There are two main paradigms to study consumer behaviour ( Bagozzi, 1993 ): the attitudinal approach, and the judgements and choices - or decision - appro­ach (henceforth, the judgement-choice approach). The former analyses consumer behaviour based on beliefs and behavioural intentions, being widely represented by the Theory of Reasoned Action - TRA ( Fishbein & Ajzen, 1975 ; Ajzen & Fishbein, 1980 ) and, the subsequent, Theory of Planned Behaviour - TPB ( Ajzen, 1991 ). The judgement-choice approach is based on the explanation of how products' attributes and prices determine the consumption level which maximizes the utility of decisions. This approach is underpinned on the classical utility economic theory of consumer behaviour (henceforth, classical-utility theory), and mainly on the Lancaster Theory ( Lancaster, 1966 , 1971 ). The theories from both approaches have evolved and included additional predictors ( e.g. , Povey et al. , 2000 ; Nicholson & Snyder, 2008 ; Fishbein & Ajzen, 2010 ; Del Giudice et al. , 2015 ), but they continue being criticized by their drawbacks.

Criticism for the attitude-behaviour or intention-behaviour gap ( Kraus, 1995 ; Hsiao et al. , 2002 ; Sheeran, 2002 ) describes an inconsistency between what consumers express and their purchase behaviour ( Kraus, 1995 ; Sheeran, 2002 ). According to Foxall (1993) and later McEachan et al. (2011) , a major weakness of models based on the abovementioned relationship is the only analysis of hypothetical pre-behavioural events. Related to the classical-utility theo­ry, it is also well-known both the divergence between monetary assessment or what is called economic costs and predicted utility ( Thaler, 1985 ; Kreps, 1990 ), as well as the failures to maximize the utility of the choices ( Kahneman & Thaler, 2006 ). Amir et al. (2008) state this may be caused by the absence of experience cues in the modelling. Bagozzi (2000) remarks the Classical-utility Theory fails by not considering aspects such as, for example, attitude, subjective norms or emotions. The goal-based choice models ( Fischer et al. , 1999 ; Van Osselaer & Janiszewski, 2012 ), based on the judgement-choice approach, also reflect part of this issue. Those assume that choices are driven by goals, which are cognitive dimensions, characterized by positive and negative forces linked to behavioural outcomes ( Van Osselaer & Janiszewski, 2012 ).

Therefore, the explicative power of both approaches may present shortcomings. The aim of this study follows this thought, trying to assess the potential of a hybrid approach to unravel consumer behaviour in agri-food markets. Concretely, the hybridization of the attitudinal approach, by means of the TRA, and the judgement-choice approach, by means of the classical-utility theory, is explored, which represents a novelty in the literature where those attempts are truly scarce for foodstuffs such as Myrland et al. (2000) and Olsen (2003) . We bring evidence on the differences between explaining behavioural intention and actions, and also research intention as driver of behaviour. The empirical application was performed using a daily product under volitional control such as olive oil, collecting data from Southern Spain consumers, and variance-based structural equation modelling by means of the Partial Least Squares (PLS) technique.

The measurement model's statistics for M1, M2, M3 and M4 are summarized in Table S2 [suppl.]. SN, SE and TT showed issues in their composition. However, we kept all the observable-variables for theoretical reasons, given that they were significant in at least one of the models.

In Table 2, the hypotheses' results are presented for M3 and M4, while Table S3 [suppl.] displays the M1 and M2 results, given that the analyses of the former are our main aim. Table 3 shows path coefficient differences between M3 and M4 and the correspon­ding significance test.

Subjective norms did not significantly influence the intention to purchase (H1 _M3 ) and actual consumption (H1 _M4 ), and there were no significant differences bet­ween both parameter estimates (H8). However, po­sitive attitude raised both intention (H2 _M3 ) and consumption (H2 _M4 ), with a similar effect size (f ² ). The similarities in the impact were confirmed by the test for the differences between the path coefficient estimates (H9). It is also worth noting that socioeconomic factors influenced intention (H3 _M3 ) and consumption (H3 _M4 ); both effect sizes are small. The relationship was positive, as the sign showed, so higher household incomes, buyers' age and having lived in rural areas improved the intention, while consumption increased in households with higher incomes, smaller size and whose members had lived in rural areas. Once more, there were not significant dif­ferences between parameter estimates (H10). Results also allowed us to confirm that the better the positive perception of products' attributes, such as price and taste, the higher the positive attitude (H4 _{M3, M4} and H5 _{M3, M4} ), intention (H6 _M3 and H7 _M3 ) and consumption (H6 _M4 and H7 _M4 ). It is relevant to highlight the medium effect of taste on attitude and the existence of significant differences in its impact on the intention and consumption, so H12 was supported. Nevertheless, we also expected price to exert a stronger and significant influence on consumption compared to intention, which was not confirmed by the statistical comparison (H11).

