Cognitions About Smoking and Not Smoking in Adolescence Laura ter Doest, PhD Arie Dijkstra, PhD Winifred A. Gebhardt, PhD Salvatore Vitale, PhD

The theory of planned behavior identifies important proximal determinants of behavior, including attitude toward the behavior, perception of subjective norms exerted by significant others, and perception of perceived control over performance of the behavior. Because research in the planned behavior tradition has focused on desirable target behaviors, it is not clear how these determinants can best be conceptualized to account for adolescents’ acquisition of health risk behaviors such as smoking. This cross-sectional study compared the explanatory power of planned behavior constructs assessed in relation to “smoking” and “not smoking” in a sample of 248 Dutch secondary students (aged 12 to 17 years; 56% girls). The results indicated that four variables—attitude toward smoking, perceived subjective norm, and perceived behavioral control over both smoking and not smoking—best explained the adolescents’ smoking intentions and smoking behavior. Methodological and practical implications for smoking interventions are discussed. Keywords: adolescents; smoking; measurement

Most smokers take up the habit by the age of 18, and adolescents who smoke regularly usually continue to smoke as adults (Office for National Statistics, 2001; Royal College of Physicians, 1992; U.S. Department of Health and Human Services [DHHS], 1995). Research has identified several important classes of variables predicting the onset of adolescent smoking: demographic characteristics (e.g., socioeconomic status), the social environment (e.g., smoking by significant others), and proximal psychological variables (Conrad, Flay, & Hill, 1992; DHHS, 1995; Moolchan, Ernst, & Henningfield, 2000). One of the most widely applied models of the proximal predictors of behavior and behavior change is Ajzen’s (1988, 1991) theory of planned behavior (TPB). In its simplest form, the TPB identifies two constructs as proximal predictors of behavior: (a) intention (i.e., one’s intention to perform a behavior) and (b) perceived behavioral Laura ter Doest, Institute for Psychological Research, Leiden University, and NDDO Institute for Prevention and Early Diagnostics, Amsterdam, Netherlands. Arie Dijkstra, University of Groningen, Netherlands. Winifred A. Gebhardt, Institute for Psychological Research, Leiden University, Amsterdam, Netherlands. Salvatore Vitale, Addiction Research Institute (IVO), Rotterdam, Netherlands. Address correspondence to Dr. Laura ter Doest, NDDO Institute for Prevention and Early Diagnostics, Noordhollandstraat 71, Amsterdam, Netherlands 1081 AS; e-mail: [email protected]. Health Education & Behavior, Vol. 36 (4): 660-672 (August 2009) DOI: 10.1177/1090198107301329 © 2009 by SOPHE

