Journal of Adolescence 37 (2014) 1319e1328
Contents lists available at ScienceDirect
Journal of Adolescence journal homepage: www.elsevier.com/locate/jado
What do temporal profiles tell us about adolescent alcohol use? Results from a large sample in the United Kingdom Michael T. McKay a, *, James R. Andretta b, Jennifer Magee c, Frank C. Worrell d a
Center for Public Health, Liverpool John Moores University, United Kingdom Child Guidance Clinic, Superior Court of the District of Columbia, USA c Faculty of Social Sciences, University of Ulster, United Kingdom d Cognition and Development, University of California, Berkeley, USA b
a b s t r a c t Keywords: Alcohol use Temporal profiles Zimbardo Time Perspective Inventory
The psychological construct broadly known as time perspective is potentially useful in understanding a range of adolescent behaviours, including alcohol use. However, the utility of the construct has been hindered by measurement and conceptual problems. To date the vast majority of studies have assessed the relationship between time perspective and other measures in a variable-focussed (correlational) rather than a person-centred way. The present series of studies used a person-centred approach to assess the relationship between temporal profiles and alcohol use in a large sample (n ¼ 1620) of adolescents from High Schools in Northern Ireland. Although a ‘Balanced’ time perspective has been suggested as optimal, the present study suggests that having a ‘Future’ temporal profile is associated with less problematic use of alcohol, while having a ‘Past Negative’ or ‘Hedonist’ profile is associated with more problematic consumption. Results are discussed in the context of the time perspective and alcohol use literatures. © 2014 The Foundation for Professionals in Services for Adolescents. Published by Elsevier Ltd. All rights reserved.
Introduction Time perspective is a useful psychological construct in attempting to understand adolescent health behaviours (e.g., Barnett et al., 2013; Rappange, Brouwer, & Van Exel, 2009), identity formation (e.g., Luyckx, Goossens, & Soenens, 2006), and educational achievement (e.g., Shell & Husman, 2001). One of the most widely used time perspective measures is the Zimbardo Time Perspective Inventory (ZTPI; Zimbardo & Boyd, 1999), which has five subscales that assess attitudes across the past, present and future (past positive; past negative; present hedonistic; present fatalistic; and future). The relationship between ZTPI scores and health-related behaviours has been studied in both adolescents and young adults, including studies on suicidal ideation (Laghi, Baiocco, D'Alessio, & Gurrieri, 2009), binge eating and binge drinking (Laghi, Liga, Baumgartner, & Baiocco, 2012), substance use (Barnett et al., 2013), and alcohol consumption (Beenstock, Adams, & White, 2011), where higher scores on past positive and future time perspectives are associated with wellbeing, with the reverse true for past negative, present hedonistic and present fatalistic time perspectives. Moreover, it was recently reported that in College students a past-negative time perspective mediated the relationship between psychological health and alcohol-use behaviours (Linden, Lau-Barraco, & Hollis, 2014). Accordingly, higher scores on past-negative were associated with a more positive relationship between poorer mental health symptoms and alcohol-related problems. * Corresponding author. Tel.: þ44 7875 778186. E-mail addresses: [email protected], [email protected] (M.T. McKay). http://dx.doi.org/10.1016/j.adolescence.2014.09.008 0140-1971/© 2014 The Foundation for Professionals in Services for Adolescents. Published by Elsevier Ltd. All rights reserved.
1320
M.T. McKay et al. / Journal of Adolescence 37 (2014) 1319e1328
Beyond health, and in a study of the relationship between attachment measures (to parents and peers) and time perspective in adolescents, Laghi, Baiocco et al. (2009) reported that adolescents who reported a secure attachment to parents had higher scores on positive past, hedonistic present, and future, while adolescents with high attachment to parent and peer and adolescents with high parent but low peer attachment had the highest scores on competence and autonomy (Laghi, D'Alessio, Pallini, & Baiocco, 2009). In a further study, Laghi, Baiocco, Liga, Guarino, and Baumgartner (2013) reported that adolescents with a more integrated identity status were more likely to be future orientated and to have a positive view of the past, whereas “diffused” adolescents reported greater negative experiences of the past, a lower future orientation and a greater inclination to fatalism (Laghi et al., 2013). Several recent studies have suggested that adolescents may have time attitude profiles akin to personality constructs that predict psychological functioning (e.g., Alansari, Worrell, Rubie-Davies, & Webber, 2013; Andretta, Worrell, Mello, Dixson, & Baik, 2013; Buhl & Lindner, 2009). Others have demonstrated that ZTPI scores can be used to create time attitude profiles (e.g., Boniwell, Osin, Linley, & Ivanchenko, 2010) that predict psychological functioning in college students (e.g., Qin et al., 2012). However, some research has questioned the validity of ZTPI scores in adolescent samples (e.g., Worrell & Mello, 2007). Thus, there were three goals in the current series of studies with adolescents: (a) to conduct a rigorous examination of the psychometric properties of ZTPI scores; (b) to examine if ZTPI scores would yield interpretable time attitude profiles; and (c) to examine if ZTPI profiles were related to alcohol use. The relationship between time measures and alcohol use has been studied in a range of population types including dependent drinkers (Smart, 1968), those in treatment services (Klingeman, 2001), university undergraduates (Beenstock et al., 2011), and adolescents (Robbins & Bryan, 2004; Wills, Sandy & Yaeger, 2001) and have all reported that more problematic use of alcohol is associated with a foreshortened future time perspective or greater present focus. Results also supported the association between measures of present and future time perspective and any alcohol consumption (Keough, Zimbardo, & Boyd, 1999) as well as total alcohol consumption (Henson, Carey, Carey, & Maisto, 2006; Levy & Earleywine, 2004) in young adults and undergraduates. Recent longitudinal evidence has suggested a more complex relationship, with increased future time perspective protecting against the use of cigarettes and a range of drugs, but not alcohol use (Barnett et al., 2013). As noted previously, concerns about the structural validity of ZTPI scores have been raised on scores in both adolescent and adult samples in several countries (e.g., Apostolidis & Fieulaine, 2004; Carelli, Wiberg, & Wiberg, 2011; Diaz-Morales, 2006; Worrell & Mello, 2007). Indeed, in a recent study, Sircova et al. (2014) reported that the best fit that they could find in support of the structural validity of ZTPI scores across 24 countries was based on 36 of the 56 items; moreover, fit indices were still suboptimal (CFI ¼ .86, RMSEA ¼ .057). These types of results cast doubt on the conclusions based on ZTPI scores. A second concern relates to the nature of the analyses previously conducted, the majority of which have been variablefocused (e.g., correlational in nature). For example, Henson et al. (2006) reported that present hedonism was associated with pleasurable health risk behaviours (e.g., alcohol use, sex) whereas present fatalism was associated with health-damaging risk behaviours (e.g., daily smoking, no seat belt use). However, because individuals hold all five temporal domains simultaneously (Boniwell & Zimbardo, 2004; Shipp, Edwards, & Schurer-Lambert, 2009), it may be more conceptually useful to assess time perspective as a