UX Research Study · Queue Management · 2026

Canteen Crowd Analysis

~18 min read · 13 sections · 14 interactive charts · 9 hypotheses · 9 design directions

Every lunch hour, hundreds of students converge on five campus canteens, and wait. This study surveyed 204 students at Dhirubhai Ambani University to measure how long they wait, how it feels, what it costs them, and whether a queue management system could fix it. Each square below is one respondent, colored by how they rated the waiting experience.

All 204 respondents

One square per person · colored by waiting-experience rating (1 = worst, 5 = best) · hover to inspect

204Respondents

81%Visit daily

64%Wait 10+ min

60%Have skipped a meal

78%Would pre-order

2.75Mean rating / 5

View hypothesis results → Design directions → Meet the archetypes →

Overview

Why study a canteen queue, and what can statistics tell us that complaints can't?

In Plain English Everyone on campus "knows" the canteens are crowded. But designing a fix requires knowing precisely what the crowd costs (skipped meals?), what actually drives the bad experience (time? chaos? seating?), who suffers most, and which interventions students would genuinely adopt. This study answers those questions with data instead of anecdotes.

Dhirubhai Ambani University runs five campus canteens serving a predominantly residential student population. Anecdotal complaints about lunch-hour crowding are constant, but anecdotes can't distinguish between a throughput problem, a queue-design problem, and a perception problem. Each implies a different design response.

This survey-based study (N=204) measures waiting behavior, waiting experience, and appetite for two candidate interventions, a pre-order system and a live wait-time display: plus willingness to pay a convenience fee. The analysis pipeline runs from descriptive statistics through nine Bonferroni-corrected hypothesis tests, two regression models, Multiple Correspondence Analysis, and K-Modes clustering that surfaces three respondent archetypes.

Study Scope The pipeline covers 7 analytical stages producing 14 interactive charts and 9 design directions. Headline result: dissatisfaction tracks perceived experience, not raw wait time: once experience rating is controlled for, wait duration no longer predicts meal-skipping. Demand for information and control is near-universal (78–80% "Yes"), but willingness to pay for it is not (13% "Yes").

Methodology

The instrument, the sample, and the seven-stage analytical pipeline.

In Plain English We asked 204 students 11 questions about their canteen habits: where they go, when, how long they wait, how it feels, and whether they'd use (or pay for) tools that reduce the wait. We then tested which patterns in the answers are statistically real, and which are just noise.

The Instrument

A structured Google Forms questionnaire with 11 substantive items: two multi-select behavioral items (canteens visited, visit time slots), three ordinal items (visit frequency, typical busy-day wait, 1–5 waiting-experience rating), one binary outcome (ever skipped a meal due to wait), three adoption-intent items (pre-order system, live wait-time display, convenience fee, each No/Maybe/Yes), one multi-select frustration-factor item, and one open-text item on waiting behavior.

Analytical Pipeline

Data Cleaning & PreparationMulti-select explosion into binary indicators, ordinal recoding with explicit ordering, derivation of frustration-count, thematic coding of open text. The single "Rarely" respondent (n=1) was merged into the "1–3 times" bucket, a stated methodological decision.
Exploratory Data AnalysisDistributions, cross-tabulations, and a frustration-factor co-occurrence matrix.
Bivariate Hypothesis TestingNine pre-registered hypotheses (H1–H9) using Spearman, Kendall, chi-square, Mann-Whitney U, and Kruskal-Wallis tests, every test reported with raw p, Bonferroni-corrected p, and an effect size.
Multivariate ModelingLogistic regression predicting meal-skipping; ordinal logistic regression predicting fee willingness; VIF multicollinearity checks.
Multiple Correspondence AnalysisThe categorical-data analog of PCA, mapping response categories into a shared 2-D space to reveal latent attitude dimensions.
K-Modes ClusteringRespondent segmentation on categorical response patterns; cost-curve model selection across K=2–5; K=3 retained.
Design SynthesisNine design directions, each traceable to a specific statistical result, plus triangulation against three stakeholder interviews.

