Abstract
An implicit association test (IAT) was used to investigate how habit strength, implicit attitudes and fruit consumption interrelate. Fifty-two participants completed a computerized IAT and provided measures of fruit consumption and related habit strength. Implicit attitudes moderated the habit strength—fruit consumption relationship; stronger relationships were observed when implicit attitudes were more positive. Amongst those with strong fruit habits, more positive associations with fruit were found for those who had recently consumed sufficient fruits compared to those who had not. Findings demonstrate the relevance of implicit positive associations in understanding the relationship between fruit consumption habits and subsequent fruit consumption.
Introduction
Public health policies have stressed the importance of healthy lifestyles, including sufficient physical activity and fibre consumption, such as an adequate intake of fruits. Despite the demonstrated beneficiary health effects of sufficient fruit intake, a fairly large proportion of population segments eat inadequate amounts of fruit, even though most people have a positive intention to eat healthy (Wammes, et al, including sufficient fruits (de Bruijn, 2010; de Bruijn et al., 2007). Both more traditional reasoned action models and more recent habit theory models have been used to identify relevant correlates of fruit consumption in order to inform health behaviour change interventions (Blanchard et al., 2009; de Bruijn et al., 2007; de Bruijn, 2010; Godin et al., 2010; Guillaumie, et al, 2010; Reinaerts, et al, 2007; Strachan & Brawley, 2008; Strachan & Brawley, 2009). Studies that have adopted an integrative perspective (i.e. combining both reasoned action models and habit theory models) have shown that measures of habit strength not only improve the amount of explained variance in fruit consumption and that habits affect fruit consumption directly (i.e. unmediated by attitudes and intention), (Brug et al., 2006; de Bruijn, 2010; Reinaerts et al., 2007) but also that habit and intention interact: intention becomes a less important predictor of fruit intake when fruit consumption increases in habit strength (de Bruijn et al., 2007; de Bruijn, 2010). These latter findings have been explained by the fact that more habituated behaviours are more impulsive and automatically initiated behaviours, thereby minimizing the effect of conscious and planned intentions on behavioural performance. Because similar results have been obtained in other health behaviours (de Bruijn & Gardner, 2011; de Bruijn, et al., 2009b; de Bruijn, et al, 2008; Rhodes et al., 2010a;), researchers have not only begun including habit strength measures in behavioural determinant studies, but also as a potential intervention effectiveness modifier (Adriaanse et al., 2010). Moreover, it has been suggested that habit strength should be viewed as an intervention outcome measure (Verplanken & Wood 2006), because behaviours that have been habituated should no longer need intervention encouragement to be maintained (Verplanken, & Wood 2006).
Nevertheless, there is some evidence indicating that, even in behaviours that have become habituated, relapses can occur. For instance, De Bruijn (2011) recently reported that some 40-60 percent of participants with strong exercise habits and intentions were not engaging in sufficient exercise at follow-up. Similar results have been reported for fruit consumption (de Bruijn, 2010) suggesting that even behaviours that have been habituated cannot be maintained consistently at sufficient levels. Further, those who report strong habits but miss a behavioural opportunity do not lack in cognitive measures, such as positive instrumental attitudes and strong intentions (de Bruijn, 2010; de Bruijn, 2011; Rhodes et al., 2010b). Therefore, cognitive constructs outlined in reasoned action models might not be sufficiently able to understand this ‘habit-behaviour’ gap.