person-centered construct, simultaneously accounting for scores on all five temporal domains. Accordingly researchers have begun to explore the effect of individuals' temporal profiles obtained from the use of cluster analyses and latent profile analyses (e.g., Andretta et al., 2013; Boniwell et al., 2010; Buhl & Linder, 2009). Person-centered analyses based on profiles allow researchers to examine the effects of time profile exemplars or personality types, if you will, on outcomes (York & John, 1992), and these types of analyses can provide additional information on how time perspective relates to health compromising and health promoting outcomes. Zimbardo and Boyd (1999) theorized a balanced time perspective, characterized by moderate to high scores on past positive, present hedonistic, and future temporal domains, and low scores on past negative and present fatalistic domains, was ideal. In a recent study using the ZTPI, person-centered analyses revealed relationships that were not evident in variablecentered analyses. Boniwell et al. (2010) identified four different ZTPI profiles that generalized across two samples: Future-Oriented, Present-Oriented, Balanced, and Negative. They reported that Future-Oriented scores had primarily low correlations with measures of wellbeing whereas individuals with a Balanced profile determined in part by above average future scores had substantially lower negative affect scores than individuals with other profiles. Person-centered analyses have the added value of being able to test whether or not the balanced time perspective is in fact optimal (Boniwell & Zimbardo, 2004; Zimbardo & Boyd, 1999). We conducted the present series of studies with several hypotheses in mind. We hypothesized that analyses would yield ZTPI subscale scores that were reliable and structurally valid. Further, we hypothesized that ZTPI scores would yield profiles similar to the ones found by Boniwell et al. (2010) and Qin et al. (2012). Third, we examined the relationship of alcohol use to ZTPI profiles. We hypothesized that correlations between alcohol use scores and ZTPI scores would be modest, with the exception of a moderate, positive correlation with present hedonistic scores. Although Zimbardo and Boyd (1999) postulated that a balanced time perspective is ideal, the extant literature on alcohol use (Beenstock et al., 2011; Robbins & Bryan, 2004) indicates that future orientation is related to less problematic alcohol use. Thus, we hypothesized that individuals with a future-oriented time perspective profile would report less problematic alcohol use. As alcohol use has been found to be associated with demographic variables, this study included indicators of gender (e.g., Flory, Milich, Lynam, Leukefeld, & Clayton, 2003; Loretto, 1994; Northern Ireland Statistics and Research Agency [NISRA], 2008) and grade level as a proxy for age (NISRA, 2008; Percy & Iwaniec, 2010). From a position where historically girls in
M.T. McKay et al. / Journal of Adolescence 37 (2014) 1319e1328
1321
NI appeared to be less likely than boys to drink problematically (e.g., Loretto, 1994), more recent evidence has suggested that a convergence has taken place so that girls are as likely as boys to drink problematically (Eisenbach-Stangl & Thom, 2009; Health Promotion Agency, 2005; McKay, Percy, Goudie, Sumnall, & Cole, 2012; Northern Ireland Statistics and Research Agency, 2008). In line with past and contemporaneous research, we hypothesized that more problematic alcohol behaviours would be associated with being older, but that gender would not be a significant predictor of drinking behaviours. Study 1 Given previous results of studies using the ZTPI with adolescents (Worrell & Mello, 2007) the first task was to examine if the structural validity of ZTPI scores would be supported. Moreover, we decided that if the ZTPI model based on 56 items was not supported, we would test an alternative model, the SZTPI-15 (Zhang, Howell, & Bowerman, 2013). In the event of the data fitting neither model well, a viable alternative would be sought from analyses of data from the present study. Participants Participants were pupils in 10 high schools in the Eastern Health Board area of Northern Ireland. Post-primary schools in NI are either Grammar or Secondary schools. Grammar schools select students at age 11 (school year 8) on the basis of higher academic ability, while Secondary schools are more comprehensive/vocational. Schools were randomly selected from the total number of schools in the area. All schools approached agreed to participate. Schools were asked to provide between 20 and 25 pupils from each of School Grades 8e12 (ages 12e16). A total of 943 questionnaires were completed. Thirty were excluded as a result of having been partially completed or spoiled (i.e., whole sections left incomplete and/or multiple responses given to answers), resulting in a final sample of 913 (97%) participants almost evenly split between male (n ¼ 455, 49.8%) and female (n ¼ 458, 50.2%). Measures The ZTPI (Zimbardo & Boyd, 1999) is a 56-item measure consisting of five subscales, past negative (PN; 10 items), past positive (PP; 9 items), present hedonistic (PH; 15 items), present fatalistic (PF; 9 items) and future (F; 13 items). Participants respond to questions using a 5-point Likert scale (1 ¼ very uncharacteristic; 5 ¼ very characteristic). Internal consistency estimates for subscale scores based on Cronbach's a ranged from 0.74 to 0.82 (Mdn a ¼ 0.79) in the initial study. The fivefactor structure was supported using exploratory factor analyses (EFAs) and inter-subscale correlations were generally low (j.09j r j.38j) (Zimbardo & Boyd, 1999). Although some have claimed that ZTPI scores have been validated for use in adolescents (e.g., Laghi, Baiocco, et al., 2009), this is based largely on a single Italian study (D'Alessio, Guarino, De Pascalis, & Zimbardo, 2003) where adolescents formed only a small proportion of the overall sample.1 Additionally, D'Alessio et al. (2003) used a shortened version of the ZTPI (22 items) and reported reliability estimates ranging from .49 to .67 (Mdn ¼ .54). Procedure An opt-out passive consent was approved by the University of Liverpool Ethics Committee. On the day of the data collection, each participating pupil was asked to assent to participate so that students also had to indicate a willingness to participate in the study. Data were gathered under examination-like conditions. Results Results of Pearson c2 analyses showed that there was a significant difference in distribution of males and females across school grades (c2 ¼ 13.91 (df4) p < .01), with a substantially greater proportion of males in School Grade 8, and a substantially greater proportion of females in School Grade 12. Model fit was first assessed using confirmatory factor analyses (CFAs) in n & Muthe n, 2010) using the MLM estimator. The MLM maximum likelihood parameter estimates with Mplus 6 (Muthe standard errors and the Satorra-Bentler mean-adjusted chi-square (c2) test statistic are robust to non-normality. A five-factor model for the ZTPI was assessed. The indices used to test model fit were c2, the comparative fit index (CFI), the TuckereLewis index (TLI), the root mean square error of approximation (RMSEA), and the standardized root mean square residual (SRMR). The values recommended by Marsh, Hau, and Wen (2004)dnamely .92 for CFI and TLI, .05 for RMSEA, and .08 for SRMRdwere employed in the present study as a guide to model fit. Results showed that several fit indices for scores on the 56-ZTPI were poor, c2(1474) ¼ 4523.03, p < .001, CFI ¼ 0.75, TLI ¼ 0.70, SRMR ¼ 0.07, and RMSEA ¼ 0.05 (0.04e0.05), and some of these also fell short of acceptable cores on the SZTPI-15, c2(80) ¼ 269.65, p < .001, CFI ¼ 0.90, TLI ¼ 0.88, SRMR ¼ 0.05, and RMSEA ¼ 0.05 (0.4e0.6).