660

ter Doest et al. / Cognitions About Smoking in Adolescence

661

control (i.e., the perceived ease or difficulty of performing the behavior). The theory also specifies three proximal predictors of intention: (a) attitude (i.e., one’s evaluation of the behavior as positive or negative), (b) subjective norm (i.e., one’s perception of whether significant others think that one should or should not engage in the behavior), and (c) perceived behavioral control. To date, the explanatory utility of the TPB has been demonstrated in numerous investigations of health-related and other behaviors, accounting for 40% to 50% of the variance in behavioral intentions, on average, and for 19% to 38% of the variance in prospectively measured behavior (Godin & Kok, 1996; Sutton, 1998; see also Armitage & Conner, 2000; Conner & Sparks, 1996). The theory was developed to explain planned or intentional behavior, and it can be applied in a straightforward way to most forms of health-enhancing behavior (e.g., exercise, dietary choice, health screening behavior). In this context, the TPB constructs can be unproblematically measured in relation to the desired behavior. However, applying the TPB to the acquisition of health risk behavior is less straightforward. When undesired behavior is being investigated, it is unclear whether the appropriate conceptual focus is the risky behavior (e.g., smoking) or the healthy behavioral alternative (e.g., not smoking). The theoretical literature and prior empirical research suggest several different approaches, which are considered and compared in the present article. Four Assessment Models The literature suggests four ways to conceptualize and operationalize the TPB determinants in relation to a health risk behavior such as smoking: (a) Assess all TPB determinants in relation to the healthy behavior (not smoking model), (b) assess all TPB determinants in relation to the risk behavior (smoking model), (c) assess all TPB determinants in relation to both healthy and risk behavior (dual assessment model), and (d) assess some TPB determinants in relation to healthy behavior and others in relation to risk behavior (mixed assessment model). Not Smoking Model. To date, the not smoking model has been used in numerous studies of smoking cessation in adult smokers, for whom the healthy behavior “not smoking” forms a logical and relevant target behavior (e.g., Godin, Valois, Lepage, & Desharnais, 1992; Hu & Lanese, 1998; Norman, Conner, & Bell, 1999). Studies of this type typically assess attitudes toward not smoking (ATT-NS), subjective norms about whether one should or should not smoke (SN), perceived behavioral control over not smoking (PBCNS), and intentions not to smoke (INT-NS). This approach is in line with operational guidelines in the literature suggesting that the planned behavior constructs should be assessed consistently in relation to a desired target behavior (e.g., Ajzen & Fishbein, 1980; Conner & Sparks, 1996; Francis et al., 2004; Godin & Kok, 1996). Smoking Model. One prior study operationalized the TPB determinants consistently in relation to the risk behavior of smoking (Maher & Rickwood, 1997) in an adolescent sample (aged 15 to 16 years; primarily nonsmokers). This study investigated ATT-S, SN, PBC-S, and INT-S. Maher and Rickwood’s (1997) use of the smoking assessment model—and, in particular, their operationalization of perceived behavioral control over smoking—is unique. Ajzen (1988) defined perceived behavioral control as the ease or difficulty of performing a behavior but suggested that the construct is primarily relevant for desirable outcome behaviors that are not entirely subject to internal volitional control. In contrast, Maher and Rickwood reasoned that perceived behavioral control over

662

Health Education & Behavior (August 2009)

smoking, an undesirable behavior, formed a relevant predictor for their adolescent sample because the construct would tap control in relation to external, environmental, and practical barriers. Dual Assessment Model. Some authors have suggested that for behaviors representing a choice between two (or more) options, it may be useful to measure the TPB determinants separately for each behavioral alternative (Ajzen & Fishbein, 1980; Sheppard, Hartwick, & Warshaw, 1988). Similarly, Conner and Sparks (1996) noted that measurement of attitudes in relation to both smoking and not smoking might offer unique insights. Nevertheless, no published study of smoking behavior has tested whether dual assessments in fact yield better prediction than assessment of single constructs. Mixed Assessment Model. A common feature of the not smoking model, the smoking model, and the dual assessment model is their consistent assessment of the TPB determinants in relation to the same target behavior(s). Nevertheless, to date, most empirical studies of the TPB and adolescent smoking have used a less consistent approach, with some constructs assessed in relation to smoking, some in relation to not smoking, and/or some in relation to both target behaviors (e.g., Carvajal, Hanson, Downing, Coyle, & Pederson, 2004; Chassin, Presson, Sherman, Corty, & Olshavsky, 1984; Flay et al., 1994; Hanson, 1997; Higgins & Conner, 2003; Hill, Boudreau, Amyot, Dery, & Godin, 1997; O’Callaghan, Callan, & Baglioni, 1999; Wilkinson & Abraham, 2004). Most research of this type has used a similar mixture of operationalizations, assessing ATT-S, SN, and PBC-NS; we refer to this as the mixed assessment model. The Present Study The present study evaluated and compared the ability of the four above-mentioned assessment models to account for adolescents’ intentions to smoke and not to smoke. Because the literature does not give grounds to favor one assessment model unequivocally over the others with regard to overall explanatory power, the model comparisons were exploratory. The literature does suggest that each model’s respective explanatory power should depend on the degree of compatibility or conceptual correspondence between the predictors and the outcome being considered (Ajzen, 1988). This led to the hypothesis that the smoking model should yield better prediction of INT-S (vs. INT-NS), whereas the not smoking model should better predict INT-NS (vs. INT-S). Finally, this study examined relationships between concurrently measured smoking behavior and TPB determinants operationalized in relation to smoking and not smoking.