From a more recent social psychological viewpoint, there appears to be merit to consider implicit positive attitudes as a potential concept that may explain this gap. Within the social psychology literature, attitudes have long been acknowledged as important predictors of behaviour (Ajzen & Fishbein, 1977; Fazio, 1990; Strack & Deutsch, 2004). More recent accounts of attitude—behaviour relationships have acknowledged that attitude can influence behaviour in an explicit and deliberate manner, but also in a more automatic and implicit way (Conner et al., 2007; Fazio, 1990; Strack & Deutsch, 2004). Additionally, two-factor attitude models have indicated that the attitude concept can be structured in both a cognitive and an affective component (Breckler & Wiggins, 1989; Rhodes & Courneya, 2003). The cognitive component of the attitude construct reflects the consideration of instrumental evaluations of behaviour, such as obtaining health benefits when eating sufficient fruits, whereas the affective component reflects experiential and emotional aspects of behavioural performance, such as the enjoyment of eating a healthy diet. Research has also indicated that the pathways that link attitudes with behaviour differ for cognitive and affective attitudes. Not only are cognitive attitudes generally weaker predictors of health behaviour than affective attitudes (Lawton, et al., 2009; Rhodes et al., 2009), affective attitudes also influence behaviour in a more direct (i.e. unmediated by planned intentions) manner than cognitive attitudes do (Lawton, et al., 2007; Lawton et al., 2009), thereby reflecting similar unmediated habit-behaviour relationships (de Bruijn & Van den Putte, 2009; Verplanken & Orbell, 2003). The direct relationship between affective attitudes and behaviour arguably occurs because evaluations of how people feel about an attitude object initiates an immediate and automatic behavioural response (Zajonc, 1980; Conner et al., in press). Moreover, some (Conner et al., in press) have argued that explicit affective attitudes may tap impulsive and automatic influences on health behaviour. This argument is based on evidence that shows incremental validity of implicit measures over cognitive attitudes in predicting health behaviour, but not over affective attitudes (Blanton & Jaccard, 2008). Thus, affective considerations of behaviour may initiate behaviour in a manner that is similar to how habitual processes influence behaviour. Not only has habit strength has been found to be medium-to-large effect sized positive correlate of explicit attitude measures towards various health behaviours, such as exercise, bicycle use and fruit consumption (de Bruijn et al., 2009b; de Bruijn & Gardner, 2011; de Bruijn, 2010; Rhodes et al., 2010a), implicit attitudes have also been found to be better predictors of health behaviour when that behaviour is strongly habitual (Conner et al., 2007). Finally, more fundamental social psychological research has shown that unconsciously linking a particular behavioural state with positive affect can promote goal attainment automatically (Custers & Aarts, 2005).
Taken together, these findings suggest a potential overlap between affective considerations and habitual behaviour, indicating that affective attitudes may be relevant in shaping and maintaining automatic and impulsive health behaviour. Although some evidence has shown that implicit attitudes are better predictors of health behaviour when that behaviour is strongly habitual (e.g. Conner et al., 2007), no research at present exists that has studied how habits, implicit attitudes and health behaviour interrelate. The present study was set up to address this issue. Based on the aforementioned evidence, it was hypothesized that more positive implicit associations with fruit consumption would be observed (1) amongst those who reported a strong fruit consumption habit and (2) amongst those who had recently consumed sufficient fruits. Further, we hypothesized that habits and implicit attitudes would interact, such that habits would be better predictors of fruit consumption at more positive levels of implicit fruit consumption attitudes.
Methods
Participants and procedures
Data from the present study was available from a convenience sample of university students (N = 52, mean age = 23.21 (SD = 4.18), 55.8% females) who were approached in university canteens. They were informed that the study would take about 10 minutes and would involve several reaction time tasks that required them respond as accurate and as fast as possible upon being presented with several words and pictures. Upon agreeing to participate, participants were led to a computer room, where implicit attitude measures were collected individually in Inquisit 3.0, which was run on an Acer computer with a 15 inch screen running at 1024*768 pixels and a refresh rate of 75 Hz. Participants were seated approximately 70 centimetres from the screen and were instructed to complete a sequence of five computerized tasks that made up the IAT measurement phase (Greenwald et al., 1998).
Briefly, the first task was used for target-concept discrimination (utensils vs. fruits), the second task for the associated attribute discrimination (positive and negative words), the third task for the initial combined task (fruit or positive words vs. utensils or negative words), the fourth task for the reversed target-concept discrimination (fruits vs. utensils) and the final task for the reversed combined task (fruit or negative words vs. utensils or positive words) (for details, see (Greenwald et al., 1998)). The target words that were used in the IAT were general positive (happiness, pleasant, friendly, joy, love) and negative words (disgust, hate, awful, sad, boring) derived from previous research on implicit attitudes (Wiers et al., 2005; de Liver et al., 2007; Greenwald et al., 1998). Pears, strawberries, apples, bananas, and lemons were used as fruit examples, as they are amongst the most commonly eaten fruits in the Netherlands (Netherlands Nutrition Centre, 1998) while glasses, boxes, towel, scissors, and paper clip were examples of utensils. In the present experiment, category labels were presented in either the upper left or upper right side of the screen, while pictures of utensils and fruits appeared in the middle part of the screen. The two critical phases are those in which target and attributes are paired (task 3 and 5), with the IAT effect a difference score that is computed from these two different pairings: a more positive score reflects a more positive association with (for this experiment) fruit consumption. When an incorrect response was given, the word ‘error’ appeared in the location of the stimulus: at the end of each trial, participants were provided with a summary of their mean response latency and the percentage of correct responses.
