Michael Karl McNeil

Last Updated: October 29, 2026

Table of Contents

• Why Safety Data Collection Fails Without a System
• What to Collect: The Core Safety Data Set
  • Incident, Inspection, and Observation Records
  • Exposure and Severity Fields That Make Trends Visible
• Building a Hazard Reporting Form That Produces Usable Data
• Near-Miss Reporting: The Early Warning Layer
• Data Quality Controls That Keep Trends Honest
  • Standardize the Schema First
  • Handle Missing Reports and Missing Fields
  • Control Classification Drift
  • Deduplicate Records
  • Correct for Underreporting and Reporting-Behavior Differences
  • The Monthly Audit
• How to Run a Trend Analysis on Hazard Data
  • Normalize Before You Compare: Rates, Not Raw Counts
  • Time-Series Comparisons: Prior Period and Same Period Last Year
  • Pareto Analysis: Find the Few Categories Driving Most Risk
  • Severity-Weighted Trends: Stop Rare Severe Events From Disappearing
  • Worked Example: From Raw Reports to a Prioritized Trend
• Safety Analytics and Leading Safety Indicators in Practice
• Turning Hazard Trends into Corrective Actions
• Frequently Asked Questions

Why Safety Data Collection Fails Without a System

Most safety programs collect plenty of paper and very little evidence. Incident reports get filed, inspection sheets get signed, and none of it adds up to a pattern anyone can act on.

The fix is not more paperwork. It is a system with defined fields, consistent definitions, and a review rhythm.

Three things break down most often:

• Definitions drift. Two supervisors classify the same event differently, so the data cannot be compared.
• Nothing is coded. Free-text narratives cannot be counted, sorted, or trended.
• No one owns review. Data gets collected and never read.

Below is a workflow, from fields to record to the corrective action that closes the loop.

What to Collect: The Core Safety Data Set

A usable safety data set rests on four record types: incidents, inspections, observations, and near misses. For guidance on which hazards are typically missed, see Identifying Hidden Workplace Safety Hazards. Each answers a different question, and together they show what happened and what nearly happened.

Incident data covers injuries, illnesses, and property damage. Inspection records capture conditions found during scheduled walkthroughs. Safety observations record behavior and conditions noted in passing. Near-miss reporting captures events with no injury but real potential.

For each record, capture a minimum field set:

• Date, time, and location
• Activity underway at the time
• Hazard category and contributing factors

A record without a location or activity cannot be trended.

Incident, Inspection, and Observation Records

Each record type needs its own form, but all should share the same hazard categories, which let you compare an inspection finding against an incident six months later.

Keep the category list short. Ten to fifteen hazard types cover most manufacturing and construction work. Long lists get abandoned in the field.

Exposure and Severity Fields That Make Trends Visible

Frequency alone tells you how often something happens, not how bad it could have been. Add two fields to every record: an exposure estimate and a severity rating.

Severity ratings should follow a simple scale: first aid, medical treatment, lost time, or fatality potential. Exposure can be recorded as people affected times hours at risk.

Building a Hazard Reporting Form That Produces Usable Data

Design the form around structured inputs, not essays: dropdowns for hazard category, location, and activity, checkboxes for contributing factors, and one short free-text narrative field.

Rules for the form:

• Required fields only where they matter. Too many mandatory fields cause abandonment.
• Plain language labels. “What was the person doing?” beats “Activity classification.”

The reporting form is part of your written safety program, so it has to plug into the larger structure.

Near-Miss Reporting: The Early Warning Layer

Near-miss reporting captures events that could have caused harm but did not.

The problem is participation. Workers may stay quiet when they fear blame or believe nothing will change. Fixing that is a culture task, not a form task. The practices in Developing an Effective Employee Safety Program drive the participation near-miss reporting depends on.

Practical steps for near-miss reporting:

• Separate near-miss reports from discipline. State it in writing.

Data Quality Controls That Keep Trends Honest

Bad data produces confident wrong answers. Before any trend analysis, put controls in place that catch the predictable errors.

