Deutsch or Denglisch –
How much English is too much?
A study into the effectiveness of code-switching
in combination with involvement
by
Michelle Hecker
University of Groningen
Faculty of Economics and Business
ABSTRACT
Foreign language use occurs frequently as a persuasion tactic in advertising in many countries. In some cases this takes the form of code-switched slogans, where a few words of the foreign language are embedded into the slogan so that consumers may attach the valence of the associations that this language use evokes onto the brand or product. This tactic has been shown to be effective for its aims, but its effectiveness is also moderated by several factors such as language proficiency. Given the status of English as an international language, it is often used in advertising, mixed with the local language. This study was conducted among a German audience, evaluating the extent of code-switching (i.e. how much English is used) under different product involvement conditions, namely high and low product involvement. Its aim was to investigate when English language use in German slogans operated mainly under a symbolic route or when its use required proficiency with English for slogans to be effective. Thus, two models were proposed: low involvement should lead to peripheral route engagement and thus a situation in which symbolic associations would drive persuasion; high involvement should lead to central route engagement, hence the message would be important, and thus English proficiency would be as well. While the effects proposed were supported to some extent, several caveats are noted as to their implications and the conclusions one can draw from them.
Key words: code-switching, English, German, (product) involvement Research theme: Persuasion tactics (code-switching)
TABLE OF CONTENTS
1. INTRODUCTION ...4
2. THEORETICAL BACKGROUND ...6
2.1 CODE-SWITCHING AND ITS EFFECTIVENESS ...6
2.2 PRODUCT INVOLVEMENT ...11
2.2.1 LOW INVOLVEMENT –SYMBOLIC ASSOCIATIONS RESPONSE ...12
2.2.2 HIGH INVOLVEMENT - MODERATION BY ENGLISH PROFICIENCY ...15
2.3 ATTITUDES TOWARD AND INFLUENCE OF ENGLISH IN GERMANY ...18
3. METHOD ...20
3.1 PARTICIPANTS ...20
3.2 DESIGN ...20
3.3 MATERIALS ...21
3.4 PROCEDURE ...24
4. DATA ANALYSIS PLAN ...26
5. RESULTS ...26
5.1 ORDER EFFECTS CHECK ...27
5.2 MANIPULATION CHECK ...28
5.3 REGRESSION ASSUMPTIONS CHECK ...30
5.4 HYPOTHESIS TESTING ...31
5.4.1 MEDIATED MEDIATION ANALYSIS -LOW INVOLVEMENT MODEL (ICE CREAM) ...31
5.4.1.1 BRAND EVALUATIONS (MEDIATOR 2) ...32
5.4.1.2 PRODUCT EVALUATIONS (MEDIATOR 2) ...37
5.4.2 MODERATED MEDIATION ANALYSIS –HIGH INVOLVEMENT MODEL (CAR) ...42
5.4.2.1 BRAND EVALUATIONS AS MEDIATOR ...43
5.4.2.2 PRODUCT EVALUATIONS AS MEDIATOR ...49
6. DISCUSSION ...55
6.1 THEORETICAL IMPLICATIONS ...55
6.1.1 LOW INVOLVEMENT ...56
6.1.2 HIGH INVOLVEMENT ...58
6.2 PRACTICAL IMPLICATIONS ...61
6.3 LIMITATIONS &IDEAS FOR FUTURE RESEARCH ...63
7. CONCLUSION ...69
8. REFERENCES ...70
1. Introduction
Vorsprung durch Technik. Even if you do not understand German, you may understand this
slogan or may remember it from Audi advertisements. You will know it speaks to the technical qualities of the car, because German and Germany are often associated with technical expertise, quality, and efficiency. These things Audi hopes will carry over to consumers’ evaluations of its brand, since this is the image the company is trying to craft. Use of languages congruent with the product’s origin is a popular marketing tactic and has been shown to be effective (Hornikx & van Meurs, 2017; Hornikx, van Meurs, & Hof, 2013). It relies on basic cues to do with the country associated with the language, usually in the case of languages other than English, or just associations with the language itself in the case of English, given its status as an international language (Pillar, 2001; Hornikx, van Meurs, & Starren, 2007). For instance, advertisements or product labels for French cheese and wine are expected to be in French and will be more effective if they use French rather than English names and slogans, since these products are generally associated with France (Hornikx et al., 2013), especially if they do in fact come from that country. Similarly, English is also used in advertising in many non-English- speaking countries.