After completing the IAT measurement phase, participants completed an online survey in which fruit consumption and fruit consumption habit strength were assessed using validated questionnaires. It was chosen to assess behavioural and habit strength measures after the experiment, because prior assessment may trigger relevant thoughts about the behaviour under study and influence response latencies (Bargh et al., 2000). Regarding fruit consumption, participants were requested to indicate frequency (how many days in the past two weeks) and usual amount on such a day for commonly eaten fruits in the Netherlands, such as apples, bananas, tangerines and oranges. This measure has been validated against seven-day diary and biomarkers (Bogers et al., 2004). Fruit consumption per day was computed by multiplying frequency and usual amount and dividing the resultant score by14. Those who consumed at least two pieces of fruit per day were coded as sufficient fruit consumers, reflecting guidelines from the Netherlands Nutrition Centre: those who consumed less were coded as insufficient fruit consumers. Habit strength was assessed regarding eating at least two pieces of fruit per day using the self-reported habit index (Verplanken & Orbell, 2003), that requested participants to indicate whether key constructs of habit strength (i.e. automaticity, lack of control) towards eating two pieces of fruit per day applied to them (+3 = totally agree; -3 = totally disagree). Items related to past behavioural frequency were omitted from this measure due to overlap with the fruit consumption measure, which may lead to inflated coefficients. Cronbach’s alpha was .92. After completion, participants were debriefed and thanked for their participation. The Institutional Review Board provided ethical approval for the study.
Analysis
The relationship between implicit attitudes, fruit consumption and habit strength was assessed using two sets of analyses. The first set focused on identifying relevant correlates of fruit consumption. This was done using correlational analysis, followed up by stepwise regression analysis in which fruit consumption in pieces per day was the dependent variable and age and gender (step 1), implicit attitude towards fruit consumption (step 2), fruit consumption habit strength (step 3), and fruit consumption habit strength * implicit attitude interaction (step 4) were the independent variables. Cohen’s r and f 2 were the effect size estimates for these analyses (Cohen, 1992) and the Aiken and West (1991) strategy was used for computing and decomposing the interaction term. The second set focused on dichotomizing fruit consumption (insufficient vs. sufficient consumers) and fruit consumption habit strength (below midscale = weak habit; above midscale = strong habit) that created four possible profiles (1 = weak habit, insufficient fruit (n = 30); 2 = weak habit sufficient fruit (n = 3); 3 = strong habit, insufficient fruit (n = 11); 4 = strong habit, sufficient fruit (n = 8)). Scores on implicit attitude measures were subjected to a 2 (fruit norm: yes vs. no) x 2 (habit strength: weak or strong) between-subjects analysis of variance with η2 as the effect size estimate (Cohen, 1992). Significant effects were followed up by univariate analysis of variance with Games-Howell post-hoc comparison to take unequal group sizes into account. Cohen’s d was used as the effect size estimate for these latter analyses (Cohen, 1992).
Implicit attitude scores were computed using Greenwald’s suggestions (Greenwald et al., 1998). Response latencies faster than 300 milliseconds were recoded into 300 milliseconds and latencies slower than 3000 milliseconds were recoded into 3000 milliseconds. Finally, response latencies were log-transformed. No participants had error rates larger than 10 percent (Greenwald et al., 1998), so all participants were retained for the analyses.
Results
Descriptives and correlational and regression analysis
Mean fruit consumption was 1.57 (SD = 1.18) pieces of fruit per day, with 11 participants (23.1%) meeting the Dutch norm of two pieces of fruit per day. Large effect-sized correlations with fruit consumption were found for habit strength and implicit attitudes, whereas a large effect-sized correlation was also found for habit strength and implicit attitudes (Table 1). The results from the linear regression analysis (Table 2) showed that, in the first step, fruit consumption was significantly correlated with gender and age, such that those who were older and of female gender consumed more fruits. In the second step, only gender remained a significant correlate: implicit fruit consumption attitude was also a significant positive correlate of fruit consumption, while in the third step, only implicit attitude and fruit consumption habit strength were significantly associated with fruit consumption. The addition of the fruit consumption habit strength * implicit attitude interaction significantly increased the amount of explained variance to a total of 61 percent indicating a large effect size (f2 = 1.56). Moreover, the final step also revealed a significant interaction term that was decomposed following standard guidelines (Aiken & West, 1991). These simple slope analyses showed that the habit strength—fruit consumption relationship increased in strength as implicit attitudes towards fruit consumption became more positive. Specifically, habit strength was a significant correlate of fruit consumption at low level of implicit attitude (β = .39, p = .02), but stronger at medium (β = .54, p < .001) and high (β = .68, p < .001) levels of implicit attitudes.
