The TPB proposes that human behaviour is guided by three categories of beliefs: behavioural, normative and control beliefs. Behavioural beliefs are about the perceived consequences of the behaviour. These beliefs influence one’s attitude towards the respective behaviour either positively or negatively. Normative beliefs are about expectations from other people, resulting in perceived, subjective social norms. Control beliefs are about the presence of factors that may facilitate or inhibit intended behaviours, giving rise to perceived behavioural control. The result of these three categories of beliefs combined, that is, attitude towards the respective behaviour, subjective norms and perceived behavioural control, leads to the formation of a behavioural intention.
Our application of the TPB followed the five consecutive stages described in the manual by Francis and colleagues (2004): (1) definition of the behaviour of interest; (2) identification of participants and context; (3) instrument development: TPB questionnaire item generation; (4) data collection; and (5) statistical analysis (38).
1. Definition of the behaviour of interest
Using the TACT (Target, Action, Context and Time) principle, we defined the behaviour of interest as: performing direct real-time observations of trainees and providing feedback by supervisors during workplace-based medical residency training (39).
Identification of participants and context
Participants were General Practice (GP) supervisors from two GP specialty training institutes in the Netherlands (Maastricht and Leiden). During years 1 and 3 of the three-year postgraduate GP training programme, trainees spend four days per week in general practice where a GP supervisor monitors, coaches and assesses their competence development.
2. Instrument development: TPB questionnaire item generation.
Our 56-item web-based questionnaire (Additional file 1) consisted of three parts, namely a general introduction that contained the definition of the behaviour of interest, five items on demographic variables and 51 TPB statements. In line with recommendations, questionnaire items addressed respondents’ attitude, subjective norms and perceived behavioural control (Figure 1) both directly, by asking questions about their attitude, perceived norms and control in general, and indirectly, by asking questions about underlying specific beliefs (38, 39). Beliefs emanated from three audio recorded focus groups with in total 21 GP supervisors (38, 40). Data generated from the focus groups were transcribed verbatim and qualitatively coded with the aid of Nvivo software (41). According to principles of qualitative data analysis, three researchers (AT, MG, LJ) independently categorized codes into themes and used the belief categories of the TPB (i.e. behavioural, normative, control) as a preliminary coding framework. Discrepancies in the coding process were resolved through constant comparison and discussion within the research team(42).After that, themes were listed in order of frequency. These themes were used to generate questionnaire items in order for the final TPB questionnaire to cover 75% of the cumulative frequency of all beliefs that were reported in the focus groups (39). As recommended, the behavioural and normative beliefs were converted into two types of items; one set of statements about behavioural beliefs and corresponding behavioural outcome evaluations and one set about normative beliefs and corresponding motivation to comply (38). Congruent with previous research studies, for instance by de Vries et al. (43), we utilised Bandura’s measures of self-efficacy to operationalise control beliefs within the TPB questionnaire (43, 44).
Since extensive research has demonstrated that past behaviour has a residual effect on intentions after controlling for other TPB measures, we chose to include ‘past behaviour’ as a determinant of the intention to perform DOs (24, 45-47). Likewise, we included perceived social pressure (a perceived urge to adopt a behaviour) and modelling (seeing others perform a behaviour) as these factors may influence people’s intention to engage in the respective behaviour, extending the impact of subjective norms (48). With the addition of these extra measures we composed an extended TPB model to explore and predict supervisors’ intentions to engage in DOs of trainee performance in the clinical workplace (Figure 1).
All items were assessed on 7-point Likert scales and had defined anchors at the extremes (e.g. good-bad) (38). The only exception were the control beliefs, and as recommended by Bandura (2006), these were rated on a 100-point scale ranging from 1 (great uncertainty) to 100 (complete certainty) with 10-unit intervals (49).
All members of the research team pre-tested a preliminary 65-item questionnaire for clarity, understanding, applicability and feasibility. Items were rephrased and when necessary deleted until no new recommendations for improvement were given. Next, the 56-item questionnaire was pilot tested by 10 GP supervisors, following which two more items were rephrased.
3. Data collection
We used consecutive sampling from a list of 472 supervisors (all active GP supervisors of the institutes in Maastricht and Leiden) to obtain 200 potential respondents (38). They were invited by email to complete the web-based TPB questionnaire between June and October 2017. Non-responders received email reminders after two and four weeks. The data collection period ended two months after the questionnaire was first emailed.
4. Statistical analysis
Since one questionnaire form had two missing values, we replaced these with the respondent’s mean score for the remaining items within that measure. Where applicable, negatively keyed items were reverse coded to ensure that all items were in the same direction (38).
We calculated descriptive statistics for the following demographic characteristics: age, gender and years of work experience as a practising GP and GP supervisor (see Table 1). For each of the indirect and direct measures of ‘attitude’, ‘subjective norms’, ‘perceived behavioural control’ and ‘intention’, we calculated item-to-total correlations with the goal of eliminating items that were not related to the same measure. Following item elimination, we estimated the internal consistency (Cronbach’s coefficient α) of the direct measures; an α of >0.60 was considered as acceptable (38). We subsequently calculated the means and standard deviations of the composite scores regarding the direct measures of attitude, subjective norms, perceived behavioural control and intention (38).
We did not perform a reliability analysis of the indirect, belief-based measures as Ajzen (24) stated that internal consistency is not a necessary feature of these measures because ‘beliefs towards a behaviour can be ambivalent when a behaviour is likely to produce both positive and negative outcomes’. As the ‘control beliefs’ measure consisted of 12 items, and self-efficacy is, according to Bandura, a multifaceted concept (49), we performed an exploratory factor analysis to check for potential separate intercorrelated subscales. We used an oblimin rotation (delta=0) in order to optimise the interrelated pattern of factor loadings of the control belief items (50). According to the guidelines, the criteria for factor loading cut-offs were >0.5 (good), >0.6 (very good) and >0.7 (excellent) (51). Identified subscales were treated as distinct measures in the analysis. For the analysis of the other indirect, belief-based measures we defined composite scores: we weighted (multiplied) each behavioural belief by the corresponding score for outcome evaluation and each normative belief by the corresponding score for motivation to comply. Finally, we summed the weighted beliefs to create a composite score for the behavioural and normative beliefs respectively.(38). As the TPB model contains theoretical measures that are assumed to be interrelated, we also explored the relationships between the TPB measures of our questionnaire by computing bivariate correlations using Pearson’s R tests.
In order to assess the impact of the respective TPB measures as predictors on the intention to perform DOs, we did a hierarchical regression analysis (39). We checked the assumptions for linear regression analysis (linearity, independence, normal distribution and equal variance of residuals (51)). We subsequently calculated standardised beta weights to examine the contribution of the different predictors to the regression equation. As a first step, the demographic variables were entered into the model, followed by the indirect measurements of ‘attitude’, ‘norms’ and ‘control beliefs’ at step two (see Figure 1). In the third step, we added the direct measurements of ‘attitude’, ‘subjective norms’, ‘social pressure’, ‘modelling’ and ‘perceived behavioural control’. Finally, we entered the ‘past behaviour’ measure into the regression equation. We performed all statistical analyses using SPSS, version 25 (52).