1 Authors did not report the exact number of adolescents, however, of the total sample (n ¼ 1507; age range 16e89 years, M ¼ 34.7 years, SD ¼ 10.55 years), a total of n ¼ 400 were within the ’16e27’ age group.
1322
M.T. McKay et al. / Journal of Adolescence 37 (2014) 1319e1328
As cluster analysis requires the use of psychometrically sound scores, we took the view that the creation of temporal profiles based on these models would not be ideal. Thus, we conducted an exploratory factor analysis on the ZTPI data with the aim of developing a better fitting short form of the ZTPI which, would in turn, form the basis for profiling. Initially, we had intended using the three items with the highest coefficients for each of the five factors from the EFA of the 56-item scale. However, the top three loading items on the Past Negative factor also had salient coefficients on other factors. Thus, we developed a short form using items that had the highest coefficients on their home factors but did not have salient coefficients (i.e., .50 and the fit indices indicated adequate to superior model fit: c2(80) ¼ 192.38, p .5 SDs above the mean), very
Fig. 1. Bars represent T-scores, where 50 is the mean and 10 is the standard deviation. Interpretation of profiles and profile labelling are based on each subscale scores' distance and direction from the subscale mean, and in relation to other subscale scores.
1324
M.T. McKay et al. / Journal of Adolescence 37 (2014) 1319e1328
low scores on F (> 1 SD below the mean), average scores on PP and PN, and moderately high scores on PF. Those in the Balanced profile were characterised by moderately high scores on PP, PH, and F (z.5 SDs above mean), and lower scores on PN and PF. Past Negatives were characterized by very low scores on PP, moderately high scores on PN (> .4 SDs above the mean), and average scores on the other three subscales. Finally, Futures were characterized by low scores (> .5 SDs below the mean) on PH, PF, and PN, average scores on PP, and moderately high scores on F. Results of Pearson c2 analyses for the whole sample showed that there was no significant difference in distribution of males and females across school grades (c2(4) ¼ 8.68 p ¼ .07). The school-level distinction in terms of academic ability between Grammar and Secondary school attendees (assessed at age 11/12) is also reflected in socio-economic status as indicated by percentage of pupils entitled to a free school meal. The mean entitlements in the present sample were: Grammar ¼ 5.56% entitlement; Secondary ¼ 28.45% entitlement. AAIS groups The following were the numbers and percentages of participants in each of the five AAIS groups: 575 abstainers (35.5%); 125 who rarely drink (7.7%); 678 who drink but do not experience problems (41.9%); 224 alcohol misusers (13.8%); and 18 alcoholic-like (1.1%). In previous studies of adolescents (e.g., Authors, blinded), we had found the normal and alcoholic-like groups have relatively few members and we have used three groups as follows: abstainers (score ¼ 0); moderate or nonproblematic drinkers (score ¼ 1e41, including normal and non-problematic drinker groups); and problematic drinkers (score ¼ 42e79, including misusers and alcoholic-like groups). We adopted this classification in the present study, yielding the following revised AAIS groups: Abstainers ¼ 575 (35.5%), Moderate Drinkers ¼ 803 (49.6%), and Problematic Drinkers ¼ 242 (14.9%). Table 1 displays percentages of participants in time attitude clusters by gender, grade in school, type of school, and AAIS groups, and the results of c2 analyses for groups by time profiles. Differences in time profiles by grade level, c2(12) ¼ 40.18, p < .001, and type of school attended, c2(3) ¼ 44.22, p < .001, were significant, but effect sizes were small as shown in the table. There were greater proportions of Grade 8 students in the Balanced and Past Negative profiles, with higher proportions of Grade 12 students in the Future profile. Greater proportions of Grammar schools attendees were in the Hedonist and Future profiles. Differences in time profiles by alcohol groups were also significant, c2 (6) ¼ 170.36, p < .001, with higher proportions of Balanced and Future participants in the abstainer group, and a higher proportion of Hedonists in the Problematic group. Males and females did not differ significantly in their representation across time perspective profiles, c2 (3) ¼ 1.31, p ¼ .73. Correlations between time attitude scores and AAIS total score were as follows (all significant at p < .01): PN (r ¼ .18), PP (r ¼ .12), PF (r ¼ .28), PH (r ¼ .40), and F (r ¼ .35). Predicting AAIS classification by demographic and time attitude groups To examine predictive covariates of AAIS group membership, a multinomial logistic regression (MLR) model was estimated in SPSS v.20. As the data were clustered at the school level (i.e., pupils within schools), it was necessary to correct the regression standard errors to take account of the non-independence of study respondents. To do this, a set of dummy variables was fitted for the clustering at school level. Alternative methods for modelling the lack of independence of observationsdfor example, including a two-level multilevel model or using a robust estimator with standard errors corrected for the non-independence of observationsdwere not implemented as it has been suggested that a sample of 50 or less clusters (schools) leads to biased estimates of the second-level standard errors (Maas & Hox, 2005; Moineddin, Matheson, & Glazier, 2007). In the MLR model, School Grade 8 was set as the default as research suggests that those in Grade 8 are the least likely to drink. Moreover, Futures were set as the default drinking group, as it was hypothesized that they would be least likely to drink based in previous research.