Mean scores, standard deviation (in parentheses) and bivariate correlations between study variables (N = 52)
p < .05; ** p < .01; *** p < .001
Results from stepwise regression analysis with fruit consumption in pieces per day as the dependent variables and age and gender (step 1), implicit fruit consumption attitudes (step 2), fruit consumption habit strength (step 3), and habit strength * implicit attitude interaction as the independent variable (N = 52)
p < .05; ** p < .01; *** p < .001
Analysis of variance
There was a main effect of fruit consumption norm, F (1,48) = 53.99, p < .001, η2 = .53, and fruit consumption habit strength, F (1,48) = 14.23, p < .001, η2 = .23 on implicit fruit consumption attitude, demonstrating that mean scores for implicit fruit consumption attitude were not only significantly different between those who were not eating sufficient fruits (M = -.18, SD = .13) and those who were (M = .23, SD = .20), but also between those who reported weak (M = -.13, SD = .16) and strong (M = .30, SD = .27) fruit consumption habits. Cohen’s d for these differences was 2.58 and 2.27, respectively, indicating large effect-sized differences. A significant habit strength x fruit consumption norm interaction was also found, F (1, 48) = 18.31, p < .001, η2 = .26, demonstrating that the effect of fruit consumption habit strength on implicit fruit consumption attitudes was stronger amongst those who recently consumed sufficient fruits as compared to those who did not. Post-hoc analyses (Table 3) showed that those who reported sufficient fruit consumption and strong fruit consumption habits had more positive implicit associations than all other categories: Cohen’s d for these differences was 5.37 and higher, indicating large effect sizes. Also, those who reported insufficient fruit consumption but strong fruit consumption habit had more positive implicit associations with fruit consumption than insufficient consumers with weak fruit consumption habits: Cohen’s d was 1.92.
Mean scores and standard deviation for IAT measures across profiles created from fruit consumption norm (yes vs. no) and fruit consumption habit strength (low vs. high) (N = 52)
p < .05; ** p < .01; *** p < .001
Discussion
The present study was set up to study the interplay between fruit consumption, fruit habit strength and implicit attitudes towards fruit consumption. Results showed relevant intercorrelated patterns that were mostly in the medium-to large effect-size range. For instance, there were strong and positive bivariate associations between implicit attitudes and fruit consumption habit strength with self-reported fruit consumption, that remained positive and in the medium effect-size range in the multivariate model. Notably, these two significant correlates were able to account for more than half of the total variance in self-reported fruit consumption. Given that a recent review of psychosocial determinants from the theory of planned behaviour and/or social-cognitive theory were able to predict 17 percent to 28 percent variance in fruit intake (Guillaumie et al., 2010), results from the present study underline the relevance of incorporating habit strength and implicit measures in health behavioural theories. Thereby, a more reflective, impulsive and/or automatic route to behaviour can be identified that may necessitate a different health intervention approach than outlined in more traditional social-psychological models (Kremers et al., 2006).
Not only were more positive associations observed for those who had consumed sufficient fruits in the past two weeks (as compared to those who did not eat sufficient fruits), those who reported strong fruit consumption habits also had more positive associations with fruit consumption than those who reported weak fruit consumption habits. However, the interplay between habits, behaviour and implicit attitudes was perhaps best demonstrated by the observed habit strength x implicit attitude interaction in the regression model when predicting fruit consumption, but also by the habit strength x fruit consumption norm interaction in the analysis of variance model. Specifically, the simple slope analysis showed that, at more positive levels of implicit attitudes, habit strength was nearly twice as strong a correlate of fruit consumption than at low levels of implicit attitudes. Notably, the observed habit strength—fruit consumption relationship at this positive level was more than 50 percent stronger than the summarized r from a recent meta-analysis on habit-behaviour relationships (fixed r = .44) (Gardner et al., in press), indicating substantial relevance of positive implicit associations to strengthen the habit strength—health behaviour relationship. This suggestion was further substantiated when response latencies were compared between profiles created from fruit consumption and fruit consumption habit strength. Even though stronger fruit habits and fruit consumption norm were independently and positively related to response latencies, they also interacted to produce stronger associations between implicit attitudes and fruit consumption habit strength amongst those who reported sufficient fruit consumption in the past two weeks as compared to those who did not eat sufficient fruits in the same reference period. This finding suggests that behaviours that have become habituated have implicit positive properties that are reinforced by recent behavioural experiences.
