2. Dual identities in forming trade preferences

In speculating on who becomes advocates and opponents to trade integration, the predominant literature has rested on an assumption that individuals form their trade preferences based on their occupations and subsequent economic considerations (e.g., Gabel, Reference Gabel1998; Scheve and Slaughter, Reference Scheve and Slaughter2001). Standard trade theories (e.g., Heckscher–Ohlin and Stolper–Samuelson models) focus on how the distributional consequences of trade produce domestic winners and losers, while theories stressing industry-level causes (e.g., the Ricardo–Viner model) as well as firm-level heterogeneities (Bernard and Jensen, Reference Bernard and Jensen1999; Bernard et al., Reference Bernard, Jensen, Redding and Schott2012) highlight the relative competitiveness of specific domestic industries or firms vis-à-vis the global economy and prospective trading partners. According to the former branch of theories, labour in capital-abundant countries will not support free trade. The latter branch of theories predicts that those working in uncompetitive industries or firms will not advocate trade liberalization.

Meanwhile, the micro-level approach has questioned whether material self-interest is solely responsible for shaping individuals' trade preferences. For instance, Hainmueller and Hiscox (Reference Hainmueller and Hiscox2006) claimed that educational attainment is an important factor, not because it impacts one's employment and income level but because education impacts one's exposure to economic theories about the benefits of trade. Furthermore, scholars adhering to sociotropic, rather than egocentric, logic contend that people's preferences are rooted in their concerns over what is best for the country. This orientation may emerge based on various causes, ranging from altruistic motivations, a willingness to sacrifice one's own interests for others (Elster, Reference Elster, Kolm and Ythier2006), to nationalist sentiment (Huddy, Reference Huddy, Huddy, Sears and Levy2013). Mansfield and Mutz (Reference Mansfield and Mutz2009) introduced sociotropic perceptions, along with out-group anxiety, as an essential factor in trade preference-formation while the relationship between the sociotropic view and public trade preference has been further investigated by numerous scholars (e.g., Rho and Tomz, Reference Rho and Tomz2017; Hearn, Reference Hearn2020). In keeping with these efforts to fine-tune and increase the sophistication of our understanding of public support for free trade, we highlight individuals' interests as both consumers and income-earners.

In fact, theoretical acknowledgement of the existence of both income-earner and consumer effects is not new (Cross, Reference Cross1993; Lury, Reference Lury1996; Slater, Reference Slater1997). In psychology in particular, rich literature has emerged on consumerism and consumer identity (e.g., Scitovsky, Reference Scitovsky1976; Schor, Reference Schor1999). However, to date, only a few scholars have addressed the multiple perspectives individuals possess with regard to free trade and specifically the importance of the consumer perspective (Baker, Reference Baker2005, Reference Baker2009; Goldstein et al., Reference Goldstein, Margalit and Rivers2008; Naoi and Kume, Reference Naoi and Kume2011, Reference Naoi and Kume2015). The predominant literature in international political economy has assumed that an individual's degree of consumption is determined by their level of income, and therefore, consumer benefits are endogenous to employment status and without independent causal force. For instance, theories stressing industry-level determinants assume that individuals' skill levels or their employment in specific industries ultimately determine income level, which is in turn a proxy for their overall economic welfare (Baker, Reference Baker2005). While this is undoubtedly true to a certain extent, research on the possible intervening effects of consumer identity has so far been lacking. Hence, this study aims to fill this lacuna by examining how the consumer-oriented perspective can be an important factor in the development of individual trade preferences.

3. Hypotheses: the consumer-oriented perspective in the Japanese context

Traditionally, Japan has been considered one of the most conservative countries with respect to governmental protections of job security, ensuring Japanese workers' financial stability via a lifetime employment system and seniority-wage system (Estevez-Abe, 2008). Although global market competition has transformed employment systems, these practices persist, especially among formal or regular employees. On a related matter, our original survey in Japan (2015) included a question on whether respondents agreed with the following statement: ‘As regular employees in Japan are strongly protected and cannot be fired easily, companies have been hiring more irregular workers. In the future, the protection of regular employees should be relaxed in order to advance the national economy and reduce inequality’. Surprisingly, only 30.39% supported this statement. In this respect, it seems reasonable to expect that the Japanese public prefers its traditional employment protections and job security. In other words, Japanese people are likely to feel threatened if they perceive open trade as a source of threat to their employment. Thus, our first hypothesis aims to confirm this inclination by testing the following hypothesis: The Japanese public opposes trade integration when their income-earner perspective is activated (H1).