Respondents

Who answered, and where and when they eat.

204Valid responses0 dropped

165Daily visitors80.9% of sample

2.3Canteens per personMulti-select mean

Lunch & DinnerPeak slots182 each (89%)

In Plain English This is not a sample of occasional visitors: four in five respondents eat at a canteen every single day. Whatever friction exists in the queue, most of these students absorb it twice daily, which is exactly why small per-visit time losses compound into meaningful weekly costs.

Visit frequency

Visits per typical week

Fig 3.1: 80.9% visit daily. The lone "Rarely" respondent is shown here raw, but merged into "1–3 times" for inferential tests.

Time slots visited

Multi-select · share of 204 respondents

Fig 3.2: Lunch and dinner are the twin peaks (89% each). Late-night demand (53%) is a notable third load window.

Canteens visited most often

Multi-select · respondents naming each canteen

Fig 3.3: Padma Kamal (3rd) and ShineStar (4th) dominate footfall (75% and 66%), concentrating queue pressure on two of five kitchens.

Key Finding Demand is doubly concentrated: in time (lunch and dinner windows, 89% each) and in space (two canteens absorb the bulk of visits). Any queue intervention that redistributes load, across time slots or across canteens, attacks the problem at its structural root.

The Wait

How long students wait, how the wait feels, what they do during it, and what it costs.

53%Wait 10–20 min109 / 204

11%Wait 20+ min22 / 204

2.75Mean rating / 5Median = 3

60.3%Skipped a meal123 / 204 · due to wait

In Plain English On a busy day, the typical student waits 10–20 minutes from arrival to food. They rate the experience below the scale midpoint. And for a striking 60%, the wait has at some point been bad enough that they skipped the meal entirely or went off campus, a direct welfare and nutrition cost, not just an annoyance.

Busy-day wait time

Arrival to receiving food

Fig 4.1: 64.2% wait ten minutes or more; only 2% get food in under five.

Waiting-experience rating

1 = very poor · 5 = very good

Fig 4.2: Mean 2.75, below midpoint. Only 10 respondents (4.9%) award a 5.

Wait time × experience rating

Row-share heatmap · each row sums to 100% of that wait bracket

Fig 4.3: The gradient runs corner to corner: longer waits shift the rating mass leftward. This visual pattern is confirmed statistically by H1 (ρ = −0.45, p < 0.0001).

What people do while waiting

Thematic coding of open-text responses

Fig 4.4: Socializing (27%) narrowly beats idle waiting (26%) and phone use (23%). A quarter of the queue is doing nothing, pure dead time.

Key Finding The wait is long, felt, and consequential: 60.3% have skipped a meal because of it. But the waiting time itself is only part of the story, the rating gradient in Fig 4.3, and the regression in Section 08, show that the felt experience of the wait matters more than its raw duration.

Frustration Landscape

Five candidate pain points, which dominate, and which travel together.

In Plain English We gave respondents five possible frustration sources and let them tick all that apply. Two questions follow: which frustrations are most common, and do they arrive alone or in bundles? The chord diagram shows the bundles, thicker ribbons mean two frustrations were ticked together more often.

Frustration factors

Multi-select · share of 204 respondents

Fig 5.1: Crowded counters (80%) and unorganized queues (62%) tower over the rest. The top two are both queue-structure problems, not kitchen-speed problems.

Co-occurrence chord diagram

Ribbon width = respondents selecting both factors · hover ribbons

Fig 5.2: Crowded counter co-occurs heavily with every other factor; it is the hub frustration. Crowding × unorganized queue is the single thickest pairing.

Frustrations per respondent

Count of factors selected (0–5)

Fig 5.3: Frustration arrives in bundles: the modal respondent ticks 2–3 factors. This count correlates with worse experience ratings (H6: ρ = −0.26, corrected p = 0.0013).