Investigation of habit strength and fruit consumption distributions showed that about a quarter of those who had strong fruit consumption habits had not eaten sufficient fruits in the two weeks prior to the experiment. Although these findings may indicate insufficiency of the habit strength measure, research on habit formation development has indicated that occasionally missing behavioural opportunities is not detrimental to habit formation (Lally et al., 2009). Rather, these and other findings ( de Bruijn 2010; de Bruijn, 2011; Rhodes et al., 2010b) provide further evidence for the existence of a habit-behaviour gap that suggests that even those who have habituated their (health) behaviour fail to act occasionally. Identifying relevant modifiable variables that can minimize this ‘habit-behaviour’ gap may thus hold great potential for public health promotion. Findings from the present study suggest that more positive (implicit) associations with fruit consumption may be such a variable that increases adherence to fruit consumption public health guidelines amongst those who already have strong fruit consumption habits. As previously outlined, there is also evidence showing that commonly applied social-cognitive variables in health educational interventions (such as planning and instrumental attitudes) are relatively unable to distinguish profiles created from behaviour and habit strength (de Bruijn, 2011). Finally, there is increasing evidence that positive affective evaluations make behavioural enactment more likely, particularly in comparison with cognitive attitudes (Conner et al., 2011; de Bruijn et al., 2009a; Keer, et al., 2010; Lawton et al., 2007; Lawton et al., 2009; Lowe, et al., 2002Van den Berg, et al., 2005).
A potential reason for the dominance of these affective evaluations is that the positive affect acts as a mental shortcut for behavioural enactment that is unmediated by cognitive considerations (Keer, et al., in press; Lawton et al., 2009) and/or as an implicit motivator of goal-direct behaviour (Custers & Aarts 2005). For instance, significant associations with implicit attitudinal measures have been found in a diverse range of health behaviours such as alcohol consumption (Pieters et al., 2010; Wiers et al., 2005) and high-fat foods (Papies et al., 2009), even when cognitive constructs such as motivation are statistically controlled for (Conroy et al., 2010). However, there is also evidence that increments in explained variance by implicit attitude measures only occurs when compared against the influence of cognitive attitudes and not when compared against the influence affective attitudes (see (Conner et al., in press) for a detailed discussion on this issue). Given that the current state of the evidence regarding the incremental value of implicit measures is relatively young and inconsistent, future research should preferably use experimental testing to indicate how and when implicit measures affect health behaviour over and above more explicit measures. This notion appears prudent, because there is at present only limited experimental intervention research on how to change health behaviour through implicit and/or explicit positive affect strategies. However, those that have been conducted suggest that these positive affect messages are more effective in changing health behaviour than cognitive messages are (Conner et al., 2011; Parrott et al., 2008).
A few limitations of the present study are worth mentioning. First, self-reported measures for fruit consumption and habit strength were employed which may have been biased by self-representation of social desirability biases. Although both survey measures have been validated against more objective measures such as diary records, biomarkers and reaction times tests, it would be preferable for future research to link implicit attitude measures with more objective behavioural and habit strength measures, such as biomarkers and reaction time tests, respectively. Second, a convenience sample of undergraduate students was used, so generalization to the population at large should be cautioned. Although the basic underlying patterns of affective associations with fruit consumption should not be necessarily different in other populations, the present study sample consisted of fairly highly educated respondents, which have been known to have better health practices and knowledge than other population segments. Third, even though the IAT measure employed in the present study has been used extensively in other studies, recent calls have been made for replacing this earlier IAT measure and its associated scoring procedure with an updated measure and scoring algorithm. This newly developed measure generally outperforms the conventional measure on the magnitude of implicit —explicit correlations (Greenwald et al., 2003). Relatedly, compatible and incompatible blocks were presented in a fixed order, with the compatible block preceding the incompatible block. Ideally, these orders should be counterbalanced, because some evidence exists that shows IAT effects to be faster when the compatible block precedes the incompatible block (Greenwald et al., 1998). Although this fixed ordering may not have influences between-groups differences in IAT-scores, replication of our study results using counter-balancing with the updated IAT measure and scoring procedure should shed some more light on the strength of the association of fruit consumption, habit strength measures and implicit fruit attitudes.
Despite these limitations, results from the present study suggest that interventions that highlight positive associations with fruit consumption should strengthen the habit—fruit consumption relationship. Although positive affect applications have been relatively absent in health promotional efforts (Conner et al., 2011; Rhodes et al., 2009), evidence from commercial advertising research has demonstrated the usefulness of emotional strategies, positive affect messages and advertisements on consumer behaviour (Monahan, 1995; Van den Putte, 2009). More research on implicit and affective considerations of fruit consumption should benefit our understanding of fruit consumption and subsequently allow for better designed interventions.