• Open Access

Time use clusters of New Zealand adolescents are associated with weight status, diet and ethnicity

Authors

  • Katia Ferrar,

    1. Health and Use of Time Group, School of Health Sciences, University of South Australia; Sansom Institute for Health Research, University of South Australia
    Search for more papers by this author
  • Tim Olds,

    1. Health and Use of Time Group, School of Health Sciences, University of South Australia; Sansom Institute for Health Research, University of South Australia
    Search for more papers by this author
  • Carol Maher,

    1. Health and Use of Time Group, School of Health Sciences, University of South Australia; Sansom Institute for Health Research, University of South Australia
    Search for more papers by this author
  • Ralph Maddison

    1. National Institute for Health Innovation, The University of Auckland, New Zealand
    Search for more papers by this author

Correspondence: Katia Ferrar, Health and Use of Time Group, School of Health Sciences, University of South Australia, Centenary Building, Room C7-42, City East Campus, North Terrace, Adelaide SA 5000; e-mail: ferke001@mymail.unisa.edu.au

Abstract

Objective : To describe New Zealand adolescent time use clusters and correlate cluster profiles.

Methods : Data were from the cross-sectional 2008/2009 National Survey of Children and Young People's Physical Activity and Dietary Behaviours, which surveyed a random sample of 10–16 year-old New Zealanders (study subset n=679). Time use data were collected using the Multimedia Activity Recall for Children and Adults, and collapsed into 17 age-adjusted variables for sex-specific cluster analysis. Cluster associations with socio-demographic, anthropometric, physical activity and dietary variables were analysed.

Results : Three time use clusters were discovered for both boys and girls. For boys, the Techno-active cluster was characterised by high levels of team sports and TV; the Quiet movers cluster by transport (active and passive) and quiet time; and the Social studious cluster by reading, study activities and social interaction. The boys’ clusters were associated with ethnicity. The girls’Social sporty cluster was characterised by sports and social interaction; the Screenie tasker cluster by TV, computer, chores and work; and the Super studious cluster by reading, study and school-based activities. The girls’ time use cluster membership was associated with weight status and serves of extra foods.

Conclusions : Distinct sex-specific time use clusters and correlate profiles exist among NZ adolescents.

Implications : These findings may assist the development of targeted time use interventions to improve adolescent health and well-being.

In recent times, the heath of adolescents and young people has been considered a priority area by governments around the world. The New Zealand (NZ) Ministry of Health has identified the current health issues in young people (12–24 years) as mental health problems, alcohol abuse, increasing rates of obesity and decreasing rates of physical activity.1 Significant health inequalities exist among NZ adolescents, with young Māori and Pacific Island people in particular experiencing poorer health outcomes than their peers.2 Māori and Pacific adolescents (10–14 years) are more likely to suffer chronic health conditions, asthma, skin conditions, poorer dental health and overweight and obesity.3 Recent data suggest that 10% of 10–18 year olds in NZ are obese and an additional 24% are overweight.4

Recent interventions to improve the health and well-being of adolescents have been implemented at government, community and individual levels.5–7 Yet it appears NZ adolescents are receiving only small benefits, with health trend data suggesting little change1 and health-related interventions yielding generally small effect sizes.8–10 Relationships between individual aspects of time use, such as screen time and health outcomes, have been investigated.11,12 However, there is a suggestion that multi-dimensional patterns of behaviour may affect health in ways not explained when such one-dimensional relationships are investigated.13 Exploring young people's multi-dimensional time use behaviours may further our understanding of the complex health and well-being relationships, and offer insights into the design of targeted health interventions.

Cluster analysis is classified as an unsupervised data mining algorithm which attempts to group the data into classes or clusters, such that ‘cases’ within the clusters are similar to each other and relatively dissimilar to the ‘cases’ in the other clusters. Cluster analysis allows empirical definition of data patterns, and does not rely on current theory or knowledge in the related field of study. In recent years, the research fields of dietary patterning14 and disease symptomology15 have successfully utilised cluster analysis to identify underlying patterns in data.16 Cluster analysis to identify adolescent time use patterns has not been a common approach to date. Only 19 adolescent time use clustering studies could be sourced as part of a recent systematic review of the literature.17 The studies reported on adolescents from different countries and relative socioeconomic backgrounds. Regardless of the differences, some similar multi-dimensional time use cluster patterns were identifiable, as were patterns of relationship with socio-demographic variables. No time use clustering studies have reported the time use clusters or patterns of NZ adolescents.

