Executive summary
This proposed paper is about inequality in the organization
of college students’ whole days, with sequence analysis as the
main empirical framework and academic time disruption (A–D–A) as a
narrower supplementary lens on study time re-entry.
The pilot supports five conclusions:
- College students’ diary-days fall into five substantively
recognizable configurations: leisure/social-centered (31.8%),
paid-work-centered (25.6%), study-centered (16.7%), class-centered
(15.4%), and domestic-demand-centered (10.4%). Study-centered and
class-centered days are empirically distinct rather than one generic
“academic” type.
- Whole-day complexity and academic opportunity are not the
same construct. Domestic-demand-centered days have the most
transitions and highest turbulence. Study-centered days provide the most
independent study and protected study, but have comparatively low
whole-day turbulence. Class-centered days have the highest entropy
because time is spread across more states; entropy is diversity,
distinct from fragmentation.
- Sequence inequality is real but not a single disadvantage
gradient. Women, parents, and part-time-employed students have
more transition-dense and turbulent adjusted days. In contrast, NH Black
and lower-income students have fewer transitions and lower turbulence
than their reference groups. Nativity differences in these continuous
sequence features are not distinguishable from zero.
- Direct whole-sequence tests show that employment and
institutional calendar matter more than any one focal social
characteristic. Employment uniquely explains 2.11% of pairwise
sequence discrepancy and weekend status 0.99%; gender and race/ethnicity
are statistically detectable but much smaller. The full adjusted set
explains 6.61%.
- The typology is exploratory, not a discovered set of natural
lifestyles. The selected five-cluster solution has a low
silhouette (0.128) and moderate agreement with ordinary Hamming
(ARI=0.664). ATUS observes one day per person. The defensible language
is therefore “diary-day configurations,” accompanied by direct
discrepancy and continuous-feature analyses that do not depend on
cluster assignment.
1. Literature, gap, and contribution
1.1 Higher-education research: from access to time poverty
Higher education often assumes that enrolled students can be
available when courses and services are scheduled and can protect
additional time for coursework. The time-poverty literature makes those
assumptions contestable. Wladis, Hachey, and
Conway (2018) connect parenthood and insufficient time for college
to persistence and academic momentum. Using 2003–2019 ATUS, Conway, Wladis, and
Hachey (2021) show that student parents—especially mothers and
parents of young children—have less discretionary time and more study
time contaminated by simultaneous childcare. Wladis, Hachey, and
Conway (2024) further link time poverty to gender and race/ethnicity
gaps in retention and credit accumulation. The Holistic Capital
Model theorizes time and body/energy as unequally distributed forms
of capital rather than individual deficits. A recent qualitative account
explicitly describes time as a classed privilege and shows how paid work
and institutional schedules compress study into makeshift intervals (Sapir and Strier,
2026).
This literature already establishes that time is an unequally
distributed resource for college. It also recognizes “quality” through
timing, contamination, and fragmentation. The remaining empirical
opportunity is to measure the ordered temporal architecture of
complete student days, not only discretionary minutes or total
academic hours.
The broader sociology of time explains why duration may be
insufficient. Offer
and Schneider (2011) show that gender inequality lies partly in the
content and affective burden of multitasking, not just total workload.
Cornwell,
Gershuny, and Sullivan (2019) identify the sequential structure of
time-stamped behavior as a major direction in time-use research. Lu (2024) connects
fragmented work schedules and role switching to subjective time
pressure. These studies do not establish that every transition is
harmful, but they motivate treating timing, order, overlap, and spell
structure as empirical dimensions in their own right.
1.2 What college-student time-use studies usually measure
Several higher-education studies are person-centered or temporally
intensive, but most do not preserve the ordered 24-hour activity
sequence:
- Fosnacht,
McCormick, and Lerma (2018) use latent profile analysis to identify
first-year student profiles from aggregate time expenditures. This is
the closest higher-education typology precedent, but latent profiles of
hours do not identify whether study occurs before, after, or between
work and care. In addition, more recent time diary data that might
reflect a pandemic-related shift has not been not included.
- Park and Swanson
(2021) examine typical-week hours in study, paid work, and social
media and connect them to a college transition program and GPA.