Key Finding No single factor explains dissatisfaction, individually, none of the five significantly predicts support for a live wait-time system (H7, all n.s. after correction). But the cumulative count of frustrations does track worse ratings (H6). Frustration is a compound load, and crowding + queue chaos, both fixable by queue design, are its two heaviest components.

Hypothesis Testing

Nine pre-registered hypotheses, Bonferroni-corrected, with effect sizes for every test.

In Plain English A hypothesis is a testable bet placed before looking at the results. Because we placed nine bets at once (plus five sub-tests inside H7), we use the Bonferroni correction, a stricter bar that prevents "significant" results appearing by pure chance. Six of nine hypotheses survive it.

ID	Hypothesis	Test	Statistic	p (raw)	p (corr.)	Effect	Verdict
H1	Longer waits → worse experience rating	Spearman ρ	−0.447	<.0001	<.0001	ρ=−0.45	Significant
H2	Longer waits → more meal-skipping	Chi-square	χ²=10.19	.0170	.153	V=0.19	Not sig.
H3	Visit frequency ↔ experience rating	Kruskal-Wallis	H=10.78	.0046	.041	ε²=0.053	Significant
H4	Meal-skippers rate the experience worse	Mann-Whitney U	U=3220	<.0001	.0001	r=0.35	Significant
H5	Meal-skippers want the pre-order system more	Chi-square	χ²=16.04	.0003	.0030	V=0.26	Significant
H6	More frustrations → worse rating	Spearman ρ	−0.263	.0001	.0013	ρ=−0.26	Significant
H7.1–.5	Each frustration factor → live wait-time support	Chi-square ×5	χ²=0.9–5.4	.066–.31	.33–1.0	V≤0.13	All not sig.
H8	Pre-order support ↔ live wait-time support	Chi-square	χ²=109.7	<.0001	<.0001	V=0.51	Significant
H9	Longer waits → more willingness to pay a fee	Kendall τ-b	−0.065	.314	1.0	τ=−0.07	Not sig.

Bonferroni: H1–H6, H8, H9 corrected as a family of 9 (H7 counted once); H7 sub-tests corrected within their own family of 5. Effect sizes: ρ/τ = rank correlations; V = Cramér's V (bias-corrected); r = rank-biserial; ε² = epsilon-squared.

Three results that matter most H1: wait duration erodes experience (ρ=−0.45, a large rank correlation for survey data). H8: support for pre-ordering and for live wait-times is essentially one attitude (V=0.51, very strong): people who want one want both. H9: even the longest-suffering waiters are not more willing to pay (τ≈0). Pain does not convert into payment.

Behavioral Flow

From wait bracket, to skipping meals, to pre-order appetite, one flow, 204 people.

In Plain English This Sankey diagram traces every respondent left to right: how long they wait → whether they've ever skipped a meal over it → whether they'd use a pre-order system. Follow the thick bands: the dominant path runs from the 10–20 minute bracket, through "skipped: yes," into "pre-order: yes."

Wait → meal-skipping → pre-order willingness

Band width = respondents · hover bands and nodes

Fig 7.1: Of 123 meal-skippers, 105 (85%) answer "Yes" to pre-ordering, versus 66% among non-skippers (H5: V=0.26). The people the queue has already failed are the system's most eager adopters.

Appetite for the three interventions

Pre-order · live wait-time display · convenience fee, share of respondents

Fig 7.2: Information and control are nearly free wins (78% and 80% "Yes"). Payment flips the picture: 58% say "No" to a fee, only 13% say "Yes".

Key Finding Adoption intent is not the bottleneck: under 5% reject either system outright. The bottleneck is the business model: the fee question inverts an 80/20 acceptance pattern into a 13/58 rejection pattern (H9 confirms pain doesn't buy willingness). Design must treat queue relief as core free infrastructure, not a premium add-on.

Multivariate Models

When all predictors compete at once, what still matters?

In Plain English Bivariate tests examine one relationship at a time; regression makes predictors compete. The question here: once we account for how a student rates the experience, does the raw wait duration still predict skipping meals? Answer: no. The feeling beats the stopwatch.