The aims of this study are to investigate the time use clusters among NZ adolescents (10–16 years) and determine which time use activity, socio-demographic, anthropometric, physical activity and diet variables best characterise each cluster. To achieve this aim, the study will cluster-analyse 24-hour recall data from a nationally representative sample of NZ adolescents.

Methods

Sample and design

The participants for this study (n=679) were a subset of randomly selected NZ boys and girls aged 10–16 years, interviewed as part of the 2008/2009 National Survey of Children and Young People's Physical Activity and Dietary Behaviours (full survey sample age 5–24 years, n=2,503; survey sample aged 10–16 n=1,093). The study used a stratified multistage sampling design using meshblocks as the primary sampling unit. Meshblocks were randomly sampled based on geographical location and rural or urban status, and households within the selected meshblocks were randomly selected to participate. Families were not obliged to participate but, if willing, only one child per participating household was surveyed. A Kish grid approach was used to select a single eligible respondent. The response rate was calculated as the total number of complete interviews divided by the total number of eligible households plus the estimated number of non-contact households that were eligible (i.e. estimated unobserved eligible). The overall response rate was 55%.18 The survey was conducted according to the ethical principles outlined in the Declaration of Helsinki and was covered by Statistics New Zealand Tier 1 ethical approval. Written consent was obtained from all participants aged 15 years and over, or their parent, depending on the age of the participant.

Measures

Socio-demographic, diet and anthropometric data were gathered during a computer-assisted face-to-face interview. Height and body mass were measured using a stadiometer [Invicta 2007246, Leicester, UK) to the nearest 0.1 cm, and digital scales (Tanita UM-070, Illinois, US) to the nearest 0.1 kg, using standard methodologies19 by trained interviewers in the participants’ homes. Participants wore light clothing and no shoes for the measurements. Body mass index (BMI) was calculated from measured height and weight [body mass (kg)/height (m)2]. Weight status was calculated as thin (Grade I, II and III combined), normal, overweight and obese according to the criteria of Cole et al.20,21 Decimal age (years) was determined based on the reported date of birth and the date of the interview. The 2006 Census ethnicity question was used to determine participant ethnicity using the total response method, which allows participants to be included in all the ethnic groups they self-identify with.22 For the purpose of data analysis, one ethnic category was assigned to each participant based on the prioritisation method in the following order: Māori, Pacific peoples, Asian, New Zealand European and Other. Meshblocks were used to assign a level of deprivation based on the 2006 New Zealand Deprivation Index.23 The New Zealand Level of Deprivation is constructed with nine variables reflecting eight types of deprivation: employment, home ownership, income, transport, qualifications, living space, communication and support.23 Participants were assigned to quintiles of deprivation (1=least deprived to 5=most deprived). Geographical area (rural or urban) was determined based on meshblocks and using Statistics New Zealand information.

Dietary data were collected during the initial face-to-face interview. A dietary habits questionnaire was modified specifically for the survey based on a Dietary Habits Questionnaire (DHQ) developed for the 2008/09 New Zealand Adult Nutrition Survey.24 No reliability or validity data were available regarding the dietary recall questionnaire. The recall period for the questionnaire was the previous month. The variables used for these analyses are servings of fruit, vegetables and a combined variable ‘extra food’ (including soft drinks, chocolate, potato chips, hot chips, meat pies and fast foods). The categorical responses to the fruit and vegetable habit questions (average serves per day) were converted into continuous data (serves/day) as follows: never = 0; less than 1 serving = 0.5; 1 serving = 1.0; 2 servings = 2.0; 3 = 3.0; and 4 or more servings = 4.0. The ‘extra foods’ frequency responses were converted to equivalent servings per week (continuous variable): never = 0; less than once per week = 0.5; 1–2 times per week = 1.5; 3–4 times per week =3.5; 5–6 times per week = 5.5; and 7 or more times per week = 7.0.

