Delivery times before and after a routing change (paired)

Average delivery time per route for 60 courier routes of a fictional parcel company, Swiftdune Couriers, measured over four weeks before and four weeks after a new route-optimisation system was introduced. Each row is one route with both measurements, so the data are naturally paired. Routes differ a lot from each other, while the improvement is small and consistent — exactly the situation where a paired analysis is far more powerful than treating the two columns as independent samples (compare with delivery_independent).

Open in the lab Download CSV

Rows: 60
Columns: 6
One row is: one delivery route (before and after)
Version: v1

Data dictionary

Column	Type	Unit	Summary	Description
`route_id` Route	id			Route identifier.
`zone` Zone	category		5 levels: North, South, East, West, Central	Delivery zone.
`distance_km` Route distance	number	km	mean 35.26, median 29.35, SD 19.71, range 11.8–120	Length of the route.
`stops` Stops	integer	stops	mean 27.45, median 28, SD 6.091, range 14–43	Average number of drop-off stops per run.
`before_min` Before	number	minutes	mean 43.09, median 41.9, SD 8.234, range 30.5–75.9	Average delivery time per parcel in the four weeks before the change.
`after_min` After	number	minutes	mean 40.4, median 38.8, SD 8.562, range 25.5–70.7	Average delivery time per parcel in the four weeks after the change.

What it is good for

Compute differences (after − before) and run a paired t-test with a confidence interval.
Contrast the paired result with an (incorrect) two-sample t-test on the same columns.
Check the distribution of differences (normality, outliers) and use the Wilcoxon signed-rank test as an alternative.

Methods: paired t-test, Wilcoxon signed-rank test, two-sample t-test (for contrast), confidence interval for a mean difference, scatter plot of before vs after.

Data quality notes

Before and after are strongly correlated (r > 0.9) because route characteristics persist.
Each value is a four-week average, not a single delivery.
No missing values.

Provenance and licence

Each route has a baseline average time driven by its distance and number of stops (route-to-route sd ≈ 12 minutes). After = before-level − 3 minutes (true saving) + route-specific variation in the saving (sd 1.5) + measurement noise (sd 2.5 for each period). Times are averages over four weeks, rounded to 0.1 minute.

Synthetic data · CC0-1.0 (synthetic data generated by Statistics Quest) · generated deterministically (seed 20241018). It does not describe real people or organisations; any resemblance is coincidental.

Preview

Dataset

Delivery times before and after a routing change (paired)

60 rows · one row = one delivery route (before and after) · synthetic data

About this dataset Open in the lab

Route	Zone	Route distance (km)	Stops (stops)	Before (minutes)	After (minutes)
RT01	North	54.9	33	47.5	51.9
RT02	Central	19.7	32	42.6	39.9
RT03	Central	41.3	37	48.7	51.1
RT04	West	65.5	31	54.9	49.5
RT05	West	35.3	24	33.6	33.5
RT06	South	20	21	42.9	33.7
RT07	South	24.5	23	38.3	36.1
RT08	East	15.3	16	33.9	28.8
RT09	Central	14.8	26	41.3	35.7
RT10	Central	17.2	21	38	38.6
RT11	East	38.7	33	51.8	45.2
RT12	South	48.1	17	41.4	39.8

Used in

Variance and standard deviation · Describing Data
Intervals for differences in means · Estimation and Confidence Intervals
One-sample t-test · Hypothesis Testing
Paired t-test · Hypothesis Testing
Effect sizes · Hypothesis Testing
Checking assumptions · Hypothesis Testing
Reporting a test responsibly · Hypothesis Testing
Repeated measures foundations · Comparing Groups and Categorical Data
Wilcoxon signed-rank test · Comparing Groups and Categorical Data
Choosing a test · Comparing Groups and Categorical Data
Did the new routing cut delivery times? · Case