2.2. The TPB questionnaire
The TPB is a theoretical framework but also comprises a set of guidelines to ensure proper measurement of the social-psychological constructs. A crucial step in applying the TPB is to define the behaviour in terms of its target, the action itself, the context in which it is performed, and when it is performed (the TACT principle). Then, to conduct a pilot study in which readily accessible behavioural outcomes, normative referents and control factors regarding the behaviour are elicited (Francis et al., 2004 & Sok et al., 2021). Our pilot study included semi-qualitative interviews with 20 dairy farmers during October-November 2018 (of them, seven dairy farmers were asked about vaccination against LSD). This gave us a set of underlying beliefs for each of the constructs (A, SN and PBC). The interviews, as well as the questionnaire construction, were based on previous studies (Francis et al., 2004; Sok et al., 2015 & Sok et al., 2016).
The questionnaire included three parts:
1. Questions measuring background factors. These factors were classified into farm and behavioural variables. The farm variables included questions on the type of the herd (family farm vs. single/double/triple cooperative farms), the herd’s yearly milk quota and the location of the herd. The behavioural variables included questions measuring perceived past experience and perceived risk as well as general questions regarding the gender, education and age of the farmer. Perceived risk was measured by the relative risk attitude and the risk perception. Risk perception was measured by multiplying two 7-point Likert-type scales, one with frequencies from ‘once in 100 years’ to ‘once in a week’ and one with the adjectives ‘No impact’ up to ‘High impact’. For measuring the relative risk attitude (with respect to animal diseases in general), each farmer was asked to compare his/her general risk perception to other farmers. This was performed by filling a 5-point Likert type scale that ranged from ‘absolutely disagree’ to ‘absolutely agree’, regarding four statements (Meuwissen et al., 2001). The TPB framework suggests that the background factors could explain variation in the intention. However, we examined whether these factors also have a direct effect on the intention.
2. Questions to obtain direct measures of the constructs of the TPB. A 5-points semantic differential scale with four different bipolar adjective pairs (e.g., unsatisfying and satisfying) was used to obtain direct measures of attitude. Injunctive subjective norm and PBC were measured with 5-point bipolar Likert-type scales with the endpoints ”disagree” to ”agree” (Sok et al., 2016). To prevent misinterpretation of questions regarding negative beliefs, two adjective pairs and one PBC statement were recoded so that higher numbers always reflect a positive attitude/self-efficacy/controllability to the target behaviour. Behavioural intention was measured by asking the farmer to scale his/her intention to vaccinate against LSD in the coming year between 1 (Will certainly not vaccinate) to 5 (Will certainly vaccinate). The survey only queried for measures of the injunctive norm.
Given the distinction between reflective and formative measurement, direct measures are required to show high internal consistency (see also Fishbein & Ajzen, 2010 & Sok et al., 2021). We tested the internal consistency of the direct indicator for each construct by using the Cronbach’s alpha scores. This analysis revealed a low internal consistency in the direct measures of the SN and PBC constructs (α=0.6, 0.2 respectively) (Table S1 ). The indicator Nd3, which was based on the question ”I feel I have social pressure to vaccinate my herd against LSD in the coming year”, had low within-construct correlations with the other two indicators (<0.3) and was not correlated with the intention indicators (Table S1 ). This can be explained by the disagreement of the farmers with the firmer wording of this statement as compared to the other two statements. Regarding the PBC, all three indicators were not correlated with each other. However, as opposed to indicators Pd1 and Pd2, the indicator Pd3, which was representing the capacity dimension, did correlate with the intention (Table S1 ). In addition, although the internal consistency was appropriate for attitude (∝C value of 0.8), only the indicators Ad1 and Ad2, which reflected experiential considerations (’necessary’, ’contribute’), had high within-construct correlations and were correlated with intention (Table S1 ). Overall, to solve for the low internal inconsistency, we assessed only the indicators which were correlated with the intention (Ad1 and Ad2 for attitude, Nd1 and Nd2 for norms, Pd3 for PBC). We calculated the mean of these indicators scores to give an overall score for each construct.
3. Questions to obtain indirect (belief-based) measures of the three social-psychological constructs (A, SN, PBC). The most frequently mentioned responses from the qualitative part of the study were used to formulate behavioural, normative, and control belief items. Consistent with the expectancy-value model, for each item the belief strength is measured at a scale of 5-point Likert type that ranged from ”not very likely” to ”very likely” for attitude’s beliefs, from ”very against” to ”very in favour for subjective norms’ beliefs and from ”not true” to ”very true” for PBC’s beliefs. Since the subjective norms’ statements had negative/positive meanings we used a bipolar scaling while we used a unipolar scaler for the attitude’s and PBC’s statements. The outcome evaluation/motivation to comply with pressure from the reference group/power of the control were measured at a scale of 5-point Likert type that ranged from ”very unimportant” to ”very important”/”not important” to ”very important”/”more difficult” to ”more easy”. This time, the power of the control received a bipolar scaling and the motivation to comply a unipolar scaling. In the analysis we changed the outcome evaluation scaling from bipolar to unipolar which was more logically fit (see Fishbein & Ajzen, 2010 & Sok et al., 2021). Certain behavioural and control belief statements with expected negative influence on vaccination were recoded as been described for the direct measurement. The resulting multiplicative products were analysed both individually and as an average.
The full questionnaire is attached to the supplementary materials .