Model 1, Who skips meals? (logistic regression)

Predictor	Odds Ratio	95% CI	p	Reading
Experience rating (per +1 point)	0.556	0.40–0.78	.001	Significant, each rating point nearly halves skip odds
Wait-time bracket	1.144	0.69–1.91	.607	n.s. once rating is controlled
Frustration count	1.176	0.93–1.49	.181	n.s.
Visit frequency	0.877	0.52–1.48	.621	n.s.

Pseudo-R² = 0.085 · LLR p = 0.0001 · all VIF ≤ 1.39 (no multicollinearity concern).

Model 2, Who would pay a fee? (ordinal logistic regression)

Predicting fee willingness (No < Maybe < Yes) from the same four predictors: no predictor reaches significance (all p ≥ .14; ORs 0.74–1.26). Willingness to pay is not explained by wait time, frequency, frustration, or rating, it appears to be a separate trait (price sensitivity), not a function of queue suffering. This independently corroborates H9.

Key Finding Perceived experience is the causal bottleneck. Wait duration influences skipping through its effect on experience (H1), not directly. Design implication: interventions that improve how the wait feels, certainty, fairness, visible progress, can reduce meal-skipping even before the kitchen gets faster.

Multiple Correspondence Analysis

Mapping every answer category into one shared attitude space.

In Plain English MCA places answer categories near each other when the same people tend to choose them together. Reading the map: the top-right corner is the "rejection corner", saying No to pre-ordering sits next to saying No to live wait-times, next to 20+ minute waits and rating 1. The lower-middle is the mainstream adoption zone.

MCA category map

Dim 1 = 14.0% inertia · Dim 2 = 13.3% · colored by variable · hover points

Fig 9.1: Three neighborhoods emerge: a rejection pole (top-right: No-No responses, 20+ min waits, rating 1), a hesitation pole (left: the two Maybes), and a mainstream zone (center-bottom: Yes-Yes, moderate waits, ratings 3–5, fee Yes). Rejecting the systems is the attitude of the worst-served, not the well-served.

Key Finding The Yes answers to pre-order and live wait-time sit almost on top of each other: the geometric confirmation of H8 (V=0.51): these are one "give me information and control" attitude, not two separate feature demands. Note the rejection pole's company: system rejection co-locates with the longest waits and worst ratings, suggesting a small, burned-out "nothing will fix this" group. 2-D inertia is modest (27.3%), typical for MCA; read neighborhoods, not exact distances.

Respondent Archetypes

Three empirically-derived segments from K-Modes clustering on categorical response patterns.

In Plain English Instead of imagining user types, we let the algorithm find groups of respondents who answered similarly across wait time, visit frequency, and all three adoption questions. Three stable segments emerged. They agree on wanting the systems, they split sharply on waiting load and willingness to pay.

Segment profiles across the five clustering variables

Mean coded value per segment (0 = lowest category, higher = more/longer/more willing) · hover bars

Fig 10.1: All three segments score ≈1.7+ on pre-order and live wait-time willingness (out of 2). The separating variables are wait-time bracket (0.99 vs 2.19) and fee willingness (0.00 vs 1.21).

₹

The Convenience Investors

Segment A · N=75 · 36.8%

Typical wait: 10–20 min
Fee willingness: Maybe→Yes (1.21/2)
Skip rate: 56%
Mean rating: 2.93 / 5
Frustrations: 2.57 avg

The only fee-open segment, every "Yes, I'd pay" respondent lives here
Prime audience for optional paid priority pickup
Adoption-ready: near-ceiling on both systems

⏳

The Frustrated Endurers

Segment B · N=77 · 37.7%

Typical wait: 10–20+ min (worst)
Fee willingness: No, unanimous (0.00)
Skip rate: 66% (highest)
Mean rating: 2.40 / 5 (lowest)
Frustrations: 2.83 avg (highest)