The intention and consumption hybrid models had high predictive power figures, with explained variances of 0.309 and 0.378 (adj- R ² ), respectively, compare to M1 and M2 (adj- R ² reached 0.253 and 0.273 respectively, see Table S3 [suppl.]). Thus, the gains in predictive power of the hybrid models are noteworthy. The explained variance accounted for the attitude resulted in 0.177 (adj- R ² ) in M3 and M4, which reveals a sufficient predictive power for a consumer behaviour study according to Falk & Miller (1992) .

Last, the results confirm that stated intention influences actual consumption (H17), which is not a surprising finding, but with only a medium effect size and an adj- R ² of 0.173.

The first significant finding is the role of subjective norms not to shape either intention or consumption, while being a variable traditionally included by an attitudinal approach. Some studies designed under the umbrella of the TRA, e.g. , those by Thompson et al. (1994) and Armitage & Conner (2001) , show that subjective norms display a low explanatory power or even no significance on intention. This has led to critics on TRA and TPB, with some authors ( Chang, 1988 ) expressing the necessity of changing the relationship of subjective norms towards the attitudes' formation more than the intention, or re-conceptualizing it ( Armitage & Conner, 2001 ). In contrast, Trafimow & Finlay (1996) and Park (2000) depict two types of intentional behaviours: those which are driven by attitudes and those driven by subjective norms. Therefore, the lack of impact in our models, as well as the lack of significant differences between path coefficient estimates, leads to the need for more empirical studies. Nonetheless, for olive oil case, this result may be influenced by the potential transfer of the beneficial effect from the extra virgin olive oil category to the broad concept of olive oil considered without category differentiation, undermining SN's predictive relevance.

Attitude is said to have a higher influence than subjective norms ( e.g. , Towler & Shepphard, 1991 ; Thompson et al. , 1994 ; Armitage & Conner, 2001 ; Webb & Sheeran, 2006 ) and, in our case, it explains consumption with the same strength as stated intention. Sheppard et al. (1988) highlights that both attitude and subjective norms predict intentions better than estimations of behaviour. However, Kim & Hunter (1993) find a slight difference in the correlations between attitude and intention compared to attitude and behaviour, while Kraus (1995) finds attitude significantly influences future behaviour, but it only accounts for 14% of the variance in behaviour on ave­rage. In this line, Glasman & Albarracín (2006) assert that strong correlations between attitude and behaviour only occur when individuals are very involved in thinking about the object, have direct experience with it, receive information about it and think their attitudes are correct. The agri-food product analysed here may influence such results, given that consumers in Southern Spain have both direct experience with olive oil in general, and they can easily receive information about it; however, it should be taken into account that we are explaining actual behaviour not future one. Therefore, consumers can be making stimulus-based and memory-based decisions, since they are able to evaluate the presented product and then access to memory for additional information about products ( Van Osselaer & Janiszewski, 2012 ).

Socioeconomic factors are particularly used in the judgement-choice approach. Those factors prove to be equally important in order to explain our dependent variables. However, studies from the attitudinal approach, which tackle the relationship between the socioeconomic factors and the intention, are truly scarce ( e.g. , Olsen, 2003 ; Michaelidou & Hassan, 2010 ), and those that analyse the whole effect of them as a latent variable are even more unusual ( Kim, 2009 ; Rodríguez-Entrena et al. , 2013 ).

