Design
A longitudinal study was conducted with a baseline measurement (T0), a follow-up after 1 month (T1) and another follow-up after 3 months (T2).
Power analysis
With the assumption of a small effect size (f2 = 0.023) for a main effect or interaction effect of implicit attitude and a test power set at 0.80 with a type I error rate of α = 0.05 for two-sided testing, power analysis revealed that 330 respondents are needed. Anticipating a drop-out rate of 20%, we aimed to conduct the first session of the study with 413 participants in order to have data from at least 330 participants at the first follow-up.
Participants and recruitment
Following approval, the study was conducted in the Behavioral and Experimental Economics Laboratory (BeeLab) of Maastricht University. Students registered in the BeeLab database were invited to participate. As most registered students were of either German or Dutch nationality, the study was conducted in these two languages. Thus, being Dutch or German was the only inclusion criterion for being invited. In total, 340 students (61% female, mean age = 21) participated in the baseline measurement. At the first follow-up, 240 students participated (71% of baseline, 64% female, mean age = 21) and a total of 128 students (38% of baseline, 69% female, mean age = 22) completed the second follow-up, 3 months after baseline.
Procedure
Potential participants registered in the BeeLab database received an invitation email containing the following information: the study aims to gain insight into the relationships of cognitions related to PA; it consists of three waves; one measurement is comprised of 2 tasks which together take 30 min to complete; there are no expected risks associated with participation; all data will be gathered and analyzed anonymously; participants will receive 15€ in cash after the first two waves and another 7,50€ in cash after participation in the third wave. Those willing to participate could select a timeslot from two given days for each wave. One day before participating, a reminder was sent. On the day of participation, participants were welcomed in the lab, received instructions, and informed consent was obtained from all individuals included in the study. In the first part, participants performed a modified version of the Single-Category Implicit Association Test (SC-IAT) [46] to assess implicit attitudes towards PA. In the second part, participants filled in a self-report questionnaire to measure explicit cognitions and PA behavior. Explicit cognitions were assessed subsequently as a prior assessment of explicit cognitions is assumed to trigger thoughts related to PA which in turn might influence the reaction time in a following task [47]. The SC-IAT and the questionnaire were available in Dutch and in German. After completion participants were thanked and if they took part in follow-ups received their incentive at T1 and T2.
Measurements
Implicit attitude assessment task
Implicit attitudes towards PA were measured with the SC-IAT. Whereas the IAT relies on the comparison of two opposite categories, e.g. men versus women, the SC-IAT does not. Regarding PA, it is difficult to define a clear opposite category as PA behavior occurs on a continuum. Moreover, the SC-IAT has proved to predict objectively-measured physical activity [38] and unintentional physical activity [38, 48]. Also, adequate internal reliability and predictive validity were demonstrated [46].
The SC-IAT consisted of two blocks, each comprising 24 practice trials and 72 test trials. In one block, “physical activity or positive” formed one category and “negative” the other category. In the other block, “physical activity or negative” was one category and “positive” the other. It is assumed that a person possesses a positive implicit attitude when he or she is quicker to categorize the displayed stimuli when “physical activity or positive” form one category than when “physical activity or negative” are one category. When this pattern is reversed, the person is assumed to hold a negative implicit attitude. The order of the two blocks was counterbalanced, meaning that the block “physical activity or positive” and “negative” had to be performed first by some participants, whereas other participants performed the block “physical activity or negative” and “positive” first. Labels for the two categories were presented on either the left or right upper part of the screen throughout the task. One by one, stimuli were presented in the centre of the screen and participants had to press e on their keyboard when the stimulus belonged to the category presented on the left or i when the stimulus belonged to the category displayed on the right. The sequence in which the stimuli were presented was randomized and words appeared an equal number of times. When an incorrect answer was given, a red X appeared on the screen until a correct answer was given.
Positive and negative words were selected from the Affective Norms for English Words (ANEW) [49] based on their valence and arousal norms. The words were translated to and from Dutch and German by German and Dutch native-speaking researchers of Maastricht University. In an informal pretest, 26 German and 22 Dutch students of Maastricht University rated the words with regard to the perceived levels of valence (1 = very negative to 9 = very positive), arousal (1 = not arousing at all to 9 = very arousing), and familiarity (1 = very unfamiliar to 9 = very familiar) in their respective mother tongue. On this basis, the following positive words were selected: love, freedom, joy, success and party (translated from German and Dutch). The selected negative words were: depression, demon, lie, infection, and poison (translated from German and Dutch). Words representing PA were carefully chosen from earlier studies in which the SC-IAT was used to assess implicit attitudes towards PA [38, 48]. These were also translated to and from German and Dutch and pretested for their representativeness for PA in both languages (1 = not representative at all, 2 = not very strongly/moderately representative, 3 = strongly representative). The seven words that were highly representative for PA were: running, biking, kickboxing, sprinting, jogging, weight-lifting, and (doing) sit-ups (translated from German and Dutch).
The SC-IAT was programmed using Inquisit by Millisecond software and the script was based on Karpinski and Steinman [46]. The implicit attitude was formed by d-scores, calculated automatically using Inquisit software by subtracting the average response time for the test block with the categories physical activity or positive/negative from the average response time of the test block with the categories physical activity or negative/positive. This score was then divided by the standard deviation of all test trials. This procedure is based on the improved scoring algorithm as described by Greenwald and colleagues [50]. D-scores can range from − 2 to + 2 with negative values representing a negative implicit attitude and positive values representing a positive implicit attitude. The higher the d-score the more positive an implicit attitude. Reliability test of the SC-IAT was calculated based on the procedure as described in Karpinski and Steinman [46] and revealed an acceptable value of r = .83.