Table 1 Demographic Representation in Time Attitude Profiles (% in parentheses). Demographic Gender Male Female School grade Grade 8 Grade 9 Grade 10 Grade 11 Grade 12 Type of school Grammar Secondary Alcohol groups Abstainer Moderate Problematic Note. N ¼ 1620.
Hedonist
Balanced
Past negative
Future
165 (18.8) 145 (19.5)
294 (33.5) 252 (34.0)
147 (16.7) 109 (14.7)
272 (31.0) 236 (31.8)
V .03
.09 33 61 83 80 53
(11.8) (19.6) (23.6) (19.9) (19.3)
131 104 98 127 86
(46.8) (33.4) (27.9) (31.5) (31.3)
48 55 54 62 37
(17.1) (17.7) (15.4) (15.4) (13.5)
68 91 116 134 99
(24.3) (29.3) (33.0) (33.3) (36.0) .17
191 (20.9) 119 (16.9)
267 (29.2) 279 (39.6)
121 (13.2) 135 (19.1)
336 (36.7) 172 (24.4)
49 (8.6) 159 (19.8) 102 (42.1)
191 (33.2) 276 (34.4) 79 (32.6)
83 (14.4) 134 (16.7) 39 (16.1)
252 (43.8) 234 (29.1) 22 (9.2)
.23
M.T. McKay et al. / Journal of Adolescence 37 (2014) 1319e1328
1325
Results by gender. Results of the MLR model are displayed in Table 2. As can be seen, females and males did not differ in their likelihood for being moderate drinkers versus abstainers. However, males did have significantly lower risk for being problematic drinkers versus abstainers and problematic drinkers versus moderate drinkers. In both cases, the effect size was modest, with only a .4% decrease in odds. Results by grade level. As can be seen in Table 2, compared to abstainers, moderate drinkers were significantly more likely to be in a higher school grade (except when comparing grade 8 and 9). The odds ratios provide a more nuanced view of these results: 10th graders were twice as likely to be moderate drinkers than abstainers, 11th graders were three times as likely to be moderate drinkers, and 12th graders were six times more likely to be moderate drinkers than abstainers. This pattern of results by grade level is repeated for problematic drinkers relative to abstainers and for problematic drinkers relative to moderate drinkers. However, in these two cases, the odds ratios are substantially higher. The risk of being a problematic drinker versus an abstainer is almost three times greater for 9th than 8th graders, and increases to 74 times higher for 12th graders. The odds ratios for being a problematic versus a moderate drinker also increases substantially across the grade levels from twice as likely for 9th graders to more than 11 times greater for 12th graders. Results for time profiles. With regard to time profiles, moderate drinkers (compared to abstainers) were significantly more likely to belong to all other time profiles than Futures. The Balanced and Past Negative profiles had similar odds ratios, with both groups twice as likely to be moderate drinkers, but the risk increased to four times for the Present Hedonists. As with the grade levels, the general pattern of results was the same for problematic drinkers relative to abstainers and problematic drinkers relative to moderate drinkers. As can be seen in Table 2, the risks are higher and similar for the Balanced and Past Negatives and substantially greater for the Present Hedonists. Discussion The present study sought to add to the increasing body of literature examining the relationship between time perspective and alcohol use. We began with validating a shortened version of ZTPI scores. Then, in addition to looking at the correlations between ZTPI scores and alcohol use, we examined adolescent alcohol use against temporal profiles. This approach is methodologically important as an individual holds more than one time attitude, at any one time. Thus, although the assessment of health behaviours against scores on any one ZTPI factor might be both interesting and useful, the personcentered approach is arguably more complete. The ZTPI (Zimbardo & Boyd, 1999) has become one of the most frequently used measures of time constructs in the literature. However, the increasing use of the instrument has parallelled a growing set of concerns about the psychometric properties of ZTPI scores (Shipp et al., 2009; Worrell & Mello, 2007), many of which remain (Sircova et al., 2014; Worrell, Table 2 Predictors associated with revised AAIS group membership: MLR (shown are OR þ95% CI). Reference category is abstainers Moderate drinkers
Gender Female Male School grade Grade 8 Grade 9 Grade 10 Grade 11 Grade 12 Time profile Future Balanced Past Negative Hedonist
Reference category is moderate Problematic drinkers
Problematic drinkers
B (SE B)
OR (95% CI)
p value
B (SE B)
OR (95% CI)
p value
B (SE B)
OR (95% CI)
p value
e .06 (.15)
e 1.06 (.80, 1.41)
e .68
e .42 (.21)
e 0.65 (.44, .99)
e .04
e .48 (.18)
e .62 (.43, .88)
e .008
e .31 (.18) .75 (.19) 1.15 (.19) 1.86 (.24)
e 1.36 (.95, .94) 2.12 (1.48, 3.05) 3.17 (2.18, 4.60) 6.40 (4.00, 10.23)
e .096
e 1.02 (.48) 2.26 (.44) 3.18 (.44) 4.31 (.47)
e 2.76 (1.08, 7.03) 9.58 (4.02, 22.81) 23.95 (10.19, 56.29) 74.53 (29.97, 185.34)
e .033
e .71 (.47) 1.51 (.43) 2.02 (.42) 2.46 (.43)
e 2.04 (.81, 5.12) 4.51 (1.93, 10.53) 7.56 (3.31, 17.30) 11.65 (4.99, 27.17)
e .131
e .69 (.15) .72 (.18) 1.38 (.20)
e 2.00 (1.50, 2.67) 2.05 (1.43, 2.93) 3.97 (2.68, 5.89)
e
Contents lists available at ScienceDirect
Journal of Adolescence journal homepage: www.elsevier.com/locate/jado
What do temporal profiles tell us about adolescent alcohol use? Results from a large sample in the United Kingdom Michael T. McKay a, *, James R. Andretta b, Jennifer Magee c, Frank C. Worrell d a
Center for Public Health, Liverpool John Moores University, United Kingdom Child Guidance Clinic, Superior Court of the District of Columbia, USA c Faculty of Social Sciences, University of Ulster, United Kingdom d Cognition and Development, University of California, Berkeley, USA b
a b s t r a c t Keywords: Alcohol use Temporal profiles Zimbardo Time Perspective Inventory
The psychological construct broadly known as time perspective is potentially useful in understanding a range of adolescent behaviours, including alcohol use. However, the utility of the construct has been hindered by measurement and conceptual problems. To date the vast majority of studies have assessed the relationship between time perspective and other measures in a variable-focussed (correlational) rather than a person-centred way. The present series of studies used a person-centred approach to assess the relationship between temporal profiles and alcohol use in a large sample (n ¼ 1620) of adolescents from High Schools in Northern Ireland. Although a ‘Balanced’ time perspective has been suggested as optimal, the present study suggests that having a ‘Future’ temporal profile is associated with less problematic use of alcohol, while having a ‘Past Negative’ or ‘Hedonist’ profile is associated with more problematic consumption. Results are discussed in the context of the time perspective and alcohol use literatures. © 2014 The Foundation for Professionals in Services for Adolescents. Published by Elsevier Ltd. All rights reserved.