With the rise of globalisation, the multi-national companies, and the mixing of cultures and languages across national borders, marketing and advertising have adapted by using English slogans across the world, particularly for global brands. In some cases this is done by using a full English slogan or by employing tactics such as code-switching, the mixing of languages within a slogan (Bishop & Peterson 2010, Luna & Peracchio, 2005a). The effect desired is for the language to become noticeable and thereby evoke favourable associations that will transfer onto evaluations of the object of the advertisement (Hornikx et al., 2007; Kelly-Holmes, 2000).
code-switching as a persuasive tool in advertising? These are questions this study will hope to answer in a German context.
Market research done by German marketing agency Endmark has shown that many Germans do not understand simple English advertising slogans. For instance, they translated the slogan employed by the German perfumery Douglas “Come in and find out” to “Komme rein und finde wieder raus” (Kaiser, 2004, citing Endmark, 2016), which in English literally means to come in and find your way back out again, not an ideal meaning to convey to potential customers. Such findings prompted several German and global companies to change their previously English slogans to German for the German market (Kaiser, 2004). Therefore, it has been suggested that English is on its way out in German advertising, as far as slogans go (Kaiser, 2004), and in terms of how influential it really is (Heine, 2014). However, culturally and with other forms of advertising, there is still English use via code-switching. Furthermore, Pillar (2001), Gerritsen et al. (2007, 2010) as well as Stefanowitsch (2002) and Hilgendorf (2007) and articles in the popular press (Heine, 2014) have recorded that the influence of English in German advertising as well as in culture, has increased over the years. Yet, even though some have recorded comprehension to have also increased (Gerritsen, 2007), slogans are still being misunderstood, as evidenced above. In addition, it has been suggested that the use of English in a non-English language among non-native speakers may take on different meanings compared to how it would be used or understood by native speakers (Stefanowitsch, 2002). And in 2018 one can still see German companies, such as Audi and Aldi, using full or modified English words in their advertising. Perhaps because, sometimes, despite getting lost in translation, advertising that uses English is still perceived positively by its German audience (Endmark, 2016). Interestingly, the same Endmark (2016) study found that easy to understand slogans were viewed as slightly boring whereas slogans that involved a play on words and were often not fully understood, were in fact perceived as more interesting.
All of this could be potentially problematic for companies using English in their advertising and thus warrants further research into whether these tactics are truly effective and what potentially limits their effectiveness.
(2001) and Kelly-Holmes (2000), to my knowledge not a great deal of empirical research exists on testing the effects of varying levels of code-switched advertising (i.e. varying the amount of English in one slogan) in combination with English proficiency and product involvement. Furthermore, many of these studies have noted that while the underlying theory may apply to any culture or language studied, the positive or negative effects found in their results may not, making it all the more important to extend research in this field to include different languages and countries or populations.
This study aims to add to the existing research and provide some insight into whether varying levels of English proficiency among a German population of consumers lead to different levels of effectiveness of code-switched advertisements (slogans) that have varying amounts of English in them, and whether the mechanisms whereby this effect takes place are affected by product type, in terms of high and low involvement with the product.
2. Theoretical Background 2.1 Code-switching and its effectiveness
Code-switching is the “alternation between two or more languages in the same piece of conversation” (Bishop and Peterson, 2010). Luna and Peracchio (2005a) narrow it down to “the insertion of a foreign word or expression into a sentence (e.g., into an advertising slogan), resulting in a mixed-language message”. This study will utilise code-switching within a sentence (intrasentential), the sentence being in the form of an advertising slogan. In line with the Matrix Language Frame model by Myers-Scotton (1995) the language that determines the syntactic structure of the code-switched sentence is referred to as the matrix language with the code-switched language being the embedded language (Luna, Lerman & Peracchio, 2005). The rules of code-switched message production outlined by this model also influence how the message is perceived. This, however, differs depending on how the message is processed, i.e. more data or more conceptually driven (Luna, Lerman & Peracchio, 2005).
(Grosjean, 1989). Some words are quite easily identified as language borrowed words, particularly technical language in domains such as computing and certain business expressions. Many Germans will, for instance, use the word ‘computer’ in everyday language as if it were a German word. Whether other words are used to an extent where they could be considered borrowed could also depend on the English proficiency of the population, and integration of words into the lexicon depends on many varying factors (Zhiganova, 2016). Thus, it may not always be clear what could be considered a borrowed word or code-switched language, and whether all people have the same views.
Foreign language use in advertising is common the world over as evidenced by the different cultures studied in this field, some specifically about code-switching, others simply comparing full English versus local language slogans. But is it always effective? This has largely been the aim of many studies in this field to document the processes by which foreign language use in advertising leads to advertising effectiveness, i.e. to uncover the underlying process and discover what conditions may limit or enhance the effectiveness of its use.