- What
makes a good study day? (2019) follows university students’ lecture
and independent-study time across days with time-series and multilevel
methods.
- Academic
Time during College (2017) uses repeated daily diaries to connect
academic time to mood, tiredness, and drinking.
- Adaptation of
student behavioural routines during COVID-19 (2023) preserves
within-day timing in 30-minute smartphone diaries and uses non-negative
matrix factorization to compare university-student routines before and
during the pandemic.
These studies establish that student time has heterogeneous profiles,
varies across days, and relates to academic and well-being outcomes.
They also show that preserving within-day timing is feasible and
important. Their samples and estimands, however, differ from a
nationally representative inequality analysis of complete U.S. student
activity sequences.
1.3 Where “sequence analysis” already appears
Sequence methods are not new to higher education, but the sequence is
usually a course trajectory or digital learning
trace, rather than the student’s full day. For example, Ngo and Velasquez
(2023) analyze ordered math course-taking between high school and
community college. Learning-analytics studies sequence clicks, platform
transitions, or task actions inside a course. Those applications answer
important educational-process questions, but they do not observe how
class and study are interleaved with paid work, care, housework, travel,
personal care, and leisure.
Whole-day sequence analysis is well established in general time-use
research. Vagni and
Cornwell (2018) use harmonized time diaries and social sequence
methods to recover recurring patterns of everyday activities across
countries and periods. General time-use research supplies direct
methodological precedents for treating the day as an ordered sequence,
although it is not centered on college students or higher-education
inequality.
1.4 Gap and proposed contribution
In the college-student literature located for this pilot, I did
not find a study that combines all four elements
below:
- a U.S. population-based sample of currently enrolled students,
including working students, parents, older students, undergraduates, and
graduate students;
- complete 24-hour activity sequences that retain timing and
order;
- a higher-education inequality question focused on gender,
race/ethnicity, nativity, socioeconomic resources, employment, and
caregiving; and
- both an interpretable diary-day typology and cluster-independent
tests of whole-sequence differences.
The contribution is therefore not “the first use of sequence analysis
in higher education.” A more accurate claim is:
This project bridges higher-education time-poverty research and
social sequence analysis to show how unequal access to college is
embedded in the organization of complete days.
Methodologically, it contributes by combining: (a) whole-day
typology, (b) multifactor sequence discrepancy analysis, (c) individual
sequence features, and (d) a narrower academic re-entry analysis.
4. Sequence construction and methods
4.1 State sequence
Each diary is converted to 96 states observed at 15-minute interval
midpoints. The mutually exclusive primary-activity states are:
- sleep;
- independent study;
- class/other education;
- paid work;
- primary care;
- domestic work/errands;
- personal care/eating;
- leisure/social activity;
- travel;
- other activity.
Secondary childcare is not a whole-day sequence state because ATUS
does not measure every secondary activity symmetrically. It is retained
in the supplementary intrusion-composition analysis.
4.2 Distances and typology
Dynamic Hamming distance compares activity states at the same clock
time and uses time-varying substitution costs derived from weighted
observed transitions. Weighted k-medoids solutions from k=2
to k=8 are assessed. The selected rule chooses the largest
eligible k ≥ 3 with every cluster at least 5% and
silhouette within 0.01 of the best eligible solution. This yields
k=5.
Distance choice matters; there is no universally optimal sequence
distance (Studer and
Ritschard, 2016). The five-cluster solution’s silhouette is
0.128, and agreement with ordinary-Hamming
k=5 is ARI=0.664. These diagnostics
support useful exploratory differentiation, not strong claims of natural
cluster boundaries.
4.3 Continuous sequence features
Features are calculated with TraMineR:
| Feature |
What it captures |
What it does not mean |
| Normalized entropy |
Diversity/evenness of time across states; ignores order |
Not academic fragmentation and not inherently “bad” |
| State transitions |
Number of boundaries between consecutive activity spells |
Does not distinguish substantively different transition types |
| Mean spell duration |
Average duration of all activity spells |
Not specifically the continuity of study |
| Normalized turbulence |
Variety of subsequences plus unpredictability in spell
durations |
Not an observed unwanted interruption |
| Complexity index |
Combination of state diversity and transitions |
Not a normative quality score |
The revised normalized turbulence follows the sequence-indicator
framework discussed by Ritschard (2022),
building on Elzinga
and Liefbroer (2007).