Carries the heaviest queue burden, and refuses to pay to escape it
Views queue relief as a fairness entitlement, not a product
Highest meal-skipping: the segment with the most welfare at stake

⚡

The Light-Wait Pragmatists

Segment C · N=52 · 25.5%

Typical wait: 5–10 min (shortest)
Fee willingness: Mostly No (0.38)
Skip rate: 58%
Mean rating: 2.98 / 5 (best)
Frustrations: 2.40 avg (fewest)

Lightest load today, likely off-peak visitors or faster canteens
Still near-ceiling on adopting free systems
Will use tools that keep their waits short, won't pay for speed

Key Finding Willingness to pay is not proportional to suffering, it inverts it. The segment waiting longest (B) is unanimously fee-averse, while moderate-wait Segment A holds all the payment appetite. Regression Model 2 corroborates: fee willingness is a price-sensitivity trait, not a queue-pain response. Naming note: segment labels are descriptive of response patterns, not personality claims.

Stakeholder Voices

Three interviews triangulate the survey: institutional, operational, and governance perspectives.

In Plain English Numbers say what students experience; operators and administrators explain why. Three structured interviews, a senior university administrator, a canteen owner, and the Deputy Convenor of the Canteen Management Committee, surfaced the constraints any design must respect.

Even if the counter moves faster, a fresh dish still takes five to six minutes in the kitchen. We tried an ordering app before, students placed orders and never came to collect them. The food was wasted, and we stopped.

, Canteen Owner (operator interview)

Paraphrased from interview notes; replace with verbatim quote before publishing

Infrastructure expansion isn't on the table this academic year. What the institute can support is better scheduling and better information, systems that spread the load rather than build new counters. And experience says students will not pay extra for campus services.

, Senior Administrative Officer, DAU

Paraphrased from interview notes; replace with verbatim quote before publishing

The committee receives the same complaints every term: lunch-hour crowding and queue discipline. We are open to piloting a student-built system, provided it doesn't add work for canteen staff and doesn't disadvantage students without smartphones.

, Deputy Convenor, Canteen Management Committee

Paraphrased from interview notes; replace with verbatim quote before publishing

Constraints for design (1) Kitchen floor: ~5–6 minutes per fresh dish; no interface can beat physics, so pre-ordering must shift when cooking starts, not promise instant food. (2) No-show risk: a prior ordering app died from uncollected orders; commitment mechanics are mandatory. (3) Zero staff burden & inclusive access: the pilot must not add operator workload or exclude non-smartphone users.

Design Directions

Nine evidence-based directions, every one traceable to a specific statistical result.

In Plain English This is where numbers become design decisions. Each card names its statistical source first, if the evidence changes, the direction changes with it.

DD-01 · H1 + Model 1

Design for the perceived wait, not the stopwatch

Source: ρ=−0.45 (H1); rating OR=0.556, p=.001, the only significant skip predictor; wait bracket n.s. after control

Experience, not duration, drives meal-skipping. Improving how the wait feels reduces harm even before throughput improves.

Show visible progress: order position, live queue length, countdown
Guarantee fairness: one enforced queue, transparent ordering
Reduce uncertainty at arrival, the worst moment is not knowing

DD-02 · H8 + MCA

One system, not two features

Source: χ²=109.7, V=0.51 (H8); MCA co-location of both "Yes" categories

Pre-ordering and live wait-times are one attitude: information + control. Ship them as one integrated experience.

Single app/surface: see live load → decide → pre-order in one flow
Don't A/B them as separate products; adoption is shared
Live data doubles as the pre-order demand signal for the kitchen

DD-03 · H9 + Model 2 + Segment B

Free core, optional priority

Source: τ=−0.07 n.s. (H9); no significant fee predictors (Model 2); Segment B fee willingness = 0.00 despite worst waits

Pain does not convert to payment. Charging for basic queue relief would exclude precisely the students most harmed.