In the meantime, Japan has endured long-lasting economic stagnation since the bubble burst in the early 1990s. This economic hardship has been exacerbated by Japan's declining industrial competitiveness in the global export market, with manufacturing workers facing increasing job insecurity. At the same time, Japanese people as consumers have enjoyed increased flows of inexpensive goods from developing countries, China in particular. Under such circumstances, Japanese consumers are considered to have become more price sensitive with the prolonged recession and the strong yen. This consumption behaviour fuelled a boom in discount retailing and a real decline in retail prices (Vogel, Reference Vogel1999). Abe's advisers even pointed out that Japanese consumers had become overly used to the declining prices of goods (Kingston, Reference Kingston2016, 233). Indeed, the Nippon Keidanren (the Japanese Business Federation; the voice of big business in Japan) emphasized benefits to Japanese companies and consumers when advocating the TPP (Jamitzky, Reference Jamitzky2015) while the former Prime Minister Abe highlighted consumer benefits as the primary attractiveness of the TPP among others (Abe, Reference Abe2013). Based on this evidence, we hypothesize that the Japanese public embraces the benefits of free trade when their consumer-oriented perspective is activated (H2).

Relying on these two opposing expectations arising from their dual identities, we are primarily interested in their intervening effects and the comparison of their relative strengths. It is important to investigate this intervening effect instead of simply confirming the independent impacts of the two perspectives because it seems unlikely that people will adhere exclusively to one perspective – either consumer or income-earner – in developing their views on free trade. In reality, it is more plausible that their perceptions are constantly constructed on the basis of multiple identities and therefore, their eventual preferences are an amalgamation of these intertwined perspectives. Following this logic, we posit that when the two perspectives are activated together, the consumer's (income-earner's) perspective will counterbalance the negative (positive) impact of the income-earner (consumer) perspective (H3).

Furthermore, economically fragile individuals face the most severe dilemma. This effect is more pronounced in a country such as Japan, which is experiencing a surge of low-paid irregular workers (e.g., temporary workers and part-timers), who now represent an unprecedented one-third of Japan's labour force (Katz and Ennis, Reference Katz and Ennis2007: 86). Common sense suggests that individuals whose income is relatively low or who experience job insecurity will oppose trade liberalization out of concern that their income will be lowered further or their jobs replaced. This will occur either through the equalization of labour prices between domestic and foreign labour markets or through domestic firms being pushed out of the market. This kind of anxiety seems to be particularly profound today in advanced economies (Hiscox, Reference Hiscox2006; Autor et al., Reference Autor, Dorn and Hanson2013). According to general equilibrium approaches, developed countries are more likely to engage in producing sophisticated and diversified goods and services in order to counter import competition. Since individuals with low incomes or job insecurity in these countries tend to hold low-skilled, labour-intensive jobs subject to foreign competition, they will be more negatively affected by imports. Ample evidence seems to support this logic. Berman et al. (Reference Berman, Bound and Griliches1992: 368–369) find that the negative effect of trade on low-skilled workers is visible across all sectors, not only those that face severe import competition. Cline (Reference Cline1997: 253) has also found that import competition increases domestic inequality by 5–15%.

Simultaneously, these economically fragile individuals are the main beneficiaries of increased imports due to lower costs of goods (Broda et al., Reference Broda, Leibtag and Weinstein2009; Handbury and Weinstein, Reference Handbury and Weinstein2015). A recent survey experiment conducted in Japan (Naoi and Kume, Reference Naoi and Kume2015) reveals that, especially in a country like Japan with small welfare programmes and high food prices, individuals experiencing higher job insecurity are more susceptible to the consumer perspective mainly due to their daily consumption patterns. Accordingly, consumer benefits tend to become more critical for these potential losers of free trade: The counterbalancing impacts of consumer and income-earner perspectives persist among individuals with economic difficulties (H4).

4. A survey experiment: design and operationalization

We conducted a survey experiment in Japan during the fall of 2015, with a sample of 1,744 respondents between the ages of 20 and 69. The survey was administered by Nikkei Research, and subjects were recruited from its registered monitors through an opt-out method. We selected Japan in 2015 for our experiment timing because this was the period when debates over the benefits of the TPP were heated and made frequent appearances in the media. Thus, Japanese people during this time were greatly exposed to discussions on the costs and benefits of free trade. As a result, the Japanese public experienced ups and downs in their support for the TPP, as mentioned earlier.