4.4 Direct sequence-discrepancy analysis
Weighted multifactor discrepancy analysis tests whether within-group
sequences are more similar than between-group sequences using the same
dynamic-Hamming distance matrix (Studer et al.,
2011). Type-II pseudo-R² values represent each
variable’s unique contribution after the other listed covariates. Pilot
p-values use 199 permutations; therefore, the smallest attainable
p-value is 0.005.
4.5 Models and controls
- Typology membership is described with weighted multinomial
standardized predictions and one-vs-rest weighted logits.
- Continuous sequence features use weighted linear models.
- All adjusted models include gender, race/ethnicity (NH White
reference), nativity (U.S.-born reference), current household
income ($100k+ reference), age, enrollment intensity, employment,
parenthood, pre/post period, weekday/weekend, summer
(
June–August), and federal-holiday diary status.
- HC1 is a heteroskedasticity-robust pilot covariance estimator. ATUS
final weights are already present in every model; HC1 changes estimated
uncertainty, not the weights.
- Stars denote † p<.10, * p<.05, ** p<.01, and ***
p<.001.
5. Main findings: whole-day typology first
5.1 Five diary-day configurations
| Diary-day configuration |
N |
Weighted share |
Defining weighted time |
| Leisure/social-centered |
814 |
31.8% |
424 leisure/social min |
| Paid-work-centered |
593 |
25.6% |
474 paid work min |
| Study-centered |
414 |
16.7% |
299 independent study min |
| Domestic-demand-centered |
284 |
10.4% |
299 domestic work/errands min |
| Class-centered |
223 |
15.4% |
353 class/other education min + 127 study min |
Weighted state-distribution chronograms for the
five diary-day configurations
The chronograms show the weighted state distribution across the
24-hour clock. They make the main substantive distinction visible:
study-centered days are organized around long independent-study
periods, whereas class-centered days are organized around scheduled
education with additional independent study. Paid-work- and
domestic-demand-centered days are not merely “low-study” variants; their
daytime structures are dominated by different obligations.
5.3 Correlations show which measures are redundant—and which are
not
These are descriptive weighted Pearson correlations across all 2,328
diaries.
| Measure 1 |
Measure 2 |
Weighted r |
| Transitions |
Turbulence |
0.993 |
| Transitions |
Complexity index |
0.964 |
| Turbulence |
Complexity index |
0.974 |
| Entropy |
Turbulence |
0.686 |
| Transitions |
Mean spell duration |
-0.880 |
| Independent-study minutes |
Protected 60-minute block |
0.757 |
| Whole-day turbulence |
Protected 60-minute block |
0.013 |
Main takeaway: transitions, turbulence, and the
complexity index are nearly interchangeable in this 15-minute
operationalization (r=.96–.99). Reporting all three as
separate “findings” would exaggerate the evidence. Entropy is related
but not identical. Most importantly, whole-day turbulence is essentially
uncorrelated with having a protected 60-minute study block
(r=0.013). General diary complexity and usable academic
time are empirically distinct.
6. Inequality in the complete sequence
6.1 Cluster-independent discrepancy analysis
| Variable |
Unique pseudo-R² |
Permutation p |
| Gender |
0.255% |
0.005** |
| Race/ethnicity |
0.434% |
0.010* |
| Nativity |
0.072% |
0.724 |
| Current household income |
0.192% |
0.065† |
| Age group |
0.435% |
0.005** |
| Enrollment intensity |
0.129% |
0.010* |
| Employment |
2.107% |
0.005** |
| Parenthood |
0.173% |
0.005** |
| Pre/post period |
0.071% |
0.784 |
| Weekend |
0.987% |
0.005** |
| Summer |
0.246% |
0.005** |
| Holiday diary |
0.270% |
0.005** |
| Full model |
6.605% |
0.005** |
Main takeaway: the complete sequence is socially
patterned, but employment and the calendar are the largest observed
organizers. Employment’s unique contribution (2.107%) is roughly eight
times the gender contribution (0.255%) and five times the race/ethnicity
contribution (0.434%). Gender and race/ethnicity remain statistically
detectable net of student roles and calendar; parenthood and enrollment
intensity are also significant but small. Income is marginal
(p=.065), while nativity and pre/post period are not
distinguishable from zero. The full model’s 6.605% means that most
diary-to-diary sequence heterogeneity remains unexplained.
