviation (SD) = square root of the variance. It is expressed in the same units as the data, which makes it interpretable. Symbols: σ (population), s (sample).

Worked example (sample): Noise readings in dBA: 85, 88, 90, 92, 95.
Step 1: Mean = 450 ÷ 5 = 90.
Step 2: Deviations: −5, −2, 0, 2, 5. Squared: 25, 4, 0, 4, 25. Sum = 58.
Step 3: Sample variance = 58 ÷ (5 − 1) = 14.5.
Step 4: Sample SD = √14.5 ≈ 3.81 dBA.
(If treated as a population: 58 ÷ 5 = 11.6, SD ≈ 3.41. Different answer, so the sample/population distinction matters.)

Interquartile range (IQR) = Q3 − Q1, the spread of the middle 50% of data. Resistant to outliers and the basis of the box-and-whisker plot. A common outlier rule flags values below Q1 − 1.5(IQR) or above Q3 + 1.5(IQR).

Coefficient of variation (CV) = (SD ÷ Mean) × 100%. A unit-free measure that lets you compare variability between data sets with different units or scales, such as comparing the consistency of recordable rates between a 50-person site and a 5,000-person site. Lower CV = more consistent.

Measures of Position

Percentiles indicate the percentage of observations at or below a value. The 90th percentile exposure means 90% of samples were at or below that level. Industrial hygiene often uses the 95th percentile to judge compliance.

Quartiles divide data into four equal parts: Q1 (25th percentile), Q2 (median, 50th), Q3 (75th).

Z-score = (x − μ) ÷ σ. It tells how many standard deviations a value lies from the mean. A z of +2.0 means the observation is two SDs above average, a statistically unusual result in a normal distribution. Example: if the mean monthly first-aid cases is 12 with SD 3, a month with 21 cases has z = (21 − 12) ÷ 3 = 3.0, a strong signal to investigate.

Distribution Shape and the Normal Curve

The normal distribution is a symmetric bell-shaped curve defined entirely by its mean and standard deviation. Many measurement errors and some exposure data approximate it, and many inferential tests assume it.

The empirical rule (68-95-99.7):
• About 68% of values fall within ±1 SD of the mean
• About 95% fall within ±2 SD
• About 99.7% fall within ±3 SD
Example: If TWA exposures average 40 ppm with SD 5 ppm and are normally distributed, roughly 95% of exposures fall between 30 and 50 ppm, and only about 2.5% exceed 50 ppm.

Skewness describes asymmetry. Safety data such as claim costs, days away, and chemical concentrations are typically right-skewed and often better described by a lognormal distribution, which is why IH statistics use the geometric mean and geometric standard deviation.

Kurtosis describes the heaviness of the tails (peakedness). High kurtosis means more extreme values than a normal curve would predict.

Graphical Descriptive Tools

• Frequency distribution / histogram – shows how many observations fall in each class interval; reveals shape, center, spread, and outliers.
• Pareto chart – a bar chart sorted from most to least frequent with a cumulative percentage line. Based on the 80/20 principle: roughly 80% of injuries come from 20% of causes. Used to prioritize corrective action.
• Box-and-whisker plot – displays median, Q1, Q3, and outliers; excellent for comparing departments side by side.
• Run chart / trend chart – data plotted over time to spot trends, cycles, or shifts; the precursor to statistical process control charts.
• Pie chart – proportions of a nominal variable (percent of injuries by body part).
• Scatter plot – shows relationship between two variables (overtime hours vs. incident count); descriptive but leads toward correlation.

Safety Rates as Descriptive Statistics

The exam expects you to calculate standardized rates, which are descriptive summaries that allow fair comparison across organizations of different sizes.

OSHA incidence rate = (Number of cases × 200,000) ÷ Total hours worked
The constant 200,000 represents 100 full-time employees working 40 hours a week for 50 weeks. Variations include:
• TRIR (Total Recordable Incident Rate) – all recordable cases
• DART rate – cases with Days Away, Restricted, or Transferred
• LTIR – lost-time cases only
• Severity rate = (Lost workdays × 200,000) ÷ Hours worked

Example: A plant with 450 employees worked 912,000 hours and recorded 14 recordable injuries, 6 of which involved days away or restriction.
TRIR = (14 × 200,000) ÷ 912,000 = 3.07
DART = (6 × 200,000) ÷ 912,000 = 1.32

Some international and insurance systems use 1,000,000 hours as the constant (frequency rate). Read the question to see which base is specified.

Also know experience modification rate (EMR) conceptually: a ratio of actual losses to expected losses; 1.0 is average, below 1.0 is better than average.

Interpreting Results Correctly

Descriptive statistics describe but do not explain. A rising TRIR could mean more hazards, better reporting, more hours of risky work, or random variation in a small population. A CSP must:
• Check sample size: rates from small groups swing wildly and should be viewed cautiously
• Consider the denominator: more hours or more employees changes rates even if raw counts are stable
• Look at dispersion as well as the center
• Identify outliers before deciding whether to include them
• Avoid confusing correlation with causation
• Present both the number and the context when reporting to management

How to Answer Exam Questions on Descriptive Statistics

Most CSP questions on this topic fall into four patterns:

Pattern 1 – Direct calculation. You are given a small data set and asked for the mean, median, mode, range, variance, SD, or a rate. Approach: write the data in order, label n, do the arithmetic in steps, and check that the answer is reasonable (the mean must lie between the minimum and maximum; the SD cannot be negative).