The greatest challenge in designing this experiment was avoiding a frame that conspicuously or forcefully manipulated respondents' perceptions in accordance with the given treatment. Instead, we hoped to implement treatments that will allow them to reach their own answers independently (without feeling that the survey was leading them to select a particular answer), as this sort of setting most closely reflects what they experience in their daily lives. Consequently, our approach in formulating questions for the respective treatments focused on agenda-setting rather than framing. The former primarily aims to provide audiences with access to information by presenting events or issues, whereas the latter involves designing the experiment so as to elicit a specific response from respondents. By avoiding framing questions, we attempted to allow respondents to independently reflect on price or employment rate change by giving them general information and leaving them to draw their own conclusions instead of explicitly summarizing the expected effects of trade.

Of course, this may weaken the potential effects of this experiment because the treatments may not work for some respondents who may make incorrect inferences from the information provided. However, in order to pursue our goal, we needed to avoid obvious framing. Moreover, people process information differently and possess differing propensities in adjusting their existing views. In order to preserve the effects of this natural variation, we designed a setting that naturally allows the respondents within each group to reconsider their perspective by asking three to six questions for each treatment. In order to preserve conditions hospitable to this variation, it was imperative to avoid cuing the respondents to the underlying motivations of the survey.

Based on these considerations, our experiment consisted of four treatment groups along with a control group as shown in Table 1. As a starting point, in order to gauge the independent effects of the consumer and income-earner perspectives, respondents in groups A and B were each presented with three questions. For group A, we asked questions intended to elicit respondents' consumer perspective by focusing on the expected effects of free trade on consumer prices: (1) [Do you think] we could obtain foreign products at cheaper prices if the market liberalized?; (2) If the market liberalized through free trade, what do you think the price of meat would be per gram?;Footnote ² and (3) If the market prohibited the inflow of imports, what do you think the price of a TV would be? For group B, we shared questions designed to stimulate citizens' perspectives as income-earners, focusing on the potential negative effects of free trade on employment conditions: (1) [Do you think] free trade brings about a negative impact on domestic industries and employment?; (2) If the market liberalized through free trade, what do you think the unemployment rate would be?; (3) If the market liberalized through free trade, what do you think the median income would be? For the second and third questions, posed to both treatment groups, we deliberately avoided using multiple-choice questions. We believe that requiring respondents to estimate, without prompting, what changes to consumer prices or employment conditions would be brought about by free trade maximizes the effectiveness of each priming by compelling respondents to engage more actively in the speculation process as they would under real-world conditions. Through this setting, we expect that respondents assigned to group A will express greater support (H2) while those assigned to group B will first become more pessimistic about free trade (H1).

Table 1. Description on treatment and control groups

Of course, it may be argued that the questions included in groups A and B risk interference from respondents' sociotropic views on the effects of trade. One may claim that the results obtained from our experiment are driven not by individuals' egocentric economic concerns based on those two perspectives but by their sociotropic views on national well-being. Nonetheless, it has been found to be very difficult to parse out the two mechanisms since they are often interlinked (Kiewiet and Lewis-Beck, Reference Kiewiet and Lewis-Beck2011). For instance, people may care about national prosperity since it benefits their own individual economic conditions in return (Kinder and Kiewiet, Reference Kinder and Kiewiet1981). In summary, it is highly likely that personal opinions about trade are generated by both individual- and national-level economic factors (Ellonen and Nätti, Reference Ellonen and Nätti2015), and therefore, this sort of priming itself makes an appreciable difference in citizens' attitudes towards trade policy. In other words, the specific pathways through which respondents reach these determinations do not ultimately impact the observed effects of our treatment as long as they both allow respondents to activate their identities as consumers and/or income-earners based on their everyday consumption and employment conditions.

Next, the treatments for groups C and D ask each group both the set of questions presented to group A and the set presented to group B, but in alternating order. Respondents in group C are first asked the questions evoking the consumer perspective, then questions priming the income-earner perspective. In contrast, those in group D first receive the income-earner treatment, and the consumer treatment follows thereafter. The core objective here is to gauge whether the consumer identity actually weakens the negative impact arising from concerns over employment or income level. A simple comparison between the results obtained from groups A and B only tells us whether these two different identities exist. This paper's primary interest, however, is their interactive relationship and how this translates into public support for trade liberalization. We speculate that the support level of respondents assigned to either group C or D will fall between those of groups A and B as the benefits they gain as consumers counterbalance their negative perception arising from employment concerns (H3).