Standardize the Schema First

Most data-quality problems are created at the form, not fixed in the spreadsheet. Lock a shared hazard data schema so every record carries the same fields and allowed values. A workable schema:

Field	Type	Allowed values / rule
Date and time	Date/time	Required
Location	Dropdown	Predefined list of areas or assets
Activity	Dropdown	Predefined task list
Hazard category	Dropdown	10-15 fixed categories
Contributing factors	Multi-select	Predefined list
People exposed	Number	Required
Exposure duration	Number (hours)	Required
Severity rating	Dropdown	First aid / medical / lost time / fatality potential
Immediate action	Short text	Required
Corrective action status	Dropdown	Open / in progress / verified closed
Reporter	Optional	Used for follow-up only

Shared dropdowns are the single biggest defense against classification drift. If two supervisors cannot pick different words for the same event, the data stays comparable across teams and time.

Handle Missing Reports and Missing Fields

Missing fields break analysis; missing reports hide risk. Run a completeness check before every review cycle: count records with each required field blank and report the percentage by source.

For missing reports, compare report volume by shift, department, and supervisor. A department with zero reports is a red flag, not a clean record. Investigate before you celebrate.

Control Classification Drift

Misclassification distorts categories. Monthly, audit 10 records against the written definitions and have a second reviewer re-classify the same sample blind. Where they disagree, the definition is usually the problem, tighten it and retrain.

Deduplicate Records

Duplicate records inflate counts. Check for repeat entries from the same event using date, location, and a short event fingerprint (activity plus hazard category). Merge duplicates before analysis and keep an audit note.

Correct for Underreporting and Reporting-Behavior Differences

Under-reporting bias hides risk, and reporting behavior varies by supervisor and shift. Two controls help:

• Track reports per employee by department and shift. A persistent gap between similar departments usually reflects reporting culture, not safety performance.
• Compare near-miss volume to incident volume. A department with incidents but almost no near misses is likely underreporting near misses, not running clean.

The Monthly Audit

A simple monthly audit catches most of this:

Start a website →

Control	What to Check	Frequency
Duplicates	Same date, location, and event fingerprint	Monthly
Classification	Blind re-classification of 10 records vs. definitions	Monthly
Completeness	Required fields populated, by source	Monthly
Reporting bias	Volume by shift and department	Quarterly
Narrative quality	Enough detail to reconstruct the event	Quarterly

These controls matter when records face outside scrutiny.

How to Run a Trend Analysis on Hazard Data

Trend analysis compares hazard records over time by category, location, or activity to find repeating patterns. Run it on a fixed schedule with a fixed method, and normalize the data before comparing it.

A workable monthly routine:

1. Export all records for the review period.
2. Normalize counts into rates using a consistent denominator (hours worked, employees, or inspections completed).
3. Group by hazard category, location, and activity.
4. Run a Pareto sort to find the few categories driving most of the risk.
5. Build a severity-weighted trend so rare but severe events are not buried by high-frequency minor ones.
6. Compare each group against the prior three periods and the same period last year.
7. Flag any group rising in rate, severity weight, or both.
8. Assign an owner to each flagged group.

Pattern recognition beats statistics here: look for repeats, clusters, and slow climbs, not p-values.

Normalize Before You Compare: Rates, Not Raw Counts

Raw counts mislead because exposure changes. A department that added a shift will show more incidents without being more dangerous. Convert every count to a rate before trending.

Common rate forms:

• Incident rate per 200,000 hours worked. This is the standard OSHA incidence-rate basis (200,000 hours equals 100 full-time workers at 40 hours for 50 weeks). It lets you compare departments of different sizes.
• Near-miss reports per 100 employees per month. Useful for tracking reporting culture, not just harm.
• Findings per inspection completed. Normalizes inspection data so a month with more walkthroughs does not look worse.

Pick one denominator per metric and keep it fixed. Changing the denominator mid-year makes every trend line meaningless.

Time-Series Comparisons: Prior Period and Same Period Last Year

A single month tells you nothing. Compare each period against two baselines: the prior three periods (to catch short-term climbs) and the same period last year (to remove seasonality).