This result is more defensible than telling the story only through
cluster membership because it uses the complete distance matrix and does
not require treating k=5 as known.
6.2 Who is likely to inhabit each configuration?
Adjusted typology membership by student
group
The heatmap contains multinomial standardized probabilities. The
following table adds selected one-vs-rest adjusted odds ratios.
| Adjusted contrast |
Configuration |
Predicted probability |
OR [95% CI] |
| Woman vs man |
Domestic-demand |
13.0% vs 6.8% |
2.11*** [1.56, 2.87] |
| Woman vs man |
Leisure/social |
28.2% vs 36.9% |
0.64*** [0.53, 0.78] |
| NH Black vs NH White |
Leisure/social |
37.4% vs 29.8% |
1.46* [1.07, 1.98] |
| NH Black vs NH White |
Domestic-demand |
5.7% vs 12.3% |
0.41** [0.24, 0.70] |
| NH Asian vs NH White |
Domestic-demand |
7.6% vs 12.3% |
0.57* [0.33, 0.98] |
| Foreign-born vs U.S.-born |
Study |
20.6% vs 16.0% |
1.42* [1.02, 1.98] |
| $50k–99,999 vs $100k+ |
Leisure/social |
28.1% vs 35.1% |
0.71** [0.56, 0.90] |
| Parent vs not parent |
Study |
14.5% vs 18.1% |
0.76* [0.59, 0.97] |
| Part-time vs full-time enrollment |
Class |
8.7% vs 17.1% |
0.42*** [0.29, 0.62] |
| Not employed vs employed full-time |
Study |
25.9% vs 10.3% |
3.31*** [2.39, 4.57] |
| Weekend vs weekday |
Leisure/social |
47.4% vs 25.9% |
2.84*** [2.30, 3.50] |
| Weekend vs weekday |
Class |
2.5% vs 20.2% |
0.09*** [0.05, 0.15] |
Main takeaway: the strongest inequality result is
gendered access to domestic-demand versus leisure/social days. Women
have a 13.0% adjusted probability of a domestic-demand-centered day
versus 6.8% for men, while study-centered membership is virtually
identical by gender. Foreign-born students are more likely to have a
study-centered day, but nativity does not explain the full sequence in
the discrepancy test; this illustrates why one typology contrast should
not be generalized to the entire diary.
Race and income results are mixed and sometimes counter to a simple
disadvantage narrative. NH Black students are more likely than NH White
students to have a leisure/social-centered diary and less likely to have
a domestic-demand-centered diary in these adjusted one-day data.
Middle-income students are less likely than the $100k+ group to have a
leisure/social-centered day; the lower-income contrast is not
significant.
Employment, enrollment, and calendar contrasts are large. Students
not employed are much more likely than full-time workers to have
study-centered days; part-time students are less likely to have
class-centered days; weekends shift diaries away from class and toward
leisure/social activity. The typology captures both inequality and the
institutional scheduling of college.
6.3 Inequality in continuous sequence features
Each cell reports the adjusted prediction for the first group versus
its stated reference; stars report adjusted difference between the two
groups.
| Contrast |
Entropy |
Turbulence |
Transitions |
Mean spell min |
| Woman vs man |
0.626 vs 0.610** |
0.159 vs 0.150** |
13.39 vs 12.44** |
112.4 vs 118.6** |
| Parent vs not parent |
0.631 vs 0.612*** |
0.160 vs 0.152** |
13.46 vs 12.68** |
111.1 vs 117.6** |
| Part-time vs full-time employed |
0.640 vs 0.610*** |
0.161 vs 0.152** |
13.58 vs 12.78* |
109.9 vs 115.5† |
| NH Black vs NH White |
0.599 vs 0.622** |
0.147 vs 0.157* |
12.09 vs 13.18** |
124.7 vs 113.4** |
| Under $50k vs $100k+ |
0.616 vs 0.620 |
0.151 vs 0.158* |
12.55 vs 13.34* |
119.0 vs 112.1* |
| Foreign-born vs U.S.-born |
0.627 vs 0.618 |
0.158 vs 0.154 |
13.26 vs 12.94 |
114.0 vs 115.2 |
Adjusted whole-day sequence features by student
group
Main takeaway: women, parents, and students employed
part-time have more diverse, transition-dense, and turbulent days than
their reference groups. NH Black students and students in households
under $50k have fewer transitions, lower turbulence, and longer spells
than NH White and $100k+ students, respectively. Foreign-born and
U.S.-born students do not differ clearly on these features.