Price and taste explained attitude, intention and consumption. As far as we know price relevance is in line with the judgement-choice approach, being the cornerstone for explaining consumer behaviour from the classical-utility theory. Indeed, the analyses of how different products' attributes modify the willingness to pay for agri-food products ( e.g. Lusk et al. , 2005 ) are usual, and, in olive oil markets, price becomes a key variable in explaining consumer behaviour ( van der Lans et al. , 2001 ; Dekhili & d'Hauteville, 2009 ). In the relationship between price and intention, Tarkiainen & Sundqvist (2005) assert that price is a kind of perceived behavioural control variable which predicts attitude. In other literature such as Michaelidou & Hassan (2010) , price is found to be a driver of attitude, intention, and even consumption ( Tarkiainen & Sundqvist, 2005 ). Nevertheless, the effect of price in consumers' decision-making process is not a well-studied variable in the attitudinal approach, being frequently included in empirical models as a single statement composing a beliefs' latent variable. A similar outcome derives from taste; some scholars ( e.g. Dekhili et al. , 2011 ) bring it into play as a fundamental factor for food choices, even finding that it has greater weight than price in explai­ning consumption ( Ward et al. , 2003 ), yet it is mainly investigated from an attitudinal point of view and by scarce literature. Some stress the influen­ce of taste on the attitude ( Stafleu, 1994 ; Shepherd, 2001 ), intention to buy a foodstuff ( Thompson et al. , 1994 ; Fotopoulos & Krystallis, 2002 ) or frequency of consumption ( Chryssohoidis & Krystallis, 2005 ); however, it is included as a single observable variable in different beliefs' latent variables.

Finally, TPB postulate that, under volitional control, behavioural intention must be the predecessor and the single best predictor of behaviour. That statement is not fully met here. To a great extent, the intention significantly explains the behaviour, as is shown by different empirical research such as Povey et al. (2000) . However, the strength in the relationship chan­ges across studies, and the level of prediction can range from 24% to 70% (of correlation) for choices of foodstuff ( Sheppard et al. , 1988 ). Our result is placed in the low range of the predictive power, which may be caused by different issues. According to Davies et al. (2002) , TRA proposes measuring actual behaviour objectively, and they cri­ticize self-reported behaviour. Our data comes from a daily product under volitional control where consumption is measured using objective actions of consumption, which may explain the low explanatory power. Young et al. (1998) depict that intentions are biased in measuring actual purchasing, either under- or overestimating it, and purchase behaviour generalization should not be based on intentions. Therefore, the bias may be corrected by assuming actual consumption will be steady in the near future and may become a proxy of future behaviour. Indeed, Wood & Quinn (2005) point out that the ability of intentions to predict behaviour relies on both the frequency of behaving and the stability of the surrounded context.

In short, both theories contribute to explaining intention and consumption in a similar way; however, by the hybrid approach, the variance in consumption behaviour is better explained suggesting a better per­formance. Evidently, the explicative factors can be extended in order to include a higher number of aspects embedded in subjects' salient beliefs, but this implies losing parsimony. Obviously, explaining actual consumption may be different than explaining future consumption, but given the gap between intention and behaviour, when ever possible, it may be interesting to truly learn which and how the determinants of actual consumption perform. Consequently, in order to expand the knowledge about consumer behaviour in agri-food markets, it may be appropriate to integrate variables from different approaches but adopting adequate de­finitions, measurements and relationships between them based on the paradigms they come from. This will allow unravelling consumer behaviour in order to fully understand agri-food markets functioning and design and implement successful commercial strategies.

This is particularly relevant in the olive oil sector where quality is still a differentiation instrument with a considerable scope for improvement as a marketing tool. In this regard, this hybrid approach, assessing the whole spectrum of predictors such as subjective norms, attitudes, perception of products' features and socioeconomic factors, can help fighting against the mainstream trend of treating the olive oil as a commodity where extra virgin olive oil (EVOO) and refined olive oil (ROO) are seen as interchangeable products by consumers instead of being fully aware of the differences between them to counterbalance these differentiation failures.

Nonetheless, the hybrid approach developed here has some limitations. A more complete model can be developed taken into account variables such as normative belief and actual control, being the latter a main novelty in RAA due to the moderating effect between the intention and behaviour relationship. In addition, to analyse whether variables from goal-based choice models may capture better the behaviour process, the hybrid models can be also extended. Second, this research only tested a concrete foodstuff, olive oil, and for consumers living in a specific area, Spain. As a consequent, further empirical research is required extending to other agri-food products and geographical areas in order to get testable results about the main predictors in agri-food product choices.

The authors wish to thank two anonymous reviewers and the editor for valuable comments on an earlier draft.