The crux of code-switching’s effectiveness seems to lie in its ability to evoke potentially desirable associations with the foreign language being switched to (Kelly-Holmes, 2000, Pillar, 2001; Hornikx & van Meurs, 2017; Luna & Peracchio, 2005b). The valence of these associations then transfers to brand and/or product evaluations - positive associations leading to positive evaluations and negative associations to negative evaluations (Luna and Peracchio, 2005a, b; Pillar, 2001). In assessing the processes that lead to code-switching’s effectiveness and in addition to the Matrix Language Frame model described earlier, researchers in this field also refer to Myers-Scotton’s Markedness Model to explain that code-switching makes the embedded language “marked” or salient and thereby leads to elaborations on the slogan related to associations made with the marked piece of language (Luna and Peracchio, 2005a, p. 761, b).
producer of many fine wines and other gastronomic delights, the French wine is the obvious pick. Ignoring any effects that pricing of luxury goods may have, a large factor involved in the purchase decision of such a product would be something called a ‘country-of-origin effect’ (Hornikx & van Meurs, 2017), something induced by the French names and terms on the label. The use of the foreign language has similar effects to explicitly mentioning the country of origin and through this effect can lead to associations, attitudes, and intentions (Hornikx and van Meurs, 2017), and will be more effective if the product and language used are congruent (Hornikx et al., 2013).
While this example had to do with product labels, foreign language use in slogans and advertisements in general would show a similar effect (Hornikx & van Meurs, 2017). Naturally then, different languages have different effects on appreciation of advertisements and therefore their persuasive effect, depending on how positive or negative the associations evoked with them are (Hornikx et al., 2007). Furthermore, if the language code-switched to is considered a minority language, i.e. the “language spoken by the group that holds less power and prestige” in the country versus a majority language, which belongs to the group with the “political, cultural and economic power” (Luna & Peracchio, 2005a, p. 44), then this will be less persuasive than the other way around. However, this is dependent on the salience of the word in the slogan and the associations with the minority language, meaning that if these associations are positive, the initial negative effect observed may be reversed (Luna & Peracchio, 2005a).
The idea of foreign language use leading to associations which influence advertising and brand and/or product evaluations, follows mainly a symbolic response logic, suggesting that its use evokes symbolic associations with the language or underlying country and culture, regardless of whether the text is understood (Pillar, 2001; Kelly-Holmes, 2000).
the spectrum (Stefanowitsch, 2002), let alone the same associations regarding what may be favourable to the advertisement. However, depending on the level of positive associations with the English language in a country, one might also expect that monolinguals or people with lower English proficiency levels, but still able to recognise the language being used, would also make positive associations with an advertisement using English and thus the brand or product being advertised, even if they do not fully understand the meaning of the English word or the context (Kelly-Holmes, 2000). This is in essence the reasoning behind the symbolic theory of language use in advertising.
While not many studies have looked into exactly what the associations with English are (and this may also depend on who is asked, which country or population of people) English has been said to be associated with “globalism, modernity, and prestige” (Hornikx et al., 2010, p. 173). Given that English is the international communication language, and it is reasonable to assume it would connote globalism, internationalisation, and modernity world-wide.
However, other studies have found that comprehension does play a role and can affect appreciation of the advertisement, and that the comprehension and appreciation of the advertisement are positively related, meaning that one cannot rely solely on the symbolic view of language use in advertising (Hornikx et al., 2010). Nonetheless, they don’t discount it entirely, given that they did measure differences in preference for the English versus, in that case, a Dutch ad in the ‘easy to understand slogan’ condition, suggesting that there must have been some positive associations evoked through the use of English.
However, English use may not always lead to more positive effects compared to local language advertising, made even more likely in cases where comprehension is low. Gerritsen et al. (2010) found that (in the Netherlands, Belgium, France, Germany and Spain) English did not lead to a more favourable product image or higher price image compared to local language advertising, as suggested in the oft-cited reasons marketers had for using English in their advertising. The same study also found that English was not as well understood as previously thought and that English advertising was more likely to be misunderstood than local language advertising. In their study, Gerritsen et al. (2010) call for further research investigating the effects of English use in advertising, among people less proficient in the language, since they assumed that their highly-educated respondents had high English proficiency.
effects compared to the local language, which they deemed to indicate that consumers are used to its use in advertising (Gerritsen et al., 2007). The study also notes that English use in advertising is on the rise, along with comprehension.