Seven to 14 days after the face-to-face interview, participants were again interviewed by telephone. All participants completed four 24-hour time use recalls, two at the time of the face-to-face interview and two during the follow-up telephone interview. Time use data were collected using the Multimedia Activity Recall for Children and Adults (MARCA),25 a computerised 24-hour recall linked to a compendium of energy expenditures. The software allows participants to recall everything they did on the previous day. The participants chose from a list of 259 activities, plus an additional 22 culturally specific Māori activities. They could recall activities in time slices as small as five minutes. The MARCA has a same-day test-retest reliability of r = 0.84–0.92 for major outcome variables [moderate to vigorous physical activity (MVPA), physical activity level (PAL) and screen time (the number of minutes spent watching television, playing videogames and using a computer)],25 and convergent validity with reference to pedometry of rho = 0.54 for PAL.26 Time use data were self-reported by all participants included in this study (age 10–16 years). The participants in this study recalled at least two days (out of the potential four) of which at least one was a school day.

Of the 1,093 eligible participants (10–16 years), only 679 met the inclusion criteria of at least two 24-hour recalls containing at least 1,400 minutes of time use data covering at least one school day. Factors such as weight status, age and socioeconomic position are commonly associated with various time use behaviours.27

To determine whether the included sample demographic differed to those participants who were excluded, statistical analysis (t-test and chi square) were conducted. The cut point of 1,400 minutes was chosen to maximise the sample size. In total, 246 boys and 233 girls included in the sample recorded more than 1,430 minutes. The 281 MARCA activities were aggregated hierarchically to form 17 variables, which were used as cluster inputs. Variable aggregation was conducted independently by two authors, with disagreements resolved via consensus, and was based on variable similarity and aimed to preserve comparability with previous time use studies. Variable aggregation reduces the degree of data skewness, and reduces the noise generated by high numbers of cluster inputs, which in turn reduces the risk of masking of the underlying cluster structure.28 Because there are differences in school and non-school activity patterns25 and because children spend about one day in two in school, school and non-school data were weighted equally. As the relationship between time use and age is non-linear, time use data were corrected for age and sex by regressing the values against age for boys and girls separately, fitting a fourth order polynomial and using the residuals in analysis. The age and sex-adjusted values are reported.

Each participant was requested to wear an ActiGraph accelerometer during waking hours for seven consecutive days. Data were included from any participant who provided at least one day of valid accelerometer data (minimum of 600 valid minutes). The inclusion criterion of one valid day's worth of data was selected to maximise the sample size while ensuring data quality. Valid minutes were defined as a recorded minutes that did not fall into a sequence of ≥20 minutes of zero activity counts.18 Accelerometers recorded data every 10 seconds.29 The ActiGraph has demonstrated acceptable validity against criterion measures (oxygen consumption and energy expenditure), with coefficients ranging from 0.50–0.89.30–32 The average daily time spent in light/moderate/vigorous intensity activities was calculated from all valid available accelerometer data using Freedson cut-off points.33 For the purpose of this study, the average daily minutes of moderate-to-vigorous activity (MVPA) as derived from the accelerometer counts was used as an objective measure of activity. Average daily accelerometer wear time (minutes/day) varied between participants, but was not considered a significant factor in average MVPA, as regression calculation suggested only a weak relationship (boys slope = 0. 07; girls slope = 0.03).

Statistical analysis

The 17 age- and sex-adjusted activity sets (residual data) were used as cluster analysis inputs. The TwoStep Cluster Algorithm was used for cluster analysis.34 This method offers some key advantages over other methods of cluster analysis such as k-means or Ward's method, in that it requires no a priori selection of expected cluster numbers, and that it utilises an algorithm similar to the BIRCH algorithm,35 which allows it to handle large data sets. All cluster analyses were conducted using IBM SPSS Statistics Version 19. As is convention, the labelling (naming) of each cluster reflects the predominant activity variable/s, see Table 2. Clusters names were developed by consensus between two researchers (KF and TO).

Table 2.  Gender-specific cluster age-and sex-adjusted mean activity set variables.
 Boys’ clustersGirls’ clusters
Cluster namesTechno-activeQuiet moversSocial studiousSocial sportyScreenie taskerSuper studious
  1. Sample size is show as n (%). Time use variables shown in age- and sex- adjusted min/day mean (SD).