A simple control-chart view helps. Plot the monthly rate and a rolling three-month average. When the rolling average crosses above the prior year’s band for two consecutive periods, that is your flag.

Pareto Analysis: Find the Few Categories Driving Most Risk

Pareto analysis sorts hazard categories by frequency or severity weight to show where a few categories account for most of the problem. Sort descending, calculate cumulative percentage, and find where the curve flattens.

Severity-Weighted Trends: Stop Rare Severe Events From Disappearing

Frequency-only trends hide low-count, high-consequence hazards. Weight each record by its severity rating before summing, so one fatality-potential event counts more than several first-aid cases.

A simple weighting scheme:

Severity rating	Weight
First aid	1
Medical treatment	3
Lost time	5
Fatality potential	10

Multiply each record’s weight by its count and sum by category. The weighted total is your risk-priority number. Review it alongside the raw frequency so you can see both the common problem and the dangerous one.

Worked Example: From Raw Reports to a Prioritized Trend

Take a fabrication shop with three hazard categories over one quarter, 24,000 hours worked.

• Struck-by: 6 incidents, severity ratings of 2, 2, 3, 1, 2, 4. Total severity 14. Weighted total: (5 × 1) + (1 × 3) = 8.
• Slip and fall: 3 incidents, ratings of 3, 3, 4. Total severity 10. Weighted total: (2 × 3) + (1 × 5) = 11.
• Chemical exposure: 1 incident, rating of 4. Total severity 4. Weighted total: 1 × 10 = 10.

Frequency says struck-by is the top problem. Severity average tells a different story: chemical exposure averages 4.0 per event, slip and fall 3.3, struck-by 2.3.

That is the judgment call analytics cannot make for you. The numbers frame it; you decide.

Safety Analytics and Leading Safety Indicators in Practice

Safety analytics means using collected hazard data to examine patterns, not just count what already happened. The key distinction is leading versus lagging indicators.

Lagging indicators measure outcomes after the fact: injury rates, lost-time days, workers’ compensation claims.

Leading indicators measure conditions that may precede harm: near-miss volume, inspection findings closed on time, training completion, hazard reports per employee, corrective action closure rate.

A practical leading indicator set:

• Near-miss reports per 100 employees per month
• Percentage of inspection findings closed within 30 days
• Hazard reports submitted per department

Turning Hazard Trends into Corrective Actions

Run it as a closed loop:

1. Assign the trend to a named owner.
2. Diagnose the root cause, not just the symptom.
3. Fix the condition, the process, or the training gap.
4. Set a due date and track it.
5. Verify the fix held by watching the trend for the next two periods.
6. Close the action only after verification.

A guard installed is not a fix confirmed.

When a trend points to a gap your team cannot close internally, outside help is an option.

---

Michael Karl McNeil built a career in fire prevention and industrial hygiene. McNeil Safety Consulting provides written safety programs, industrial hygiene monitoring, and Cal/OSHA citation defense for employers across California and the Western states.

Frequently Asked Questions

What safety data should employers collect to identify hazard trends?

Collect incident reports, near-miss reports, inspection findings, safety observations, exposure measurements, and corrective action records. Each record needs consistent fields: date, location, task being performed, hazard category, severity, and root cause. Without those fields, safety data stays anecdotal and trend analysis becomes guesswork. A standardized set also lets you compare months and sites.

How can near-miss reporting help reveal workplace safety trends?

Near-miss reporting captures events that did not cause injury. Reviewing reports may help identify recurring patterns. Encourage reporting without blame and keep the form short.

What should be included on a workplace hazard reporting form?

A hazard reporting form should capture the date and time, exact location, task being performed, equipment involved, hazard category, a description of what happened or could happen, severity and likelihood ratings, and the reporter’s name. Add a field for immediate action taken and space for photos. Keep the form concise.

How often should safety data be reviewed for trends?

Employers can review near-miss and hazard reports, inspection findings, and injury metrics on a regular schedule. The review cadence can be adjusted to incident volume.