METHOD Respondents The participants in this study were adolescents attending a public secondary school in a medium-sized city in the Netherlands. Questionnaires were distributed in eight classes representing three successive grade levels (equivalent to the seventh, eighth, and ninth grades) as well as different academic tracks ranging from general secondary education to college preparatory education. The sample included no vocational or special education students. (The latter comprise about 20% of Dutch secondary students; Rijksinstituut voor Volksgezondheid, 2002.) The sample may be regarded as a convenience

ter Doest et al. / Cognitions About Smoking in Adolescence

663

sample of middle- and upper-level Dutch secondary students. A sample size of about 250 was selected because prior research suggested that this sample size would yield sufficient power to detect relationships involving planned behavior constructs and adolescent health risk behavior. Procedure The adolescents completed the questionnaire at school during class hours. Their teacher and a researcher jointly introduced the research project to the students. It was explained that all responses provided in the questionnaire would be anonymous and that participation was voluntary. All students in class at the time of the data collection agreed to complete the questionnaire. Prior permission for the study was obtained from the school principal, in conformity with informed consent practices at the time of the study. Measures TPB Predictors. The TPB predictor variables (ATT-S, ATT-NS, SN, PBC-S, PBCNS) were operationalized in accordance with guidelines in the theoretical literature (e.g., Ajzen, 1988; Conner & Sparks, 1996). ATT-S and ATT-NS were measured using two 7-item semantic differential scales scored from 1 to 7 (1 = unhealthy; 7 = healthy). The seven items measuring ATT-S were preceded by the text “I think that smoking is…,” and those measuring ATT-NS by “I think that not smoking is…” Higher scores reflected more positive attitudes toward the target behavior. Both measures showed satisfactory reliability (αATT-S = .89; αATT-NS = .95). SN was measured with two items tapping perceived norms exerted by friends and parents, respectively: for example, “My friends think that I (1 = definitely should not smoke; 7 = definitely should smoke). This measure showed satisfactory reliability (α = .76). PBC-S and PBC-NS were each assessed with a single item (i.e., “If you wanted to, would it be difficult or easy to smoke [or: not to smoke]?”; 1 = difficult; 7 = easy). Smoking Intentions. Both INT-S and INT-NS were measured with two-item scales: “How strong is your intention to smoke [or: not to smoke] in the next 6 months” (1 = not strong; 7 = very strong); and “How likely is it that you will smoke [or: not smoke] in the next 6 months” (1 = not likely; 7 = very likely). Both measures showed satisfactory reliability (αINT-S = .87; αINT-NS = .77). Smoking Behavior. Smoking behavior was measured with a single question (“Which description fits you best?”) with five possible answers: “I have never smoked a cigarette, not even a puff” (never smokers); “I’ve never really smoked, but I have tried a few puffs” (experimenters); “I used to smoke, but now I’ve stopped” (ex-smokers); “I don’t smoke every day but every week”; “I smoke every day.” The latter two groups were combined to form a single subgroup (current smokers).

RESULTS Sample Characteristics The sample included 248 adolescents aged 12 to 17 years (M = 14.04 years, SD = 1.27 years; 44% boys, 56% girls). Most participants reported being of Dutch nationality

664

Health Education & Behavior (August 2009)

(96% Dutch vs. 3% non-Dutch). Self-reported smoking behavior was as follows: 9% current smokers (3% weekly, 6% daily), 7% former smokers, 25% experimenters, and 59% never smokers. Data Analysis Three sets of data analyses were conducted. First, bivariate and partial correlations were used to examine the strength of associations between the various TPB constructs as well as the degree to which unique variance in smoking intentions was accounted for by constructs assessed in relation to smoking as compared with not smoking. To evaluate the four assessment models, a series of four hierarchical regression analyses was carried out twice, once for each measure of smoking intentions (INT-S, INT-NS). In Step 1 of each analysis, variables were entered simultaneously to represent, respectively, the not smoking model (Analysis 1), the smoking model (Analysis 2), the dual assessment model (Analysis 3), or the mixed model (Analysis 4). In Step 2 of Analyses 1, 2, and 4, all remaining predictors were added in a stepwise fashion to test whether the remaining predictors could improve the regression solution. In Analysis 3, this second step was not carried out because all predictors had been entered in Step 1. A final series of analyses assessed whether the various TPB constructs could reliably distinguish between subgroups of adolescents defined in terms of their current smoking behavior (viz., current smokers, ex-smokers, experimenters, never smokers). In these analyses, tests for overall group differences were followed up with planned comparisons between adjacent categories. Data Screening Three cases were omitted because of missing values, yielding a sample of 245 for the analyses. Because departures from normality were observed on some variables, all correlational analyses made use of nonparametric correlations (Spearman’s ρ) and nonnormally distributed outcome variables were transformed prior to the regression analyses. Because of differences in sample size and variance between the four categories of self-reported smoking behavior, a nonparametric test (Kruskal-Wallis) was used to evaluate group differences. Correlational Analyses Means, standard deviations, and bivariate correlations are displayed in Table 1. To characterize the magnitude of observed correlations, we apply Cohen’s (1988) criteria for small (r ≥ .10), medium (r ≥ .30), and large (r ≥ .50) effect sizes. Intention to Smoke Versus Not to Smoke. All five TPB predictor variables (ATT-S, ATT-NS, SN, PBC-S, PBC-NS) showed significant correlations in the predicted direction with both measures of smoking intentions (INT-S and INT-NS). The intercorrelation between INT-S and INT-NS was large and highly significant. Attitude Toward Smoking Versus Not Smoking. The intercorrelation between ATT-S and ATT-NS was large, negative, and highly significant. Partial correlation analyses indicated that ATT-S was more strongly related to smoking intentions (i.e., INT-S and INT-NS) than ATT-NS: With ATT-NS partialled out, ATT-S still showed highly significant associations with both INT-S (Spearman’s ρ = .47, p < .001) and INT-NS (Spearman’s