Introduction Time perspective is a useful psychological construct in attempting to understand adolescent health behaviours (e.g., Barnett et al., 2013; Rappange, Brouwer, & Van Exel, 2009), identity formation (e.g., Luyckx, Goossens, & Soenens, 2006), and educational achievement (e.g., Shell & Husman, 2001). One of the most widely used time perspective measures is the Zimbardo Time Perspective Inventory (ZTPI; Zimbardo & Boyd, 1999), which has five subscales that assess attitudes across the past, present and future (past positive; past negative; present hedonistic; present fatalistic; and future). The relationship between ZTPI scores and health-related behaviours has been studied in both adolescents and young adults, including studies on suicidal ideation (Laghi, Baiocco, D'Alessio, & Gurrieri, 2009), binge eating and binge drinking (Laghi, Liga, Baumgartner, & Baiocco, 2012), substance use (Barnett et al., 2013), and alcohol consumption (Beenstock, Adams, & White, 2011), where higher scores on past positive and future time perspectives are associated with wellbeing, with the reverse true for past negative, present hedonistic and present fatalistic time perspectives. Moreover, it was recently reported that in College students a past-negative time perspective mediated the relationship between psychological health and alcohol-use behaviours (Linden, Lau-Barraco, & Hollis, 2014). Accordingly, higher scores on past-negative were associated with a more positive relationship between poorer mental health symptoms and alcohol-related problems. * Corresponding author. Tel.: þ44 7875 778186. E-mail addresses: [email protected], [email protected] (M.T. McKay). http://dx.doi.org/10.1016/j.adolescence.2014.09.008 0140-1971/© 2014 The Foundation for Professionals in Services for Adolescents. Published by Elsevier Ltd. All rights reserved.
1320
M.T. McKay et al. / Journal of Adolescence 37 (2014) 1319e1328
Beyond health, and in a study of the relationship between attachment measures (to parents and peers) and time perspective in adolescents, Laghi, Baiocco et al. (2009) reported that adolescents who reported a secure attachment to parents had higher scores on positive past, hedonistic present, and future, while adolescents with high attachment to parent and peer and adolescents with high parent but low peer attachment had the highest scores on competence and autonomy (Laghi, D'Alessio, Pallini, & Baiocco, 2009). In a further study, Laghi, Baiocco, Liga, Guarino, and Baumgartner (2013) reported that adolescents with a more integrated identity status were more likely to be future orientated and to have a positive view of the past, whereas “diffused” adolescents reported greater negative experiences of the past, a lower future orientation and a greater inclination to fatalism (Laghi et al., 2013). Several recent studies have suggested that adolescents may have time attitude profiles akin to personality constructs that predict psychological functioning (e.g., Alansari, Worrell, Rubie-Davies, & Webber, 2013; Andretta, Worrell, Mello, Dixson, & Baik, 2013; Buhl & Lindner, 2009). Others have demonstrated that ZTPI scores can be used to create time attitude profiles (e.g., Boniwell, Osin, Linley, & Ivanchenko, 2010) that predict psychological functioning in college students (e.g., Qin et al., 2012). However, some research has questioned the validity of ZTPI scores in adolescent samples (e.g., Worrell & Mello, 2007). Thus, there were three goals in the current series of studies with adolescents: (a) to conduct a rigorous examination of the psychometric properties of ZTPI scores; (b) to examine if ZTPI scores would yield interpretable time attitude profiles; and (c) to examine if ZTPI profiles were related to alcohol use. The relationship between time measures and alcohol use has been studied in a range of population types including dependent drinkers (Smart, 1968), those in treatment services (Klingeman, 2001), university undergraduates (Beenstock et al., 2011), and adolescents (Robbins & Bryan, 2004; Wills, Sandy & Yaeger, 2001) and have all reported that more problematic use of alcohol is associated with a foreshortened future time perspective or greater present focus. Results also supported the association between measures of present and future time perspective and any alcohol consumption (Keough, Zimbardo, & Boyd, 1999) as well as total alcohol consumption (Henson, Carey, Carey, & Maisto, 2006; Levy & Earleywine, 2004) in young adults and undergraduates. Recent longitudinal evidence has suggested a more complex relationship, with increased future time perspective protecting against the use of cigarettes and a range of drugs, but not alcohol use (Barnett et al., 2013). As noted previously, concerns about the structural validity of ZTPI scores have been raised on scores in both adolescent and adult samples in several countries (e.g., Apostolidis & Fieulaine, 2004; Carelli, Wiberg, & Wiberg, 2011; Diaz-Morales, 2006; Worrell & Mello, 2007). Indeed, in a recent study, Sircova et al. (2014) reported that the best fit that they could find in support of the structural validity of ZTPI scores across 24 countries was based on 36 of the 56 items; moreover, fit indices were still suboptimal (CFI ¼ .86, RMSEA ¼ .057). These types of results cast doubt on the conclusions based on ZTPI scores. A second concern relates to the nature of the analyses previously conducted, the majority of which have been variablefocused (e.g., correlational in nature). For example, Henson et al. (2006) reported that present hedonism was associated with pleasurable health risk behaviours (e.g., alcohol use, sex) whereas present fatalism was associated with health-damaging risk behaviours (e.g., daily smoking, no seat belt use). However, because individuals hold all five temporal domains simultaneously (Boniwell & Zimbardo, 2004; Shipp, Edwards, & Schurer-Lambert, 2009), it may be more conceptually useful to assess time perspective as a person-centered construct, simultaneously accounting for scores on all five temporal domains. Accordingly researchers have begun to explore the effect of individuals' temporal profiles obtained from the use of cluster analyses and latent profile analyses (e.g., Andretta et al., 2013; Boniwell et al., 2010; Buhl & Linder, 2009). Person-centered analyses based on profiles allow researchers to examine the effects of time profile exemplars or personality types, if you will, on outcomes (York & John, 1992), and these types of analyses can provide additional information on how time perspective relates to health compromising and health promoting outcomes. Zimbardo and Boyd (1999) theorized a balanced time perspective, characterized by moderate to high scores on past positive, present hedonistic, and future temporal domains, and low scores on past negative and present fatalistic domains, was ideal. In a recent study using the