Sample size181 (52%)63 (18%)104 (30%)81 (25%)89 (27%)161 (49%)
Activity set       
Sleep627.1 (62)575.6 (78)571.1 (56)615.9 (70)597.7 (44)602.5 (60)
TV159.2 (90)111.5 (76)98.7 (60)87.4 (53)183.8 (86)95.6 (58)
Videogames39.1 (66)19.0 (35)39.2 (52)34.8 (67)9.5 (19)5.8 (12)
Computer15.1 (22)49.1 (69)15.9 (24)9.1 (15)37.8 (49)13.2 (18)
Social interaction8.8 (16)6.6 (15)27.7 (41)37.8 (54)13.8 (20)8.7 (13)
Grooming40.3 (16)34.0 (11)51.4 (20)59.0 (21)58.2 (26)55.7 (18)
Eating60.5 (18)57.2 (19)77.2 (27)62.4 (18)69.4 (21)67.0 (17)
Quiet time40.7 (32)94.2 (72)50.1 (41)76.2 (56)61.8 (43)77.4 (65)
Play56.1 (52)46.0 (38)57.6 (53)35.5 (38)35.7 (27)60.5 (62)
Team sports58.3 (57)24.4 (35)23.2 (30)31.7 (41)18.6 (27)12.4 (17)
Non-team sports25.4 (31)29.3 (35)48.0 (60)70.5 (64)22.9 (31)36.8 (35)
Active transport37.0 (30)91.1 (72)37.7 (30)51.6 (41)38.5 (28)44.3 (36)
School138.0 (62)162.8 (87)146.6 (62)99.3 (53)110.6 (53)164.2 (64)
Study/homework/music25.7 (31)24.2 (32)71.5 (69)20.1 (27)26.7 (31)54.9 (50)
Reading19.8 (24)25.3 (33)47.5 (48)28.0 (36)20.5 (25)42.2 (42)
Passive transport34.3 (28)52.7 (54)44.7 (38)50.7 (42)38.4 (28)45.5 (30)
Chores & work52.6 (50)35.6 (44)30.8 (35)68.3 (52)94.5 (70)51.8 (35)

Classification Trees with Tree Boost using the DTREG Program36 were used to determine how ‘well’ the time use variables classify participants into the assigned clusters (classification error value). The process has two stages: first, a Classification Tree is developed (training stage); and second, the result is validated by v-fold cross-validation (validation stage), and is reported as the training and validation misclassification error. V-fold cross-validation is used to determine the optimal model, whereby random (v) samples are drawn from the original data set, analysed and (v) results are averaged. In addition, the overall importance values for each variable are generated, and reflect how important each variable is in discriminating cluster membership.

ANOVA (continuous variables), chi-square (nominal variables) and Kruskal-Wallis (ordinal variables) tests, with cluster membership as the independent variable, explored univariate associations. Tamhane tests (unequal variances) and Bonferroni tests (equal variances, ANOVA), standardised residuals (chi-square), and pairwise comparisons of subgroups37 (MedCalc version 12.1.1.) Kruskal-Wallis post hoc tests were performed. Alpha was set at 0.05 and IBM SPSS Statistics Version 19 was used. Classification Trees with Tree Boost were used to determine the predictive importance of the correlates to cluster membership.

Results

Sample demographics

Table 1 details the demographic characteristics of the participants. When the included participants’ (n=679) socio-demographic details were compared to the excluded participant details (n=414), there was no statistical differences. The two datasets were similar with regard to age (included = 13.2 years and excluded = 13.3 years; p=0.38), level of deprivation (p=0.48) and weight status distribution (p=0.11).

Table 1.  Demographic characteristics of the sample.
 MalesFemales
  1. Body mass index (BMI) was calculated from measured height and weight. Age, height and weight shown as mean (SD) and the remainder of variables are shown as proportions. Adolescent weight status was classified according to Cole et al. 20,21 criteria. Level of deprivation was based on the 2006 New Zealand Deprivation Index.23

n348331
Age (years)13.2 (2.0)13.2 (2.0)
Height (m)1.60 (0.13)1.57 (0.10)
Weight (kg)55.73 (17.88)54.53 (16.79)
Weight status
Thin
Normal
Overweight
Obese

5.8
57.1
23.8
13.3

4.0
59.2
24.5
12.3
Region
Urban
Rural

84.2
15.8

82.2
17.8
Level of Deprivation
1 (least deprived)
2
3
4
5 (most deprived)

18.2
24.0
22.3
14.7
20.8

23.0
16.7
18.8
17.9
23.6
Ethnicity
Māori
Pacific
Asian
NZ/European

18.7
8.6
14.9
57.8

17.5
8.8
10.6
63.1

Time use clusters

Three clusters were generated for each of the sexes using the 17 age- and sex-adjusted residual values. Using the cluster membership derived from the cluster analysis the age- and sex-adjusted real value means and standard deviations for each activity set variable for each sex-specific cluster were calculated (Table 2). Across the top of Table 2 are the three boys’ clusters and the three girls’ clusters, and the 17 activity variables are listed down the left hand side.