ter Doest et al. / Cognitions About Smoking in Adolescence

665

Table 1. Descriptive Statistics and Bivariate Correlations for the Complete Sample

1. Gender 2. Age 3. ATT-S 4. ATT-NS 5. SN 6. PBC-S 7. PBC-NS 8. INT-S 9. INT-NS

1.

2.

3.

— –.03 –.10 .06 –.14* –.04 .00 –.02 .01

— .29*** –.16* .35*** .30*** .06 .26*** –.28***

— –.51*** .49*** .28*** –.24*** .58*** –.54***

4.

— –.36*** –.25*** .23*** –.39*** .36***

5.

— .28*** –.11 .42*** –.46***

6.

— .02 .26*** –.21**

7.

— –.31*** .31***

8.

— –.71***

9.

M

SD

Scale Range

—

1.57 14.04 1.63 6.07 1.97 4.71 6.06 1.66 6.17

0.50 1.27 0.87 1.40 1.17 2.31 1.75 1.48 1.62

1-2a — 1-7b 1-7b 1-7c 1-7d 1-7d 1-7e 1-7e

NOTE: N = 245. Correlations are Spearman’s ρ. ATT = attitude; SN = subjective norm; PBC = perceived behavioral control; INT = intention; –S = construct assessed for smoking; –NS = construct assessed for not smoking. Higher scores represent the following: a = female gender, b = positive ATT, c = SN favoring smoking, d = high PBC, e = strong INT, and f = more smoking. *p < .05. **p < .01. ***p < .001.

ρ = –.45, p < .001). In contrast, when ATT-S was partialled out, ATT-NS maintained only a small association with INT-S (Spearman’s ρ = –.13, p < .05) and a nonsignificant association with INT-NS. Perceived Behavioral Control Over Smoking Versus Not Smoking. Unlike the two attitude constructs, PBC-S and PBC-NS did not show a statistically significant intercorrelation. Partial correlation analyses indicated that both PBC-NS and PBC-S accounted for unique variance in smoking intentions: With PBC-NS partialled out, PBC-S maintained a significant correlation with INT-S (Spearman’s ρ = .28, p < .001) and INT-NS (Spearman’s ρ = –.22, p < .01); with PBC-S partialled out, PBC-NS maintained a significant correlation with INT-S (Spearman’s ρ = –.31, p < .001) and INT-NS (Spearman’s ρ = .31, p < .001). Evaluation of the Four Assessment Models Multiple regression analyses were used to evaluate the ability of the four assessment models to account for the adolescents’ smoking intentions (i.e., INT-S and INT-NS). As can be seen in Table 2 (Step 1, Analyses 1 to 4), all four models yielded significant regression solutions (p < .001) for both measures of smoking intentions (INT-S, INTNS), with relatively low explained variance for the not smoking model (26% to 28%) and relatively high explained variance for the smoking model, the mixed model, and the dual assessment model (41% to 48%). Step 2 in Analyses 1, 2, and 4 evaluated whether the variance in smoking intentions explained, respectively, by the not smoking model, the smoking model, and the mixed model could be improved by adding any of the remaining TPB determinants. For two models, this was found to be the case: The smoking model was improved by adding PBC-NS, yielding a 1% increase in explained variance in both INT-S (βPBC-NS = –.11, p < .05) and INT-NS (βPBC-NS = .12, p < .05); the not smoking model was improved by adding ATT-S, yielding a 22% increase in explained variance in INT-S (β = .60, p < .001) and a 15% increase for INT-NS (β = –.47, p < .001). The hypothesis concerning compatibility was weakly supported by the observed levels of explained variance: The smoking model accounted for 6% more variance in INT-S than in INT-NS, and the not smoking model accounted for 2% more variance in

