T Test
When should a supply-chain analyst use a paired t-test?
A paired t-test is appropriate when the same operational entity is measured twice, such as the same warehouse before and after a process change or the same route under two dispatch policies.
What does the p-value mean in a t-test for inventory operations?
The p-value estimates how likely the observed mean difference would be if there were actually no real process effect; a result below the chosen alpha level, often 0.05, is treated as statistically significant.
Why is the t-distribution used instead of the normal distribution?
The t-test is built for situations with limited sample sizes and unknown population variance, which is common in pilot warehouse changes, supplier trials, and line-side experiments.
A t-test is a statistical hypothesis test that determines whether the mean of one numeric sample differs from a hypothesized value or whether the means of two groups differ significantly. In supply chain and inventory analytics, it validates process changes by comparing KPIs such as lead time, putaway time, or cycle count accuracy against targets or across operational groups.
On a manufacturing shop floor, t-tests are used to decide whether a process change produced a real shift in a numeric KPI rather than random variation. Typical applications include comparing dock-to-stock time before and after slotting, picker travel time across zones, replenishment lead time between suppliers, or line-side stockout duration after a kanban change. When the same line or warehouse is measured before and after a change, the paired t-test is appropriate because the measurement units are linked over time. When comparing two distinct groups such as shifts, docks, or warehouses, an independent two-sample t-test applies. In raw-material tracking, a one-sample t-test can compare a supplier's average delivery delay against a contractual target. The test converts the observed mean difference into a t-statistic using sample means, standard deviations, and sample sizes, then compares that statistic to a p-value threshold such as 0.05. This is especially useful with small samples.
Wrong test variant: Teams compare the same warehouse before and after a process change with an independent t-test, ignoring dependency between linked measurements. This distorts significance results and can hide or invent an effect.
Ignoring variance or outliers: Rush orders, downtime days, and receiving exceptions inflate spread. If assumptions are not checked, the t-test loses validity and real shifts may be missed.
Non-comparable KPIs: A plant compares average receiving time for a small palletized SKU versus a bulk commodity as if equivalent. Different handling modes and labor content make the comparison misleading.