For boys, the misclassification error was 4.0 and 14.1% (training; validation) for the Classification Tree Boost analysis. For girls, the misclassification error was 2.7 and 14.3% (training; validation) for the Tree Boost analysis. For the boys, the top five cluster inputs ranked by overall importance were active transport, sleep, team sports, quiet time and TV; and TV, school, computer, chores and work, and social interaction for girls.

As is convention, the clusters were named based on the activity set(s) that appeared to dominate each cluster, i.e. those with large differences in the means across the clusters (Table 2), and the activity variables already identified as important to cluster membership (variable of importance ranking).

The sample sizes of the resultant boys’ clusters were uneven, ranging from 63 to 181 (18% to 52% of the sample). The boys’ time use Cluster 1 was named Techno-active, members of which were identified by the highest levels of team sports, TV viewing and sleep. Boys’ Cluster 2 was named Quiet movers due to the highest values for both active and passive transport and quiet time. The third boys’ cluster was named Social studious as the members reported the highest mean values for reading, study/homework/music and social interaction (Table 2).

The girls’ cluster sample sizes were also somewhat uneven, ranging from 81 to 161 (25% to 49% of the sample). When compared to the boys’ clusters, the girls’ clusters were characterised by different activity patterns. The girls’ Cluster 1 was named Social sporty due to the high reported values of social interaction and team and non-team sports. Cluster 2 was named Screenie tasker due to the highest recorded values of TV, computer, chores and work. Finally the girls’ Cluster 3 was named Super studious due to the large amount of reported time spent reading and doing homework, music and study and school activities.

The results of these univariate association tests between cluster membership and correlates are displayed in Table 3.

Table 3.  Cluster-correlate associations for both boys’ and girls’ clusters.
Variable categoryVariable nameBoys’ Clusters Girls’ Clusters 
  Techno-activeQuiet moversSocial studious p valueSocial sportyScreenie taskerSuper studious p value
  1. Note: Body Mass Index (BMI), Moderate to Vigorous Physical Activity (MVPA), age and diet variable values are shown as means, and the remaining variables are shown as cluster-specific proportions. Bolded proportion values indicate post hoc testing attributed significance (>2 or <−2 standardised residuals). Values with same superscript in the same row are significantly different. *= Kruskal-Wallis post hoc pairwise comparisons showed significant differences between cluster pairs.

Anthropometry          
 BMI (kg.m−2)21.821.520.90.3023.022.321.30.078
 Weight status   0.062   0.016*
 Thin3.96.58.2 3.73.54.4 
 Normal49.561.356.4 49.454.766.7 
 Overweight29.416.121.8 30.926.720.1 
 Obese17.216.113.6 16.015.18.8 
Accelerometry          
 MVPA (min)111.1105.595.90.9290.775.080.20.088
Socio-demographics          
 Age (yrs)13.013.613.30.05712.913.313.20.371
 No. of people in house4.544.514.520.994.574.134.490.112
 Level of Deprivation   0.199   0.084
 1 (least deprived)17.214.522.1 23.517.026.1 
 221.129.026.0 13.617.018.0 
 321.730.618.3 16.017.021.1 
 415.016.113.5 16.023.915.5 
 5 (most deprived)25.09.720.2 30.925.019.3 
 Area   0.903   0.989
 Urban83.485.784.6 82.782.082.0 
 Rural16.614.315.4 17.318.018.0 
 Ethnicity   0.007   0.273
 Maori18.814.321.2 25.918.013.0 
 Pacific 12.2 1.6 6.7 8.67.99.3 
 Asian10.514.3 23.1  7.413.510.6 
 NZ Euro58.669.849.0 58.060.767.1 
Dietary intake          
 Extra foods (serve/week)10.710.99.60.23210.9a9.58.8a0.04
 Fruit (serve/day)2.32.12.20.5352.32.22.40.472
 Vegetable (serve/day)2.32.52.20.2322.12.22.40.054

For the boys, only ethnicity was significantly associated with cluster membership (Table 3). The Social studious cluster was over-represented by Asian adolescents. The Techno-active cluster was over-represented and the Quiet movers cluster was under-represented by Pacific participants.