The two opposing orders of priming are introduced due to both theoretical and practical considerations. Theoretically, people tend to have a limited memory capacity, and thus, it requires us to demonstrate why and how people remember certain items more than others. Studies on the serial position effect posit that items presented in the beginning (primacy effect) and at the end (recency effect) are generally remembered the best while middle items the worst when people observe a number of items or pieces of information (Ebbinghaus, Reference Ebbinghaus1913). Scholars have long sought to identify whether the primacy or recency effect prevails over the other (e.g., Welch and Burnett, Reference Welch and Burnett1924; Murdock, Reference Murdock1962; Li, Reference Li2010). Their findings still seem inconclusive yet have consistently confirmed the importance of both effects (Wiswede et al., Reference Wiswede, Rüsseler and Münte2007; Jones and Oberauer, Reference Jones and Oberauer2013). Therefore, it is important to consider which information people receive in which order so that we can identify whether this order makes a difference in their trade preference formation.

Practically, the order of activation of the two perspectives is also important as our main interest is to accurately identify the actual conditions under which individuals develop their trade preferences. People are exposed to a plethora of information, commonly through broadcast, print, or social media. Therefore, it is safe to speculate that their attitudes on trade integration are shaped by that information to a certain degree. Furthermore, that information is based on diverse perspectives, ranging from positive impacts (usually linked to consumer benefits) to negative effects (usually linked to employment concerns) of trade. Under these conditions, people orient themselves based on a reiteration of learning between the benefits and costs of trade. This repetitive practice is what we hope to capture in our study by altering the order of activation of consumer and income-earner perspectives. Specifically, this experimental setting allows us to test whether the primacy or recency effect holds more power in the development of individual trade preferences or whether the order matters at all.

The categorization of these groups ranging between A and D is treated as our main independent variable (TREATMENT). They are unordered categorically, and group E is assigned as a control group (base). After respondents receive their respective treatments, the question that serves as our dependent variable was presented to all respondents: Do you support liberalization of the market via free trade? (FREETRADE). Possible responses are as follows: (1) oppose, (2) somewhat oppose, (3) neither, (4) somewhat support and (5) support. The higher (lower) value indicates greater support (opposition) towards trade liberalization.Footnote ³ We deliberately asked about free trade instead of the TPP in particular in order to deduce the impact of conflicting identities of people in viewing international trade in general.

The remaining questions are related to the characteristics and propensities of the respondents that have been emphasized by past studies. Among them, two questions aim to test the impact of economic fragility. INCOME records overall income level of respondents' households while SECURITY asks about future prospects about their jobs. The former variable ranges across eight levels, ranging between ‘below 2,000,000’ and ‘above 14,000,000’ Japanese yen, a higher number indicating a higher level of household income. The latter variable measures whether respondents believe that it will be easy to get another job with conditions similar to the one they currently have. The answers are ordered categorically with five classifications, with a higher value representing the respondents' greater sense of positive prospects. These two variables are later applied to our sub-group analysis on individuals with economic difficulties (H4). They are widely adopted indicators in gauging people's economic status and speculating winners and losers (e.g., Mayda and Rodrik, Reference Mayda and Rodrik2005; Walter, Reference Walter2015; Nguyen, Reference Nguyen2017). Thus, based on these measures, we suspect that the counterbalancing impacts of consumer and income-earner perspectives will persist among respondents who earn low income or feel job insecurity. The additional eight questions ask about respondents' socio-geographic and political information, such as respondents' political affiliation, education level and gender. We asked these questions in order to conduct a balance check across the treatment and control groups. The detailed information on these screening questions and descriptive statistics is presented in Appendix Tables A1 and A2.

5. Statistical analysis I: overall treatment effects

Based on our first three hypotheses, we expect that, compared to the mean of the control group, the mean of group A will be more positive while that of group B will be more negative. In an effort to specifically capture the interplay between consumer and income-earner perspectives, we change the order of priming for groups C and D. Regardless of the order, we suspect that their mean values will largely fall between those of groups A and B, generating ones that are closer to the control group. In our experimental setting, the 1,744 respondents were randomly assigned to group A (351 respondents), group B (373), group C (355), group D (330) and the control group (335). The balances of demographics across the five groups can be found in Appendix Table A3.

