Northeast Freight Time one clock, four readings

Averages Hide the Problem

Dwell is reported as an average and the average is close to useless, because dwell distributions are not symmetric. They have a long right tail, and the tail is where the cost, the disputes and the safety exposure sit. For another workplace-measurement reference, identifying mouse jigglers.

What a two-hour average usually contains

Two facilities can both report two hours. For complementary transport data and research, see Transport Topics.

One is genuinely slow at everything. Most loads between ninety minutes and two and a half hours. A capacity problem, and it needs more doors, more labour, or fewer appointments per hour.

The other is fast with a tail. Most loads at forty minutes and a handful at six or seven. A failure-mode problem, and it needs to know what those handful have in common — a commodity, a door, a shift, a carrier, a time of day.

The average is identical and the fixes are opposite. Anyone acting on the mean alone will spend money on the wrong thing.

The four numbers to report instead

Cheap to compute from the same data.

Median. The typical load. Usually well below the mean, and the gap between them is itself informative — a large gap means a tail.

75th percentile. What a bad-but-normal load looks like. The number a carrier plans around.

95th percentile. The tail. This is what generates detention charges and consumes hours-of-service.

And the proportion over your free-time allowance. The only figure that predicts your invoice volume, and almost nobody computes it.

Four numbers, one spreadsheet column, and they turn an unactionable average into a description of what is actually happening.

Why the tail matters more than the middle

Charges are generated by the tail. Free time exists precisely so ordinary variation is unbilled, so the mean sits below the threshold and contributes nothing to the invoice total.

Hours-of-service failures come from the tail. A driver loses a delivery to a six-hour wait, not to a two-hour one.

And the safety association is about the tail too. The federal finding relates increases in average dwell to expected crash rate, and averages move because tails lengthen.

So the operational question is never what is our average. It is what happens on our worst ten per cent of loads, and what do they have in common.

How to find the tail's cause

Take the slowest ten per cent of visits over a month and cross-tabulate. Five cuts, and one of them usually explains most of it.

By hour of arrival. Morning spikes are the commonest single pattern.

By door. One slow door drags everything and is invisible in aggregate.

By commodity or load type. Reefer and anything needing a lumper behave differently.

By day of week. Mondays and the day after a holiday.

And by carrier. Occasionally it genuinely is one carrier arriving unprepared, and knowing that is as useful as the other four.

What to be careful about

Small samples. Twenty visits will not give you a stable 95th percentile. A month is a floor for a busy dock and a quarter for a quiet one.

Truncated data. If your system starts at check-in, your distribution is missing the queue, and the tail is exactly where that omission is largest.

And comparing your distribution to published averages. Industry figures are self-reported, undefined and coarse. Compare your own months to each other, which is the only comparison where the definition is held constant.

The one-line version

Report the median, the 95th, and the share over your allowance. Ignore the mean — and think twice before putting any of it on a screen.

Any facility that reports those three has more usable information about its own dwell than an industry survey can provide, and it costs one afternoon to set up.

The short version