ZTPI, person-centered analyses revealed relationships that were not evident in variablecentered analyses. Boniwell et al. (2010) identified four different ZTPI profiles that generalized across two samples: Future-Oriented, Present-Oriented, Balanced, and Negative. They reported that Future-Oriented scores had primarily low correlations with measures of wellbeing whereas individuals with a Balanced profile determined in part by above average future scores had substantially lower negative affect scores than individuals with other profiles. Person-centered analyses have the added value of being able to test whether or not the balanced time perspective is in fact optimal (Boniwell & Zimbardo, 2004; Zimbardo & Boyd, 1999). We conducted the present series of studies with several hypotheses in mind. We hypothesized that analyses would yield ZTPI subscale scores that were reliable and structurally valid. Further, we hypothesized that ZTPI scores would yield profiles similar to the ones found by Boniwell et al. (2010) and Qin et al. (2012). Third, we examined the relationship of alcohol use to ZTPI profiles. We hypothesized that correlations between alcohol use scores and ZTPI scores would be modest, with the exception of a moderate, positive correlation with present hedonistic scores. Although Zimbardo and Boyd (1999) postulated that a balanced time perspective is ideal, the extant literature on alcohol use (Beenstock et al., 2011; Robbins & Bryan, 2004) indicates that future orientation is related to less problematic alcohol use. Thus, we hypothesized that individuals with a future-oriented time perspective profile would report less problematic alcohol use. As alcohol use has been found to be associated with demographic variables, this study included indicators of gender (e.g., Flory, Milich, Lynam, Leukefeld, & Clayton, 2003; Loretto, 1994; Northern Ireland Statistics and Research Agency [NISRA], 2008) and grade level as a proxy for age (NISRA, 2008; Percy & Iwaniec, 2010). From a position where historically girls in
M.T. McKay et al. / Journal of Adolescence 37 (2014) 1319e1328
1321
NI appeared to be less likely than boys to drink problematically (e.g., Loretto, 1994), more recent evidence has suggested that a convergence has taken place so that girls are as likely as boys to drink problematically (Eisenbach-Stangl & Thom, 2009; Health Promotion Agency, 2005; McKay, Percy, Goudie, Sumnall, & Cole, 2012; Northern Ireland Statistics and Research Agency, 2008). In line with past and contemporaneous research, we hypothesized that more problematic alcohol behaviours would be associated with being older, but that gender would not be a significant predictor of drinking behaviours. Study 1 Given previous results of studies using the ZTPI with adolescents (Worrell & Mello, 2007) the first task was to examine if the structural validity of ZTPI scores would be supported. Moreover, we decided that if the ZTPI model based on 56 items was not supported, we would test an alternative model, the SZTPI-15 (Zhang, Howell, & Bowerman, 2013). In the event of the data fitting neither model well, a viable alternative would be sought from analyses of data from the present study. Participants Participants were pupils in 10 high schools in the Eastern Health Board area of Northern Ireland. Post-primary schools in NI are either Grammar or Secondary schools. Grammar schools select students at age 11 (school year 8) on the basis of higher academic ability, while Secondary schools are more comprehensive/vocational. Schools were randomly selected from the total number of schools in the area. All schools approached agreed to participate. Schools were asked to provide between 20 and 25 pupils from each of School Grades 8e12 (ages 12e16). A total of 943 questionnaires were completed. Thirty were excluded as a result of having been partially completed or spoiled (i.e., whole sections left incomplete and/or multiple responses given to answers), resulting in a final sample of 913 (97%) participants almost evenly split between male (n ¼ 455, 49.8%) and female (n ¼ 458, 50.2%). Measures The ZTPI (Zimbardo & Boyd, 1999) is a 56-item measure consisting of five subscales, past negative (PN; 10 items), past positive (PP; 9 items), present hedonistic (PH; 15 items), present fatalistic (PF; 9 items) and future (F; 13 items). Participants respond to questions using a 5-point Likert scale (1 ¼ very uncharacteristic; 5 ¼ very characteristic). Internal consistency estimates for subscale scores based on Cronbach's a ranged from 0.74 to 0.82 (Mdn a ¼ 0.79) in the initial study. The fivefactor structure was supported using exploratory factor analyses (EFAs) and inter-subscale correlations were generally low (j.09j r j.38j) (Zimbardo & Boyd, 1999). Although some have claimed that ZTPI scores have been validated for use in adolescents (e.g., Laghi, Baiocco, et al., 2009), this is based largely on a single Italian study (D'Alessio, Guarino, De Pascalis, & Zimbardo, 2003) where adolescents formed only a small proportion of the overall sample.1 Additionally, D'Alessio et al. (2003) used a shortened version of the ZTPI (22 items) and reported reliability estimates ranging from .49 to .67 (Mdn ¼ .54). Procedure An opt-out passive consent was approved by the University of Liverpool Ethics Committee. On the day of the data collection, each participating pupil was asked to assent to participate so that students also had to indicate a willingness to participate in the study. Data were gathered under examination-like conditions. Results Results of Pearson c2 analyses showed that there was a significant difference in distribution of males and females across school grades (c2 ¼ 13.91 (df4) p < .01), with a substantially greater proportion of males in School Grade 8, and a substantially greater proportion of females in School Grade 12. Model fit was first assessed using confirmatory factor analyses (CFAs) in n & Muthe n, 2010) using the MLM estimator. The MLM maximum likelihood parameter estimates with Mplus 6 (Muthe standard errors and the Satorra-Bentler mean-adjusted chi-square (c2) test statistic are robust to non-normality. A five-factor model for the ZTPI was assessed. The indices used to test model fit were c2, the comparative fit index (CFI), the TuckereLewis index (TLI), the root mean square error of approximation (RMSEA), and the standardized root mean square residual (SRMR). The values recommended by Marsh, Hau, and Wen (2004)dnamely .92 for CFI and TLI, .05 for RMSEA, and .08 for SRMRdwere employed in the present study as a guide to model fit. Results showed that several fit indices for scores on the 56-ZTPI were poor, c2(1474) ¼ 4523.03, p < .001, CFI ¼ 0.75, TLI ¼ 0.70, SRMR ¼ 0.07, and RMSEA ¼ 0.05 (0.04e0.05), and some of these also fell short of acceptable cores on the SZTPI-15, c2(80) ¼ 269.65, p < .001, CFI ¼ 0.90, TLI ¼ 0.88, SRMR ¼ 0.05, and RMSEA ¼ 0.05 (0.4e0.6).