Clearly, foreign language use in advertising is not only an interesting topic with a highly relevant application in a globalised world, but also a complex topic with no one single answer when it comes to designing the most effective slogan or advertisement. Proficiency with the language or comprehension of the ad is only one of many possible moderators that may affect the effectiveness of code-switched advertising. It may also be that for certain products or brands, a code-switching strategy makes sense since it is better received, for instance for multinational companies or luxury versus necessity products, where differences in advertising effectiveness can be observed between English and local language advertising (Krishna & Ahluwalia, 2008).
Further to this point of ensuring a good fit, language use that is congruent with the consumption context portrayed in the advertisement, makes it more likely for the advertisement to elicit the “thoughts about family and friends” that are said to drive the positive effects of native language use in ads targeted at bilinguals, given that these thoughts can be considered similar to self-referent thoughts which lead to higher elaboration and can therefore lead to greater persuasion (Noriega & Blair, 2008). And, as mentioned earlier, ensuring that the language used matches the product advertised, in terms of the country it is commonly associated with, will also lead to more effective use of foreign languages in advertising (Hornikx et al., 2013).
This study, therefore, will look further into what conditions may determine when English takes a largely symbolic function in advertising effectiveness, or when its symbolic function becomes less important and instead English proficiency plays a role in determining effectiveness.
a high and low involvement product, for instance, a car or ice cream (Petty, Cacioppo, & Schumann, 1983).
In addition, this study hopes to discover if there are also differences depending on the amount of English used in a slogan for either the associations evoked (e.g. is there a greater effect with more English?) in a low involvement situation, or the interaction with English proficiency in a high involvement situation.
The next section will elucidate the general involvement, describing associated processes in more detail and their application to this study before deriving hypotheses for two proposed models.
2.2 Product Involvement
This study proposes two models for code-switched advertising based on two different routes to persuasion, which in turn depend on involvement with the product. Previous studies have looked into motivation to engage cognitive thinking and how this impacted language processing in bilinguals (Luna & Peracchio, 2002) as well as involvement with the advertisement and a person’s local or global identity in looking at the effectiveness of code-switched advertising (Lin & Wang, 2016). They found that internal and external motivation and involvement with the ad were strong moderators and mediators for its success.
This study attempts to discover whether perhaps levels of involvement with the product can lead to one form of response mechanism being used over another, i.e. the symbolic or affective response or a more functional, comprehension reliant response. This is proposed to happen via peripheral or central routes to persuasion being engaged, given the different levels of involvement, as per the Elaboration Likelihood Model (Petty et al., 1983).
According to the Elaboration Likelihood Model put forward by Petty et al. (1983), when people deem something important or relevant, for instance if they are highly involved with a product, they will be more motivated to assess it cognitively and form an evaluation of the thing at hand. They propose that this “diligent consideration” (p. 143) forms the central route to persuasion but that this also requires an ability to process the information. This is where proficiency in the English language comes in as a potential moderator of the effectiveness of the code-switched message leading to product and brand evaluations and positive or negative purchase intention.
attitude or behaviour as an indication of persuasive effectiveness of the code-switched message, however respondents did consider the code-switched message more carefully than the one without code-switching. The present study, on the other hand, will look into how existing involvement (with the product, thus ad), will impact the effect code-switched advertising has on persuasion (i.e. brand and product evaluations and purchase intention).
This leads to the following hypotheses:
Hypothesis 1: low à if product involvement is low, the peripheral route to persuasion will be
engaged, where heuristic cues will be employed to derive evaluations and reach persuasion.
Hypothesis 2: high à if product involvement is high, central route processing will be engaged
so that the persuasive effect of code-switching on purchase intention will be mediated by brand and/or product evaluations (as separate mediators).
Thus, under low involvement, the associations, operationalized through a language favourability measure, i.e. people’s positive or negative affective evaluation of the foreign language use, become the driving feature in deriving product and brand evaluations and thus purchase intention. This means that the associations people have with the English language and perhaps the underlying countries, become highly important.
This also means that the effects of any differences between slogans using either more or less English, may be negligible in conditions where English proficiency is largely irrelevant, but may be more impactful when it does play a role, for instance under a high involvement condition.
2.2.1 Low Involvement – Symbolic associations response
For a code-switched slogan advertising a low-involvement product (e.g. ice cream) processing fluency is posited to have less of an influence, due to a low level of involvement with the product initiating a peripheral route to persuasion. That is, people in this condition are more likely to rely on heuristics rather than engage in a focused processing of the meaning of the slogan. Therefore, English proficiency is less likely to be important and have an effect under this condition, i.e. for low involvement products.
in terms of the slogan is also low (Petty et al., 1983). Thus, the use of English will trigger either positive or negative peripheral associations with the language, which will then affect brand and product evaluations (and purchase intention) in the same direction as the valence of the association (Luna & Peracchio, 2005a, b; Hornikx et al., 2007).
