Where does performance differ?
Compare the four terminals to locate differences in first-bag arrival time.
OPERATIONS ANALYTICS / CASE STUDY 01
Less waiting.
Better evidence.
What can two years of baggage data tell us about a better airport experience?
A quantitative study of Melbourne Airport’s four terminals, exploring where handling performance changed after automated scanners were introduced.
MEL / BAGGAGE OPERATIONS2018 → 2019
SCANNERS ADDED AT T2 + T3Run the interactive lab ↗
04Terminals compared
02Years · 2018–2019
03Research questions
01Goal · Understand efficiency
01 / THE QUESTION
The first bag is
part of the journey.
Passengers see a carousel. Analysts see a process with measurable differences.
The study examines daily average time until the first bag reaches the carousel and daily baggage volume. In the study scenario, T2 and T3 received automated scanners on 1 January 2019; T1 and T4 did not.
The challenge: separate the changes we can observe from the explanations we still need to test.
Academic case study based on a supplied research report. This is an independent portfolio project, with no claimed affiliation with Melbourne Airport.
What changed after automation?
Compare 2018 and 2019 at T2 and T3, checking time alongside workload.
What happened elsewhere?
Use T1 and T4 as context for changes without the new scanners.
02 / EXPLORE THE FINDINGS
One airport.
Four different stories.
Select a terminal. Compare the change in time with the change in workload.
2018 · Before 2019 · After
Terminal 1
Mean first-bag arrival time · source time units · lower is better
Source: report Figure 1, printed p. 5. Bars share a zero baseline across terminals.
See all values & download the chart data
| Terminal | Time 2018 | Time 2019 | Time change | Bags/day 2018 | Bags/day 2019 |
|---|
Time units are not explicitly specified in the report tables, so no conversion to seconds or minutes is made. Percentages are calculated from the displayed means. Values are rounded; see the downloadable CSV for the transcribed values.
Download summary CSV ↗03 / WHAT IS BEING IMPROVED?
Follow the bag.
Find the better question.
Make the baggage journey respond to you. Add bags, change demand, and see where a queue builds.
INTERACTIVE BAGGAGE LAB
ILLUSTRATIVE MODELChange the controls. Watch the queue react. Select a bag to follow its journey.
Select any moving bag to follow its journey.
0Waiting to scan
0Reached carousel
0Bags in system
Model assumptions: one scanner; an illustrative service rate that is faster in automated mode; user-controlled arrivals. Speeds, queues and counts are demo outputs, not observations or estimates from the study. Switching modes keeps the current bags; Reset starts a fresh run.
Start with a baseline.
2018 establishes the reference point for each terminal. The interactive lab lets you vary demand and processing speed; its queues are illustrative, not observed airport data.
The research results above are separate from this lab. The report does not measure individual stage durations, scanner processing speed, staffing levels, or passenger waiting time.
04 / THE ANALYTICAL APPROACH
From observations
to defensible decisions.
Three research questions guide the statistical investigation.
Describe the system
Summarise time and volume across four terminals and two years. Compare centres, spread and distributions before interpreting change.
Reported tools: means, medians, SD, coefficients of variation, heatmaps, scatter plots and boxplots. Figure 16 gives N = 730 daily observations per terminal over the two-year period.
Check the assumptions
Inspect whether the observations follow a normal distribution before choosing how to compare groups.
The report uses Shapiro–Wilk tests (reported p < .001) and Q–Q plots. A paired t-test requires checks on paired differences and a defensible pairing rule; testing each year's distribution alone does not establish those assumptions.
Compare and interpret
Compare terminals, then compare years within terminals. Read the time changes alongside changes in baggage volume.
Reported tests: Kruskal–Wallis across terminals, paired t-tests for T2/T3 and year-based ANOVA for T1/T4. These are reported results, not independently rerun tests; the raw daily dataset was not supplied.
Audit the conclusion
Check the tables against the narrative. Distinguish an observed improvement from a proven effect of automation.
The website recalculates percentage changes from Figure 1. It corrects the interpretation of t-statistics as time savings and flags the inconsistency between T1’s reported means and the narrative conclusion.
05 / THE EVIDENCE
The result is only as good
as its interpretation.
Read what the report found, what it means, and what remains uncertain.
ATerminals have different arrival-time distributions.Reported p < .001
What the output says
Kruskal–Wallis: H = 2455.110, df = 3, p < .001. The reported follow-up comparisons show T3 vs T4 as the exception (p = .798).
How to read it
At least one terminal differs. The omnibus test does not establish which process causes the differences. The shown Tukey comparisons should be checked against a rank-based post hoc procedure with multiplicity adjustment.
Figures 16–18 · printed p. 20 · source tables PDF, page 3.
BT2 and T3 recorded lower times in 2019.10.9% / 12.3% lower
What the output says
T2: t(364) = −42.829, p < .001. T3: t(364) = −47.002, p < .001. The mean time fell by 0.465 and 0.934 source units, respectively.
How to read it
The values 42.829 and 47.002 are t-statistics, not seconds saved. Lower time coincided with lower volume at T2 and higher volume at T3. Volume, schedules and staffing may help explain the changes.
Figures 1, 19–20 · printed pp. 5, 22 · source tables PDF, pages 1 and 4.
CT1 worsened; T4’s small change was inconclusive.Check the direction
What the output says
T1 mean rose from 4.769 to 4.952 (+3.8%); reported F(1,728) = 182.682, p < .001. T4 mean fell from 7.170 to 7.137 (−0.5%); the section 3.3 output reports p = .088.
How to read it
T1’s displayed means show a longer arrival time, despite the narrative calling it an improvement. The T4-labelled discussion refers to tables incorrectly captioned T2. At a .05 threshold, p = .088 does not provide sufficient evidence of a change; it does not prove equivalence.
Figures 1, 22–25 · printed pp. 5, 25 · source tables PDF, pages 1 and 5.
A NOTE ON ANALYTICAL HONESTY
A before-and-after study identifies change. Establishing the cause takes more.
The original daily data and analysis files were not provided. This website visualises transcribed report summaries and reports the original statistical outputs. It does not claim to reproduce the tests or independently verify the historical scanner installation. Time units are retained as unspecified source units.
06 / WHAT I WOULD DO NEXT
Turn a finding
into a better decision.
Observed changes suggest where to investigate. A stronger follow-up would test whether the improvement holds under comparable conditions.
Take the summary dataTHE ANALYST BEHIND THE STUDY
Good analysis makes
complexity useful.
I’m Md Tanvir Rahman. This case study brings together quantitative research, visual exploration and careful interpretation to make an operational question understandable.
Research designExploratory analysisStatistical testingData storytelling
EXPLORE MORE OF MY WORK
Visit my portfolio Website source on GitHub Read the Python notebook Review the study tables