1 Authors did not report the exact number of adolescents, however, of the total sample (n ¼ 1507; age range 16e89 years, M ¼ 34.7 years, SD ¼ 10.55 years), a total of n ¼ 400 were within the ’16e27’ age group.
1322
M.T. McKay et al. / Journal of Adolescence 37 (2014) 1319e1328
As cluster analysis requires the use of psychometrically sound scores, we took the view that the creation of temporal profiles based on these models would not be ideal. Thus, we conducted an exploratory factor analysis on the ZTPI data with the aim of developing a better fitting short form of the ZTPI which, would in turn, form the basis for profiling. Initially, we had intended using the three items with the highest coefficients for each of the five factors from the EFA of the 56-item scale. However, the top three loading items on the Past Negative factor also had salient coefficients on other factors. Thus, we developed a short form using items that had the highest coefficients on their home factors but did not have salient coefficients (i.e., .50 and the fit indices indicated adequate to superior model fit: c2(80) ¼ 192.38, p .5 SDs above the mean), very
Fig. 1. Bars represent T-scores, where 50 is the mean and 10 is the standard deviation. Interpretation of profiles and profile labelling are based on each subscale scores' distance and direction from the subscale mean, and in relation to other subscale scores.
1324
M.T. McKay et al. / Journal of Adolescence 37 (2014) 1319e1328
low scores on F (> 1 SD below the mean), average scores on PP and PN, and moderately high scores on PF. Those in the Balanced profile were characterised by moderately high scores on PP, PH, and F (z.5 SDs above mean), and lower scores on PN and PF. Past Negatives were characterized by very low scores on PP, moderately high scores on PN (> .4 SDs above the mean), and average scores on the other three subscales. Finally, Futures were characterized by low scores (> .5 SDs below the mean) on PH, PF, and PN, average scores on PP, and moderately high scores on F. Results of Pearson c2 analyses for the whole sample showed that there was no significant difference in distribution of males and females across school grades (c2(4) ¼ 8.68 p ¼ .07). The school-level distinction in terms of academic ability between Grammar and Secondary school attendees (assessed at age 11/12) is also reflected in socio-economic status as indicated by percentage of pupils entitled to a free school meal. The mean entitlements in the present sample were: Grammar ¼ 5.56% entitlement; Secondary ¼ 28.45% entitlement. AAIS groups The following were the numbers and percentages of participants in each of the five AAIS groups: 575 abstainers (35.5%); 125 who rarely drink (7.7%); 678 who drink but do not experience problems (41.9%); 224 alcohol misusers (13.8%); and 18 alcoholic-like (1.1%). In previous studies of adolescents (e.g., Authors, blinded), we had found the normal and alcoholic-like groups have relatively few members and we have used three groups as follows: abstainers (score ¼ 0); moderate or nonproblematic drinkers (score ¼ 1e41, including normal and non-problematic drinker groups); and problematic drinkers (score ¼ 42e79, including misusers and alcoholic-like groups). We adopted this classification in the present study, yielding the following revised AAIS groups: Abstainers ¼ 575 (35.5%), Moderate Drinkers ¼ 803 (49.6%), and Problematic Drinkers ¼ 242 (14.9%). Table 1 displays percentages of participants in time attitude clusters by gender, grade in school, type of school, and AAIS groups, and the results of c2 analyses for groups by time profiles. Differences in time profiles by grade level, c2(12) ¼ 40.18, p < .001, and type of school attended, c2(3) ¼ 44.22, p < .001, were significant, but effect sizes were small as shown in the table. There were greater proportions of Grade 8 students in the Balanced and Past Negative profiles, with higher proportions of Grade 12 students in the Future profile. Greater proportions of Grammar schools attendees were in the Hedonist and Future profiles. Differences in time profiles by alcohol groups were also significant, c2 (6) ¼ 170.36, p < .001, with higher proportions of Balanced and Future participants in the abstainer group, and a higher proportion of Hedonists in the Problematic group. Males and females did not differ significantly in their representation across time perspective profiles, c2 (3) ¼ 1.31, p ¼ .73. Correlations between time attitude scores and AAIS total score were as follows (all significant at p < .01): PN (r ¼ .18), PP (r ¼ .12), PF (r ¼ .28), PH (r ¼ .40), and F (r ¼ .35). Predicting AAIS classification by demographic and time attitude groups To examine predictive covariates of AAIS group membership, a multinomial logistic regression (MLR) model was estimated in SPSS v.20. As the data were clustered at the school level (i.e., pupils within schools), it was necessary to correct the regression standard errors to take account of the non-independence of study respondents. To do this, a set of dummy variables was fitted for the clustering at school level. Alternative methods for modelling the lack of independence of observationsdfor example, including a two-level multilevel model or using a robust estimator with standard errors corrected for the non-independence of observationsdwere not implemented as it has been suggested that a sample of 50 or less clusters (schools) leads to biased estimates of the second-level standard errors (Maas & Hox, 2005; Moineddin, Matheson, & Glazier, 2007). In the MLR model, School Grade 8 was set as the default as research suggests that those in Grade 8 are the least likely to drink. Moreover, Futures were set as the default drinking group, as it was hypothesized that they would be least likely to drink based in previous research.