The correct conclusion is therefore multidimensional
inequality, not “every disadvantaged group has more fragmented
days.” The project can contribute precisely by showing that paid work,
unpaid obligations, class schedules, and academic opportunity produce
different forms of temporal constraint that are not captured by a single
high-complexity score.
7. What typology membership is—and is not—linked to
The adjusted academic columns in Section 5.2 establish strong
concurrent characterization:
- Study-centered days predict about 290 independent-study minutes and
an 87.7% probability of a protected 60-minute study block.
- Class-centered days predict about 395 broad-academic minutes but
only 119 independent-study minutes.
- Domestic-demand-centered days predict about 34 independent-study
minutes and an 18.6% protected-block probability.
These associations validate the substantive labels, but they are not
external outcome tests because both the typology and the academic-time
measures are constructed from the same diary. ATUS does not contain GPA,
credits earned, retention, persistence, or learning outcomes. It would
be incorrect to say that typology membership “predicts academic success”
in the present data.
For this paper, the outcome is the organization and usability
of time itself. A later study with repeated diaries plus
administrative outcomes could test whether day configurations predict
performance or persistence beyond total study minutes.
8. Undergrad/graduate and institution-type sensitivity
8.1 Student level
| Sample |
N |
Leisure/social |
Paid work |
Study |
Domestic demand |
Class |
| All college/university students |
2,328 |
31.8% |
25.6% |
16.7% |
10.4% |
15.4% |
| Likely undergraduate (PEEDUCA <=42) |
1,508 |
34.3% |
20.9% |
16.2% |
10.9% |
17.7% |
| CPS-linked undergraduate |
433 |
29.9% |
17.1% |
23.0% |
9.3% |
20.7% |
| CPS-linked graduate |
178 |
18.6% |
38.3% |
26.2% |
4.0% |
13.0% |
The likely-undergraduate restriction is close to the full-sample
distribution. In the smaller CPS-linked sample, graduate students are
much more likely to have paid-work- and study-centered days and less
likely to have leisure/social- or domestic-demand-centered days. Because
CPS level is measured in a linked October supplement rather than
necessarily on the diary date, these are sensitivity descriptions, not
the core estimand.
8.2 Two-year versus four-year linked undergraduates
| Institution |
N |
Leisure/social |
Paid work |
Study |
Domestic demand |
Class |
| 2-year |
115 |
33.6% (n=42) |
19.7% (n=26) |
17.2% (n=22) |
9.0% (n=10) |
20.5% (n=15) |
| 4-year |
318 |
28.8% (n=105) |
16.3% (n=53) |
24.8% (n=73) |
9.4% (n=40) |
20.8% (n=47) |
Four-year linked undergraduates are more study-centered than two-year
students (24.8% vs 17.2%), while the two-year group is somewhat more
leisure/social- and paid-work-centered. However, the total two-year
sample is only 115 and four of its five typology cells have fewer than
30 observations. These estimates show the potential of adding more waves
to test for institutional heterogeneity.
9. Supplementary analysis: academic A–D–A re-entry
9.1 Construction
Two academic anchors are retained:
- Independent study: ATUS
060301,
degree/certification research or homework.
- Broad academic: class (
060101) +
independent study (060301) + other degree-related education
(060401).
The motif is academic primary activity → nonacademic primary
activity → return to the same academic anchor within 60
minutes. The primary risk set requires at least 60 total
minutes in the relevant academic anchor that day; it does not condition
on already observing two study spells. Models additionally control total
same-day anchor minutes. Return windows of 30 and 90 minutes and
exposure thresholds of 30 and 90 minutes are robustness checks.