Girls' cluster membership was associated with weight status and serves per week of extra foods. The Super studious cluster was over-represented by normal weight participants. In addition, a greater number of Social sporty members are reported as overweight or obese, and report the highest intake of non-core ‘extra foods’.

Compared to the time use residual data, the correlate classification errors were much higher. A higher classification error suggests the correlates were less accurate than the time use residual data at discerning cluster membership. The ordinary Decision tree misclassification error was 48.3 and 65.9% (training data; validation data) for the boys, and 47.7 and 50.0% for the girls. Tree Boost analyses did not improve the predictive power. The two highest-ranked correlates were age and serves/week of extra foods for the boys, and serves/week of extra foods and accelerometer values for the girls.

Discussion

This study is the first to investigate and identify time use clusters among NZ adolescents. Three time use clusters were discovered for both NZ boys and girls. The boys’ clusters were associated with ethnicity. The girls’ time use cluster membership was associated with weight status and serves/week of extra foods.

On the whole, the cluster characteristics and patterns of co-occurring behaviours in this study can be considered sex-specific, but a few gender similarities were evident. Both boys and girls were identified as having a primarily cognitive-based cluster: the girls’Super studious and the boys’Social studious clusters. Of the few previous time use cluster studies that conducted sex-specific analyses, one identified a study-based cluster for both sexes in a sample of UK adolescents.38 Interestingly, the boys in this study participated in more reading and study each day than did the girls (68.1 vs. 54.9 min/day), a pattern not often reported in the literature.39 Relative to their peers in other clusters, the health of the members of these cognitive-based clusters may benefit, as relatively higher levels of participation in cognitive activities could potentially delay the onset of neurological decline such as Alzheimer's later in life.40

Normal-weight girls were over-represented in the Super studious cluster, and the cluster members reported the lowest intake of non-core ‘extra foods’. On the other hand, the Social sporty girls consume the most non-core ‘extra foods’ each week which may be related to the higher rate of overweight and obesity.41 Despite the high levels of reported sports participation, it appears possible the Social sporty girls may not incur a weight-related health advantage over their Super studious peers.

The Techno-active cluster type is the one adolescent cluster type that is overwhelmingly reported to be associated with boys.17 It is characterised by the co-occurring behaviours of high physical activity and high screen time participation. The Techno-active boys in this study participated in more team sports, watched more TV and slept more than their peers. Cluster analysis literature supports the Techno-active cluster type, but the relationships with socio-demographic variables and other correlates are inconsistent. A recent systematic review of adolescent time use clustering studies reported consistent relationships between the Techno-active cluster type and boys, positive physical activity orientation and negative school orientation, whereas inconsistent relationships were reported for ethnicity, psychosocial health and antisocial behaviours.17 These relationship inconsistencies may be due to between-study differences in sample characteristics (e.g. ethnic representation), age range, time use measurement tools and correlate definitions. The Techno-active cluster type is similar to the Active Couch Potato phenomena among adults, who report high sedentary time (e.g. TV watching) and meet physical activity guidelines, yet are still at increased risk of poor health outcomes including high waist circumference and cholesterol.42 Interestingly, TV viewing has not been reported to increase cardiometabolic risk among adolescents,43 but it is plausible the cardiometabolic effects may culminate and be more apparent in adulthood. The Techno-active cluster was associated with a high proportion of Pacific adolescents. Pacific adolescents are reported in the literature to be among the more physically active ethnic groups in NZ,44 which is reflected in the reported higher levels of team sport.

On average, the members of the Quiet movers clusters participated in almost double the amount of quiet time (e.g. listening to music, sitting/talking quietly and lying awake) each day compared to their peers. The same cluster was under-represented by Pacific adolescent boys. Culture-based explanations for this finding are purely speculative, but it is possible the highly social, family-centred structure associated with Pacific culture 45 may reduce the opportunities for individual quiet time behaviours.