Before comparing responses across the treatment and control groups, our experiment result demands confirmation on whether the treatments actually activated respondents' consumer and/or income-earner perspectives. This is because those that were administered the consumer or income-earner treatments were exposed to three questions while those that underwent double treatments were exposed to a total of six questions. This may have caused information overload, inadvertently generating confusion, indifference, or information dumping. In this sense, although the core purpose of our experiment design was to capture respondents' reactions after answering all the given questions and thus the actual response to each specific question is less important, it is nonetheless worth checking how they answered these questions.

As shown in Appendix Table B1, we first checked the mean score for each question. It seems that the majority responded in the expected directions. For instance, the third question in the consumer priming treatment asked respondents to guess the price of TVs once the market prohibited the inflow of imports. In total, 71.4% responded with prices that were higher than the current average price. Even among those exposed to both treatments, answers followed the expected directions regardless of the order of treatments. Additionally, we also disaggregated respondents based on the number of questions they answered in the intended direction so that we could gauge the degree of understanding they possessed vis-à-vis consumer benefits or employment threats. We dichotomized responses so that 1 implies answering two out of three questions under the treatment as we expected (relevant perspective under operation) and 0 indicates answering only one or none of the three questions as expected (relevant perspective not operated). The results show that the majority of respondents understood both the benefits and harms that may be brought about by free trade while they seem to be more aware of consumer effects rather than employment effects. For instance, 81% of respondents answered two out of three questions in an expected manner in group A while 62% did so in group B. The detailed description as well as classification rules can be found in Appendix Table B2. Overall, it seems reasonable to claim that most respondents sufficiently understood the intentions of the questions based on the consumer benefits and (or) employment threats arising from free trade.

Next, Figure 1 shows a descriptive analysis based on the distribution of responses in our dependent variable (FREETRADE), by the treatment and control groups (TREATMENT). The vertical axis captures responses to FREETRADE across its five answer choices. The horizontal axis shows the percentages of responses to each answer choice within each treatment group, and the dotted lines represent standard errors. When we compare responses across answer choices at a glance, it is noticeable that the answer choices, ‘somewhat support’ and ‘neither’, are most widely selected, followed by ‘support’, ‘somewhat oppose’ and ‘oppose’. This straightforward observation suggests that not many Japanese people in general are opposed to free trade. Intuitively, even respondents who are exposed to income-earner priming (groups B, C and D) tend to choose the answer category of ‘somewhat support’. This result indicates that a certain number of people at least moderately support free trade even when they are aware of the negative impacts that it can bring about to their economy and labour market.

Figure 1. Distribution of responses to trade liberalization.

Note: Dotted lines indicate standard errors.

When this figure is examined more closely, the noticeable patterns across these answer choices are that, first, pro-free trade answers (‘support’ and ‘somewhat support’) are least supported by respondents in group B while anti-free trade answers (‘oppose’ and ‘somewhat oppose’) are least supported by those in group A. This observation hints at the positive (negative) impact of consumer (income-earner) perspective (H1 and H2). Second, for relatively radical responses such as ‘support’, ‘somewhat oppose’ and ‘oppose’, the percentages of these response selections in groups C, D and control fall in between those in groups A and B. This pattern reinforces our theoretical expectation on the intervening effects of the consumer and income-earner perspectives (H3).

Although Figure 1 displays many interesting patterns, it is difficult to dismiss the fact that the differences across treatment groups are frequently very small. In fact, respondents assigned to group B seem to most clearly and consistently express their negative views towards trade liberalization. They most outspokenly express their views by least selecting the ‘support’ category while most actively selecting ‘somewhat oppose’ and ‘oppose’. In order to specify these tendencies in greater detail, we applied an ordered logistic regression model with robust standard errors. Table 2 summarizes the key findings, mainly coefficients and standard errors of the four treatment groups, based on various classifications of respondents (for actual regression results, please refer to Appendix Tables A4, B3 and B4, particularly M(1) and M(2) from each table). M(1) exhibits the treatment effects when all respondents are included in the regression analysis. M(2) and M(3) change the composition of observations by following the binary coding used in Appendix Table B2. The former model only includes respondents who answered two out of three questions in the expected direction for each treatment while the latter includes those who did not. We speculate that these two models will show the difference between those who are well primed and those who are not.Footnote ⁴

Table 2. Ordered logistic regression results

Notes: The table displays three separate ordered logistic regression models. They do not include control variables since the results do not make a difference. The detailed results can be found in Appendix Tables A4, B3 and B4; ***P < 0.001, **P < 0.01, *P < 0.05; robust standard errors in parentheses; control group (group E: 335 respondents) as base.