666

Health Education & Behavior (August 2009)

Table 2. Hierarchical Regression Summary: TPB Determinants Predicting Adolescents’ Intention to Smoke and Intention Not to Smoke INT-S

Analysis 1 Step 1: Not smoking model ATT-NS SN PBC-NS Step 2: Other TPB determinants ATT-S PBC-S Analysis 2 Step 1: Smoking model ATT-S SN PBC-S Step 2: Other TPB determinants ATT-NS PBC-NS Analysis 3 Step 1: Dual assessment model ATT-S ATT-NS SN PBC-S PBC-NS Analysis 4 Step 1: Mixed model ATT-S SN PBC-NS Step 2: Other TPB determinants ATT-NS PBC-S

R2

Δ R2

.26***

.26***

INT-NS β

R2

Δ R2

.28***

.28***

–.09 .42*** –.20*** .48***

.22***

.07 –.44*** .20*** .43***

.15***

.60*** n.s. .47***

.47***

–.47*** n.s. .41***

.41***

.60*** .11 .05 .48***

.01*

–.51*** –.20** –.01 .42***

.01*

n.s. –.11* .48***

.48***

n.s. .12* .43***

.43***

.57*** –.03 .12* .06 –.11* .48***

.48***

–.47*** .02 –.20** –.01 .12* .42***

.42***

.58*** .12* –.11* .48***

.00

β

–.48*** –.20** .12* .42***

n.s. n.s.

.00 n.s. n.s.

NOTE: N = 245. TPB = theory of planned behavior; ATT = attitude; SN = subjective norm; PBC = perceived behavioral control; INT = intention; –S = construct assessed for smoking; –NS = construct assessed for not smoking; n.s. = nonsignificant (i.e., construct was not entered because stepwise procedures revealed no significant improvement of regression solution). *p < .05. **p < .01. ***p < .001.

INT-NS than in INT-S. These results should, however, be interpreted with caution because it is not possible to evaluate differences between these nonnested models with an inferential statistical test. TPB Constructs and Smoking Behavior As can be seen in Table 3, the Kruskall-Wallis tests showed strong associations between the adolescents’ current smoking behavior and all seven TPB constructs (ps < .001).

ter Doest et al. / Cognitions About Smoking in Adolescence

667

Table 3. TPB Constructs and Current Smoking Behavior Current Smokers (n = 22)

Ex-Smokers (n = 18)

Experimenters (n = 62)

Never Smokers (n = 143)

M

SD

M

SD

M

SD

M

SD

KruskallWallis χ2 (3 df)

ATT-S

3.47

0.73

2.36

0.88

1.51

0.55

1.31

0.55

84.47***

ATT-NS

5.45

0.93

5.63

0.82

5.99

1.45

6.26

1.47

42.63***

SN

3.50

0.87

2.78

1.24

1.91

1.15

1.65

0.97

49.19***

PBC-S

6.27

1.52

6.33

1.28

4.90

2.15

4.17

2.38

27.43***

PBC-NS

4.41

2.36

6.11

1.53

6.34

1.44

6.19

1.68

18.77***

INT-S

5.07

1.58

2.22

1.61

1.40

0.99

1.17

0.73

109.61***

INT-NS

2.63

1.63

5.50

1.75

6.34

1.37

6.72

0.80

99.45***

Pairwise Comparisons Cur >*** Ex >*** Exp >** Nev Cur = Ex ** Exp = Nev Cur = Ex >*Exp >* Nev Cur *** Ex >* Exp >** Nev Cur