Previous studies have reported associations between girls’ screen-based clusters and increased weight status,46,47 which was not observed for the Screenie tasker girls in the current study. However, it is important to note that the Screenie taskers were characterised by both high levels of screen time and high levels of chores and work, a combination not identified in previous cluster-analysis studies. It seems possible that the different cluster-weight status relationships between studies may be explained by between-study sample differences, methodological differences resulting in identification of different behaviour clusters. The fact that previous cluster analyses have identified an association, while many simple correlate analyses find no association (skinfolds)43 or, at best, small and weak associations with adolescent weight status48 and TV viewing; supports the use of multi-dimensional analyses.

Implications

This study has demonstrated there are distinct gender-specific time use clusters among this sample of NZ adolescents. The results of this study reinforce the findings from previous time use cluster studies, suggest that some time use clusters members may be at greater health risk, and offer possibilities related to targeted intervention design that may help improve NZ adolescent health and well-being.

Targeted interventions are developed for a subgroup of a population that takes into account characteristics shared by the subgroup's members.49 To date, targeted interventions have been largely based on shared psychological variables such as readiness to change, and less so on time use patterns. Interventions can also be targeted at other health risk behaviours such as diet, smoking and drinking alcohol, which commonly cluster with certain time use behaviours.50–52 Information regarding cluster patterns and profiles could be used at various stages of an intervention; the participant selection, intervention design or delivery method. Cluster-specific time use interventions could incorporate Pacific cultural sensitivities to target the Pacific members of the Techno-active cluster. For example, materials could be written in Pacific languages or illustrations could depict Pacific adolescents. All of these ‘specificities’ have been previously reported to improved access, involvement and outcomes in health services and health-change interventions among Pacific communities.2 The Social sporty girls were identified as potentially being at greater health-risk (due to increased weight status and unfavourable dietary patterns); this group may benefit from a healthy eating intervention that could be delivered via sporting clubs and associations. Given the clusters’ affinity for sporting behaviours, healthy eating patterns could be encouraged in the context of ‘boosting sporting performance’ and ‘having fuel to fully enjoy’ sports.

Cluster analysis of time use patterns of NZ adolescents is in its infancy. Future research is essential to explore the potential of cluster analysis and should explore: better and more wide-ranging measures of health such as cardiometabolic markers and psychological measures; school versus non-school day patterns; longitudinal cluster stability and dynamic cluster trajectories; and clustering of multiple health-related behaviours. In other fields of medicine and health, it has revealed powerful patterns of co-occurring behaviours and health effects, such as the Mediterranean diet.53 Clustering has been used previously to predict the use of health services54 and the response to health-related interventions.55

Strengths and limitations

Components of the study methodology contribute to its strength, including the substantial sample drawn from a large national survey and the use of a 24-hour time use measurement tool (MARCA), which eliminates the need to pre-select variables and reduces the risk of over- and under-reporting. In an attempt to avoid several commonly reported cluster analysis issues,14 three components of cluster analysis were deliberately included: sex-specific and age-adjusted analyses; aggregated inputs reducing potential analytical noise; and the use of a cluster algorithm developed to handle large data sets which required no a priori selection of cluster number.

The 2008/2009 National Survey of Children and Young People's Physical Activity and Dietary Behaviours in New Zealand was cross-sectional in design, thus no causation can be inferred from the reported cluster-correlate relationships. As participation in this survey was voluntary, it is possible that non-responder bias was introduced. As no data were collected from those who declined to participate, it is impossible to make comparisons. The selection of the subsample based on school-day data may have introduced some bias, thus reducing the generalisability of the results, yet this is unlikely as no socio-demographic differences were detected.

Conclusion

The results of this study suggest that cluster analysis has the ability to discover meaningful and potentially useful time use clusters among NZ adolescents. The results of time use clustering studies such as this one could be used to predict adolescent health outcomes and the use of health services, and develop targeted health-related time use interventions to improve adolescent health and well-being.

Acknowledgements

The authors thank the Health Sponsorship Council for providing access to these survey data, and John Petkov for his extensive statistical knowledge and advice regarding data analysis.

The 2008/2009 National Survey of Children and Young People's Physical Activity and Dietary Behaviours in New Zealand study was funded by Sport and Recreation New Zealand (SPARC) in partnership with the Ministry of Health and the Ministry of Education and with support from the Ministry of Youth Development.

Ancillary