Self-report assessment
All explicit cognitions referred to adequate physical activity. Adequate PA for adults was defined as being moderately physically active five times a week for at least 30 min. Moderately active is described as, for instance, brisk walking with an increase in heart rate [51]. This definition was presented to the participants and could be re-read at any time while answering the questionnaire. The questions to measure explicit cognitions were based on the I-Change model [14]. For the full questionnaire, see Additional file 1.
Explicit attitude was assessed using 20 items that were rated on a 5-point Likert Scale. Ten items assessed the perceived cons of adequate PA (Cronbach’s α = .83) and 10 items assessed the perceived cons of adequate PA (Cronbach’s α = .77). One example item for pros is “When I am adequately active it is” with answer options ranging from (1) “very good for my health” to (5) “not good for my health”. Items were reversed so that higher values represent the perception of more pros. An example for cons is “When I am adequately active it is” with answer options from (1) “too time-consuming” to (5) “not time-consuming”. Items were reversed, so that lower scores represent the perception of fewer cons. One scale score for perceived pros and one for perceived cons were created for the analyses.
Social norms and social modeling were assessed by four questions. Answers were given on a 5-point Likert scale and assessed the norms about adequate physical activity of family members, partners, and friends (Cronbach’s α = .74) and their PA behavior (Cronbach’s α = .48). An item representing norms was “Most of my friends” (1) “certainly think that I need to be adequately active” to (5) “certainly do not think that I should be adequately active”. An additional answer option: “I don’t have any friends/Not applicable” was given as a sixth option. A modeling item was “Most of my friends are adequately physically active” with answer options from (1) “totally agree” to (5) “totally disagree”. The additional answer option “I don’t have any friends/Not applicable” was also available. These answers were not included in the analyses. Norms and modeling items were reversed with higher scores representing stronger norms or modeling. The mean scale scores for norms and modeling were included in the analyses.
Self-efficacy was measured by nine items, also on a 5-point Likert scale (Cronbach’s α = .74). These items enquired about the extent to which respondents thought they would be able to be adequately physically active in different situations. For instance “I find it difficult/easy to be adequately physically active when I am tired” with answer options from (1) “very difficult” to (5) “very easy”. Questions were based on those used in former studies about PA [15, 52, 53]. Higher scores indicate higher self-efficacy. The mean scale score was included in the analyses.
Intention was measured by three items on a 5-point Likert scale (Cronbach’s α = .87). The first item assessed whether respondents intended to become adequately physically active within the next 3 months, ranging from (1) “yes, absolutely” to (5) “no, not at all”. The second item asked whether respondents were motivated to become adequately physically active within the next 3 months with answer options ranging from (1) “totally agree” to (5) “totally disagree”. The third item measured how high the chances were of becoming adequately physically active within the next 3 months. Answer options ranged from (1) “very little” to (5) “very high”. The first two items were reversed, so that higher scores represent a stronger intention. The mean score of all three items was included as scale score for intention in the analyses.
Physical activity levels were measured by the Short Questionnaire to Assess Health-enhancing physical activity (SQUASH). This has been proven to be a reliable and valid tool for assessing PA levels among Dutch adults [54, 55] and has been applied in former studies [15,16,17, 53]. Completing the SQUASH takes around 5 min; it assesses different domains of PA, namely commuting activities, activities at work, household activities, and leisure time activities. For each activity, frequency (days per week), duration (minutes per day) and intensity (light/moderate/intense expressed in metabolic equivalent of task, MET) were measured. MET values for sport activities were derived from Ainsworth and colleagues [56]. Based on the procedure of Wendel-Vos and colleagues [54], the total minutes of an activity were calculated by multiplying frequency by duration. These were then multiplied by the intensity in order to obtain an activity score for each activity. A total activity score was calculated by summing all activity scores. The higher the score, the more physically active a person is.
Additionally, participants gave information about their age, gender, use of drugs, alcohol or medications that could influence their reaction time, and whether they were able to be physically active in the recent past.
Analyses
Differences between the German and Dutch version of the tests were tested in advance. No significant differences were found. Descriptive analyses were conducted to describe the sample. To assess whether study variables changed significantly over time, linear mixed models were used. Logistic regression analysis was used to evaluate whether dropout was predicted by age, gender, perceived pros, perceived cons, social norms, social modeling, self-efficacy. All analyses were done with SPSS version 23.
For the first hypothesis, two hierarchical multiple regressions were performed: one with PA behavior after 1 month, and a second with PA behavior after 3 months as dependent variable. Baseline variables were included as predictors in three steps. In step 1 we entered age and gender, in step 2 perceived pros, perceived cons, social norms, social modeling, self-efficacy and intention, and in step 3 implicit attitudes as predictor. For hypothesis 2, there was a fourth step, entering all interaction terms between implicit attitude and the explicit cognitions. If there were significant interaction terms, follow-up stratified analyses were conducted [57]. In this case, implicit attitude was categorized into positive, neutral, and negative based on the tertiles of its score distribution. Implicit attitude scores ≤ − .053 were categorized as negative, implicit attitude scores > − .053 and ≤ .285 were considered neutral, and scores > .285 as positive. To test whether the interactions found added significantly to the prediction of PA after 1 month or after 3 months, another hierarchical regression analysis was performed, only adding the significant interaction terms. To test hypothesis 3, hierarchical multiple regressions, similar to those carried out for question 2, were performed, but this time with intention at baseline, after 1 month and after 3 months as dependent variable. In step 1, we again entered age and gender; in step 2, perceived pros, perceived cons, social norms, social modeling, self-efficacy and implicit attitudes; and in step 3, all interaction terms between implicit attitude and the explicit cognitions. All predictors were mean-centered before entering into the models. Cases with missing values were not included in the analyses.