9.2 Feasibility and threshold robustness
| Specification |
Independent study |
Broad academic |
| Any anchor on all diary-days |
44.2% |
53.2% |
| Mean anchor minutes on all diary-days |
98.7 |
158.6 |
| 60-minute A–D–A among days with ≥60 anchor min |
25.9% |
32.3% |
| Minimum anchor minutes |
Independent N |
Independent A–D–A |
Broad N |
Broad A–D–A |
| 30 |
927 |
24.8% |
1,083 |
31.5% |
| 60 |
884 |
25.9% |
1,043 |
32.3% |
| 90 |
785 |
28.8% |
954 |
34.5% |
| Return window |
Independent academic days |
Broad academic days |
| 30 |
13.8% |
17.5% |
| 60 |
24.3% |
31.2% |
| 90 |
26.9% |
36.3% |
A–D–A is sufficiently frequent and changes smoothly across the
prespecified thresholds. That supports measurement feasibility. It does
not prove that every observed return is a harmful or
unwanted interruption. A lower A–D–A rate can mean continuous study, no
return after leaving, or insufficient opportunity to return.
9.3 What appears inside broad-academic A–D–A blocks?
Percentages are weighted and multi-label; an intervening block can
contain both an unpaid obligation and personal/leisure activity.
Secondary childcare is added to unpaid obligations. Paid work remains
separate.
| Group |
A–D–A day N |
Paid work |
Unpaid obligation |
Personal/leisure |
| Man |
141 |
4.0% |
31.8% |
85.7% |
| Woman |
165 |
1.9% |
23.0% |
88.4% |
| NH White |
179 |
2.1% |
27.5% |
86.9% |
| NH Black |
33 |
8.4% |
23.3% |
93.3% |
| NH Asian |
42 |
4.1% |
21.9% |
83.9% |
| Hispanic |
45 |
1.0% |
38.3% |
86.4% |
| NH other/multiracial |
7 |
0.0% |
7.1% |
100.0% |
| U.S.-born |
249 |
3.0% |
27.7% |
87.1% |
| Foreign-born |
57 |
2.0% |
23.0% |
87.7% |
| Under $50k |
126 |
4.7% |
31.4% |
81.0% |
| $50k–99,999 |
84 |
1.1% |
18.0% |
88.2% |
| $100k+ |
96 |
2.4% |
31.2% |
93.8% |
Paid-work intrusion appears on only seven complete-case
broad-academic A–D–A days, so subgroup paid-work comparisons are not
stable. Unpaid-obligation estimates are more feasible but remain
selected on first experiencing an A–D–A event. The table is best used to
motivate mechanisms, not to rank groups.
Raw A–D–A prevalence is concentrated in study- and class-centered
days because those configurations contain much more academic exposure.
Once same-day academic minutes are adjusted, independent-study A–D–A
predictions across the five configurations narrow to roughly 23%–30%.
This is another reason to keep A–D–A supplementary rather than use it as
the paper’s central typology validation.
10. Overall interpretation and paper contribution
The sequence-centered results make a coherent narrative:
- The unit of inequality is the organized day. Higher
education participation is embedded in qualitatively different
configurations of study, class, paid work, domestic obligations, travel,
and leisure.
- Academic opportunity and whole-day complexity are
separable. A study-centered day can be academically intensive
and structurally simple; a domestic-demand day can be highly turbulent
and contain little academic exposure.
- Different social positions map onto different temporal
mechanisms. Gender is most visible in domestic-demand versus
leisure/social membership; parenthood and part-time work in sequence
complexity; enrollment and employment in academic configurations; race
and income in mixed, non-monotonic patterns.
- Institutions organize time alongside students’ social
circumstances. Weekday/weekend, academic season, holidays, and
enrollment intensity are substantive parts of the temporal inequality
story, not nuisance controls.
- The methodological contribution is triangulation.
Typology makes the configurations interpretable; discrepancy analysis
tests the full sequence without clusters; entropy/turbulence/spells
identify form; A–D–A isolates one academic re-entry process.
A defensible paper claim is:
Inequality in higher education concerns not only the amount of time
students can devote to college, but the kinds of days through which
academic participation must be accomplished.