Core queue info + pre-order: free for all students
Monetize only optional add-ons (priority window) aimed at Segment A (36.8%)
Never paywall the live wait-time display

DD-04 · H5 + Sankey

Meal-skippers are the beachhead

Source: V=0.26 (H5); 85% of 123 skippers say Yes to pre-ordering

The 60% who have skipped meals are the most motivated adopters, recruit the pilot cohort from them.

Launch messaging: "never skip lunch again", not "save time"
Measure success as reduced skipping, a welfare metric juries respect
Their H4 pattern (worst ratings) means quick wins land hardest here

DD-05 · H6 + H7 + Fig 5.1

Fix the compound, lead with the queue

Source: ρ=−0.26 (H6) count effect; all single-factor H7 tests n.s.; crowding 80%, unorganized queue 62%

No single fix clears the frustration load, but its two heaviest components are both queue-structure problems.

Physical: marked lanes, single-file rails, clear pickup vs order zones
Digital tokens remove bodies from the physical crush
Track frustration count, not just satisfaction, as the KPI

DD-06 · Fig 4.4

Design the wait itself

Source: waiting behavior, socializing 27%, idle 26%, phone 23%

A quarter of the queue is pure dead time. Occupied, informed waiting feels shorter, classic queue psychology, now with local evidence.

Social waiting zones with seating near (not in) the queue
Order-status screens give the idle 26% something that matters to watch
Phone users (23%) are the natural first users of push status updates

DD-07 · Fig 3.2/3.3 + operator floor

Orchestrate the peaks

Source: lunch/dinner 89% each; two canteens absorb most visits; kitchen floor ≈5–6 min/dish

Demand is concentrated in two windows and two kitchens; the kitchen floor caps burst throughput. Spread the load.

Pre-order pickup slots stagger arrivals across the peak hour
Live cross-canteen load view redirects overflow to Honey/Non-Veg/Ashapura
Slot capacity derived from the 5–6 min/dish floor, not optimism

DD-08 · Operator interview

No-show-proof the pre-order

Source: prior app failure from uncollected orders (operator); committee constraint: zero staff burden

The last ordering app died from no-shows. Commitment mechanics are a launch requirement, not a v2 feature.

Prepayment or token deposit at order time
Short, explicit pickup windows; auto-release + strike policy after
Cooking triggered by proximity check-in, so no order starts unbacked

DD-09 · Clustering + committee constraint

Segment-tiered rollout

Source: K-Modes segments (Fig 10.1); inclusive-access constraint (Deputy Convenor)

Three segments, one free core, different on-ramps, including a non-smartphone path.

Segment B (37.7%): free access + fairness framing, win trust first
Segment A (36.8%): optional paid priority window funds operations
Physical token + printed queue display for non-smartphone access

Limitations & Methods Not Applied

What this study can't claim, and why some famous methods were deliberately left out.

Self-reported, cross-sectional data. Wait times are recalled estimates, not stopwatch measurements; associations are correlational, not causal.
Single-campus sample. Findings describe DAU's five canteens; transfer to other campuses is plausible but unverified.
Coverage gaps. Only one "Rarely" visitor (merged into "1–3 times", a stated decision); only 4 respondents in the "Under 5 minutes" bracket, so its estimates are unstable.
Modest MCA inertia. The 2-D map explains 27.3% of variance, normal for MCA, but read neighborhoods, not precise distances.
K selection. K=3 was chosen from the K-Modes cost curve; categorical clustering lacks a single definitive index, so the archetypes are a useful segmentation, not a natural law.
Paraphrased stakeholder quotes. Interview excerpts are paraphrased from notes pending verbatim verification.

Why no CFA, IRT, or network psychometrics? Those methods require a multi-item psychometric scale, many questions measuring one latent construct (as in a TEQ-style instrument). This survey deliberately uses heterogeneous single-item measures of distinct behaviors and intents. Running factor analysis or IRT here would produce numbers without meaning. The categorical-data equivalents were used instead: MCA in place of factor analysis, K-Modes in place of latent profile analysis, with Bonferroni-corrected tests and effect sizes throughout.