According to M(1) on the overall statistical results, compared to control group, group A yields positive impacts while group B does the opposite. The treatment for groups C and D also seems to have produced a negative impact but the scope is much smaller compared to that of group B. Most importantly, the coefficient of group B is the only treatment effect that turns out to be statistically significant. These results reveal that the income-earner perspective more strongly and significantly impacts individual attitudes towards international trade, echoing findings of previous studies (e.g., Scheve and Slaughter, Reference Scheve and Slaughter2001; Bernard et al., Reference Bernard, Jensen, Redding and Schott2012). This outcome may be due to either the respondents' stronger identity as income-earners in viewing free trade or a high uncertainty about the potential effects of free trade on their occupations, which increases people's overall anxieties as ‘uncertain losers’ (Naoi and Urata, Reference Naoi and Urata2013).

Next, we compare the effects of groups A and B between M(2) and M(3). The former model displays that both coefficients of groups A and B are statistically significant. This indicates that both consumer and income-earner primings are effective among respondents where primings were successfully activated. It also implies that the initial null effect of the consumer treatment from the first model is largely driven by the 68 respondents who did not answer the questions in the intended direction. M(3) precisely focuses on those 68 respondents. Despite the small number of observations, the result nevertheless shows a strongly significant and negative impact of the consumer treatment. It suggests that the consumer effect is much larger and detrimental when people do not recognize or appreciate consumer benefits. Similarly, compared to the overall results from M(1), the negative impact of the income-earner treatment becomes much stronger among respondents who foresee employment threats (group B in the second model) while those who are not properly primed express a similar level of support to that of the control group. In summary, these results confirm our hypotheses on the negative impact of the income-earner perspective (H1) and the positive impact of the consumer perspective (H2).

We now turn to the comparison between the effects of the treatments for groups C and D. We find that the results for both groups are statistically insignificant regardless of the categorization of respondents. In other words, while both the consumer and income-earner perspectives independently exhibit meaningful impacts, the combined treatments lead to a similar level to the control group in supporting trade liberalization.

While this result numerically confirms our expectation of the values of groups C and D falling between those of groups A and B (H3), we need to confirm that this result is indeed driven by the nullifying effect of the consumer perspective on the income-earner perspective. To observe the distribution of respondents in more detail, we again refer to Appendix Table B2, which categorizes respondents based on the type of activated perspective and their mean values of FREETRADE accordingly. Regardless of the order of priming, respondents in groups C and D reveal similar tendencies. Almost half of the respondents (46% in group C, 55% in group D) are aware of both consumer benefits and employment harms while one-third of them (32% in group C, 27% in group D) are only clearly aware of consumer benefits. Among the remaining 20% of respondents, two-thirds (14% in group C, 12% in group D) are only clearly aware of employment threats while the remaining one-third (8% in group C, 6% in group D) were not susceptible to any of the priming. The mean values for FREETRADE across these four classified groups depending on the activation(s) of primings, on average between groups C and D, are 3.45, 4.05, 2.85 and 3.02, respectively.

Put together, those who are only primed by the consumer treatment express the highest support for free trade while those who are only primed by the income-earner treatment express the lowest. The fact that the former group on consumer treatment involves a much larger number of respondents suggests that people tend to more easily recognize consumer benefits. Meanwhile, when both perspectives are successfully activated, the mean value becomes closer to that of the control group (3.58). Considering that this group involves the largest number of respondents, we can infer that the majority of respondents are aware of both the positive and negative sides of free trade. When their consumer perspective is activated, they will then more strongly advocate free trade while the opposite will occur when their income-earner perspective is activated. When both perspectives are equally activated, their support level falls between those of the two perspectives. The disaggregation of responses seems to suggest that the consumer treatment alleviates negative impacts arising from the income-earner treatment.