Table 1 Demographic Representation in Time Attitude Profiles (% in parentheses). Demographic Gender Male Female School grade Grade 8 Grade 9 Grade 10 Grade 11 Grade 12 Type of school Grammar Secondary Alcohol groups Abstainer Moderate Problematic Note. N ¼ 1620.
Hedonist
Balanced
Past negative
Future
165 (18.8) 145 (19.5)
294 (33.5) 252 (34.0)
147 (16.7) 109 (14.7)
272 (31.0) 236 (31.8)
V .03
.09 33 61 83 80 53
(11.8) (19.6) (23.6) (19.9) (19.3)
131 104 98 127 86
(46.8) (33.4) (27.9) (31.5) (31.3)
48 55 54 62 37
(17.1) (17.7) (15.4) (15.4) (13.5)
68 91 116 134 99
(24.3) (29.3) (33.0) (33.3) (36.0) .17
191 (20.9) 119 (16.9)
267 (29.2) 279 (39.6)
121 (13.2) 135 (19.1)
336 (36.7) 172 (24.4)
49 (8.6) 159 (19.8) 102 (42.1)
191 (33.2) 276 (34.4) 79 (32.6)
83 (14.4) 134 (16.7) 39 (16.1)
252 (43.8) 234 (29.1) 22 (9.2)
.23
M.T. McKay et al. / Journal of Adolescence 37 (2014) 1319e1328
1325
Results by gender. Results of the MLR model are displayed in Table 2. As can be seen, females and males did not differ in their likelihood for being moderate drinkers versus abstainers. However, males did have significantly lower risk for being problematic drinkers versus abstainers and problematic drinkers versus moderate drinkers. In both cases, the effect size was modest, with only a .4% decrease in odds. Results by grade level. As can be seen in Table 2, compared to abstainers, moderate drinkers were significantly more likely to be in a higher school grade (except when comparing grade 8 and 9). The odds ratios provide a more nuanced view of these results: 10th graders were twice as likely to be moderate drinkers than abstainers, 11th graders were three times as likely to be moderate drinkers, and 12th graders were six times more likely to be moderate drinkers than abstainers. This pattern of results by grade level is repeated for problematic drinkers relative to abstainers and for problematic drinkers relative to moderate drinkers. However, in these two cases, the odds ratios are substantially higher. The risk of being a problematic drinker versus an abstainer is almost three times greater for 9th than 8th graders, and increases to 74 times higher for 12th graders. The odds ratios for being a problematic versus a moderate drinker also increases substantially across the grade levels from twice as likely for 9th graders to more than 11 times greater for 12th graders. Results for time profiles. With regard to time profiles, moderate drinkers (compared to abstainers) were significantly more likely to belong to all other time profiles than Futures. The Balanced and Past Negative profiles had similar odds ratios, with both groups twice as likely to be moderate drinkers, but the risk increased to four times for the Present Hedonists. As with the grade levels, the general pattern of results was the same for problematic drinkers relative to abstainers and problematic drinkers relative to moderate drinkers. As can be seen in Table 2, the risks are higher and similar for the Balanced and Past Negatives and substantially greater for the Present Hedonists. Discussion The present study sought to add to the increasing body of literature examining the relationship between time perspective and alcohol use. We began with validating a shortened version of ZTPI scores. Then, in addition to looking at the correlations between ZTPI scores and alcohol use, we examined adolescent alcohol use against temporal profiles. This approach is methodologically important as an individual holds more than one time attitude, at any one time. Thus, although the assessment of health behaviours against scores on any one ZTPI factor might be both interesting and useful, the personcentered approach is arguably more complete. The ZTPI (Zimbardo & Boyd, 1999) has become one of the most frequently used measures of time constructs in the literature. However, the increasing use of the instrument has parallelled a growing set of concerns about the psychometric properties of ZTPI scores (Shipp et al., 2009; Worrell & Mello, 2007), many of which remain (Sircova et al., 2014; Worrell, Table 2 Predictors associated with revised AAIS group membership: MLR (shown are OR þ95% CI). Reference category is abstainers Moderate drinkers
Gender Female Male School grade Grade 8 Grade 9 Grade 10 Grade 11 Grade 12 Time profile Future Balanced Past Negative Hedonist
Reference category is moderate Problematic drinkers
Problematic drinkers
B (SE B)
OR (95% CI)
p value
B (SE B)
OR (95% CI)
p value
B (SE B)
OR (95% CI)
p value
e .06 (.15)
e 1.06 (.80, 1.41)
e .68
e .42 (.21)
e 0.65 (.44, .99)
e .04
e .48 (.18)
e .62 (.43, .88)
e .008
e .31 (.18) .75 (.19) 1.15 (.19) 1.86 (.24)
e 1.36 (.95, .94) 2.12 (1.48, 3.05) 3.17 (2.18, 4.60) 6.40 (4.00, 10.23)
e .096
e 1.02 (.48) 2.26 (.44) 3.18 (.44) 4.31 (.47)
e 2.76 (1.08, 7.03) 9.58 (4.02, 22.81) 23.95 (10.19, 56.29) 74.53 (29.97, 185.34)
e .033
e .71 (.47) 1.51 (.43) 2.02 (.42) 2.46 (.43)
e 2.04 (.81, 5.12) 4.51 (1.93, 10.53) 7.56 (3.31, 17.30) 11.65 (4.99, 27.17)
e .131
e .69 (.15) .72 (.18) 1.38 (.20)
e 2.00 (1.50, 2.67) 2.05 (1.43, 2.93) 3.97 (2.68, 5.89)
e
