Whether the Data Is Any Good

A behaviour system reports confidently regardless of whether the device is mounted properly. These four checks take an afternoon.

Event rate by vehicle

Events per hundred miles, per vehicle, within a vehicle type.

A vehicle far above its type is a mounting or calibration problem before it is anything about the driver.

Check it before anyone sees a ranking, because a miscalibrated device produces a driver who looks reckless and is not.

Re-run after any device replacement, which resets the calibration.

Driver and vehicle separated

Swap two drivers' vehicles for a week and see whether the event rates follow the driver or the vehicle.

If they follow the vehicle, the data is describing equipment.

This is the single most informative test in the subject and almost nobody runs it.

Speeding events against real limits

Take fifty events and check the actual limit at that location.

Map databases are wrong on recent changes, temporary limits and minor roads.

An error rate above a few percent means speeding data cannot be used for anything individual until the map is corrected or those roads excluded.

Clip trigger accuracy

Review a sample and classify: genuine event, defensive manoeuvre, road surface, false trigger.

The proportion in the last two categories is your false positive rate.

Above a quarter, the reviewer will start dismissing without watching, which is the failure mode that hollows out a video programme.

Reporting them

On the same page as any score or ranking.

With the vehicle type and route type stated, because both dominate.

And with the exposure visible, so a small denominator is apparent.

Acting on them

High event rate on one vehicle: check the device before the driver.

Rates following vehicles: recalibrate and re-mount, then re-baseline everything.

Speeding errors: exclude the affected roads and report the correction.

High false trigger rate: retune the thresholds, which is the month of silent logging again and is worth repeating annually.

Swap the vehicles

The most informative test in the subject.

Two drivers, two vehicles, one week, exchanged.

If the event rates follow the vehicle rather than the driver, the data is describing equipment.

Almost nobody runs it, and it is the difference between a defensible outlier conversation and an indefensible one.

Connect policy and data

The choices in this note can be compared with online timesheets for auditable work records. Keep the written purpose in control and enable only the information needed at this boundary.

Independent reference

For a thematic point of reference, see the National Safety Council. This established source is a useful independent reference for the issue discussed here.