6. Statistical analysis II: conditional treatment effects

The next question is whether this effect will hold across the board such that the consumer perspective overcomes negative attitudes towards free trade among comparative losers in the domestic economy. Our conditional hypothesis (H4) speculates that the counterbalancing effect persists as respondents with lower incomes and/or a higher perception of threat to their employment will oppose trade liberalization more strongly; however, their protectionist attitudes will lose force once the consumer-oriented perspective is activated. In this sense, the interaction terms between the treatments and economic fragility will be either statistically insignificant due to an unperceivable difference between winners and losers or significant due to a stronger tendency among people with economic vulnerabilities.

In order to capture how individuals respond differently based on their employment status, we rely on two questions that the respondents answered in our survey. INCOME records overall income level of respondents' households while SECURITY asks about future prospects related to their jobs. We ran an ordered logit model with interaction terms (one interaction between treatment groups and INCOME, and the other interaction between treatment groups and SECURITY). To ease the interpretation of the results, we only graphically present predicted probabilities based on marginal effects as shown in Figure 2 (for actual regression results, refer to Appendix Table A4, particularly models 3 through 6).Footnote ⁵

Figure 2. Predicted probabilities by subgroups.

With respect to INCOME (upper graph of Fig. 2), it appears that respondents with high incomes are likely to express a greater favouritism towards free trade when they are exposed to consumer priming only (group A) or consumer priming following income-earner priming (group D). Their support levels decrease when they are exposed to either income-earner priming only (group B) or income-earner priming following consumer priming (group C). These results seem to indicate that high income-earners are more susceptible to recency effects: their responses are more heavily influenced by the last information they receive. Meanwhile, respondents with low incomes reveal a pattern that more closely resembles the overall treatment effects found in Table 2. Consumer priming (group A) increases pro-free trade attitudes the most while income-earner priming (group B) decreases them the most. The impacts of combined primings (groups C and D) fall between these two boundaries set by consumer and income-earner treatments.

The lower graph of Figure 2 depicts predicted probabilities based on the interaction between treatment groups and job security. One noteworthy observation is that income-earner priming (group B) most negatively affects respondents' opinions on trade liberalization for both those with and without a sense of job insecurity to a similar magnitude, while consumer priming (group A) alone yields greater positive impacts among those with job stability. Both combined effects (groups C and D) equivalently yield positive impacts similar to that of consumer priming among the respondents feeling secure about their jobs. The respondents feeling insecure about their jobs show a pattern that is similar to that among low-income earners. The impacts of combined primings (groups C and D) fall between those for groups A and B.

Nevertheless, it should be noted that confidence intervals of the aforementioned predicted values either based on income level or job insecurity overlap greatly across targeted subgroups (e.g., between high- and low-income earners or those with and without job insecurity), and therefore, it is difficult to claim that the classification of respondents based on their economic or employment conditions yields any meaningful difference. In other words, while most of these predicted values follow the general tendency (negative impact by income-earner priming, which weakens when combined with consumer priming), those with low incomes or job insecurity are not necessarily more susceptible to either consumer or income-earner priming compared to their counterparts.

The flip side of this null finding is that regardless of income level or job (in)security, the employment-oriented perspective significantly reduces support for liberal trade policy, compared to the initial values of the control group. However, when they are introduced to both the consumer and employment primings simultaneously, their significantly negative attitudes are considerably mitigated in comparison to when they are exposed to employment priming alone. These findings confirm our hypothesis on the resilient impact of the consumer treatment even among those whom standard models would expect to be firmly anti-globalization.

7. Robustness checks

In order to confirm these findings, we ran additional regression analyses. First, we changed model specifications by excluding control variables that turn out to be statistically significant in Appendix Table A4, which is used to generate Table 2 (model 1) and Figure 2, to confirm that the results are not biased by multicollinearities among independent and control variables (see Appendix Table C1). Second, for additional sensitivity analysis, we altered model specifications by inserting the prefecture variable into the model instead of treating it as a fixed effect. To do so, we logged prefectural level agriculture and forestry revenue (Statistics of Japan 2015) for each prefecture, labelled as AGRICULTURE (see Appendix Table C2). Third, we recoded our dependent variable (FREETRADE) from five ranges (1 oppose, 2 somewhat oppose, 3 neither, 4 somewhat support, 5 support) to three categories: (1) oppose (combining ‘oppose’ and ‘somewhat oppose’), (2) neutral (replacing ‘neither’) and (3) support (combining ‘support’ and ‘somewhat support’) (see Appendix Table C3). Fourth, we applied an OLS regression model instead of an ordered logit model (see Appendix Table C4). These moderations of the original model do not change the implications, hence supporting the robustness of the main findings.