Pattern 2 – Choose the appropriate statistic. You are told about the data type or shape and asked which measure is best. Approach: nominal data → mode; ordinal data → median; skewed data or outliers → median and IQR; symmetric interval/ratio data → mean and SD; comparing variability across different scales → coefficient of variation.

Pattern 3 – Interpretation. You are given a mean and SD and asked what percentage lies in a range, or given a z-score and asked what it means. Approach: apply the empirical rule or z-score formula, draw a quick bell curve, and shade the area.

Pattern 4 – Concept recognition. Definitions such as descriptive vs. inferential, population vs. sample, parameter vs. statistic, or which chart serves which purpose. Approach: eliminate distractors that describe prediction, estimation, or hypothesis testing when the question is about description.

Exam Tips: Answering Questions on Descriptive Statistics for Safety Data

1. Sort the data first. Nearly every error with medians and quartiles comes from skipping this step. Rewrite the numbers in ascending order before doing anything else.

2. Count n carefully. Whether the median is a single middle value or the average of two depends on whether n is odd or even. Miscounting by one changes the answer.

3. Sample means n − 1. When a question says sample, use n − 1 in the variance denominator. If it explicitly says population or all incidents in the entire company, use N. If the answer choices include both versions, this distinction is the whole question.

4. Do not take the square root too early or forget it entirely. Variance is in squared units; standard deviation is the square root. Answer choices will often include the variance as a distractor when the SD is requested, and vice versa.

5. Memorize the 200,000 constant and the rate formula. Also be ready for a question that gives you the rate and hours and asks you to back-calculate the number of cases: Cases = (Rate × Hours) ÷ 200,000.

6. Watch for the outlier trap. When one value is far from the others, the exam is signaling that the mean is distorted and the median is the better descriptor. Questions asking for the most representative or typical value with an obvious outlier want the median.

7. Know the skew direction rule. The mean is pulled toward the tail. If mean > median, the distribution is right (positively) skewed. Injury severity and cost data are almost always right-skewed.

8. Lock in 68-95-99.7. Be able to apply it in one direction too: about 16% lies above +1 SD, about 2.5% above +2 SD, about 0.15% above +3 SD.

9. Match the data level to the statistic. If a question describes a 1-to-5 hazard severity ranking and asks for the best measure of central tendency, the answer is the median, not the mean, because the data are ordinal.

10. Match the chart to the purpose. Prioritizing causes → Pareto. Showing shape of a distribution → histogram. Trend over time → run chart. Comparing spread between groups → box plot. Proportions of a whole → pie chart.

11. Check units and reasonableness. A standard deviation larger than the range, a mean outside the data span, or a rate in the hundreds for a normal workplace signals an arithmetic slip. Recalculate.

12. Separate descriptive from inferential wording. Words like summarize, describe, average, spread, and chart are descriptive. Words like estimate, predict, infer, significant, confidence interval, and hypothesis are inferential. Many questions can be answered by this distinction alone.

13. Use the calculator deliberately. On the computer-based exam, write intermediate results (sum, mean, each deviation squared) rather than chaining long keystroke sequences. Transcription errors are more common than formula errors.

14. Read what is being asked twice. A question may present six incidents with lost days and ask for the rate, the severity, the median, or the range. The data set is the same; the answer is entirely different. Underline the key noun before computing.

15. Manage time. Statistics questions are typically straightforward but computation-heavy. If a calculation is going long, flag it, move on, and return. Never leave it blank; use estimation and elimination to pick the most reasonable choice.

Quick Reference Formulas

• Mean = Σx ÷ n
• Median = middle value of sorted data (average of two middle values if n is even)
• Mode = most frequent value
• Range = Max − Min
• Sample variance s² = Σ(x − x̄)² ÷ (n − 1)
• Population variance σ² = Σ(x − μ)² ÷ N
• Standard deviation = √variance
• IQR = Q3 − Q1
• CV = (s ÷ x̄) × 100%
• z = (x − μ) ÷ σ
• Incidence rate = (Cases × 200,000) ÷ Hours worked
• Severity rate = (Lost workdays × 200,000) ÷ Hours worked
• Empirical rule: 68% within ±1σ, 95% within ±2σ, 99.7% within ±3σ

Summary

Descriptive statistics turn safety data into safety information. For the CSP exam, master the vocabulary (levels of measurement, central tendency, dispersion, position, shape), practice the handful of formulas until they are automatic, and learn to recognize which statistic fits which situation. Pay special attention to the sample-versus-population distinction, the effect of outliers and skew on the mean versus the median, the empirical rule, and the standardized OSHA rate calculation. With sorted data, careful counting, and a reasonableness check on every answer, these questions become some of the most reliable points you will earn on the exam, and the same skills will make you a more persuasive and effective safety professional in practice.