Tentative support for this process is provided by Noriega and Blair’s (2008) study, which, while not able to conclusively say that mediation effects existed from language through thoughts to advertising outcomes, did find constituent effects, i.e. from language to thoughts, and from thoughts to outcomes, finding some positive effects from thoughts on attitude toward the ad and brand, and purchase intention. More definitive support is provided by Luna & Peracchio (2005b), who show that code-switching effectiveness works through evoking associations with the language, which are elaborated on, with the valence of these elaborations transferring onto brand and product evaluations.
Therefore, brand and product evaluations and purchase intention will act as proxies for the persuasive effect of code-switched advertisements in this study.
Hypothesis 1.1: positive (negative) language associations will lead to higher (lower)
brand/product evaluations and thus higher (lower) purchase intention.
Under a low involvement condition, adding more English to the slogan should not have any differential effects on brand and product evaluations, since elaboration likelihood for the slogan is low and people are considering the slogan and advertisement under a peripheral route. This means that the likelihood that people should exert more cognitive effort as English usage increases is low. Since their motivation to do so in a low involvement condition is already low, adding more English, which would require more cognitive effort, would thus only decrease motivation in this condition (Petty et al., 1983).
messages is less cognitively demanding than L2 messages and processing L2 messages takes more cognitive effort, meaning that when processing the latter, consumers are more likely to take a peripheral route to persuasion and rely on cues other than the language, for instance the visuals in the advertisement. The ease of the task can also interact with these effects, in that when the task was easy both people high and low in NFC were able and motivated to process it, suggesting that L1 ads would be recognised just as well by those high in NFC as those low in NFC.
Thus, the slogan with no English words (i.e. the full German slogan) should be similarly well received by both those in the high and low involvement conditions, in so far as German is their most fluent language, which is expected to be the case here.
For people in the low involvement condition, their motivation to process the code-switched message is low (Petty at al., 1983), and thus code-code-switched messages that include their L2 would require more cognitive effort to process than they might be willing to expend, thus reinforcing their use of the peripheral route to persuasion.
However, under high involvement, motivation to process the message and L2 words is higher (Petty et al., 1983), and thus correctness of the slogan will be slightly more important given that a central route to persuasion will be engaged (Luna, Lerman & Peracchio, 2005; Petty et al., 1983).
Others have also supported the notion that English words possibly function symbolically as peripheral cues but that little to no empirical research exists on this and therefore on what role the type of processing of the message may have in determining the effectiveness of English use in advertising (Van Meurs, Korzilius & Bergevoet, 2015). This study will hopefully be able to go some way in answering this. Indeed, some studies have already shown that under the peripheral route, linguistic correctness of the message becomes less important (Luna, Lerman, & Peracchio, 2005) and therefore one might assume that under a low involvement product condition actual comprehension of the message is a secondary factor when it comes to consumers being persuaded by the advertisement. Therefore, using more English in a slogan for a low involvement product would only make it more likely that a peripheral route to persuasion is engaged for both monolinguals and bilinguals.
proficiency does not matter as much. Its only role in this condition is predicted to be reinforcing the use of a peripheral route to persuasion for monolinguals due to their inability to understand and process the meaning of the message properly by using the words alone, and make no real difference in the case of bilinguals since they can process and understand both languages equally well but take the peripheral route due to their low product involvement.
Hypothesis 1.2: Under low involvement, increasing the number of English words in the slogan
(extent of code-switching) affects product/brand evaluations and subsequently purchase intention in the same direction as the valence of the associations, regardless of English proficiency, with no differential effects between code-switching conditions.
Thus, the following conceptual model is derived for the low involvement condition (Figure 1).
Low Involvement - Conceptual Model 1 (H1, H1.1, H1.2)
Figure 1
2.2.2 High Involvement - moderation by English proficiency
proficiency and comprehension should matter more in deriving product and brand evaluations from either code-switched slogan, the one low in English and the one high in English, with proficiency becoming more important as the amount of English in the slogan increases.
Luna, Lerman and Peracchio (2005) looked into the effects of linguistic structuring of the code-switched message and its persuasive effectiveness. While in the previous condition, the grammatical correctness of the message did not impact persuasiveness due to peripheral processing being at play, in a high involvement condition, a linguistically correct message is more persuasive than one which does not follow the “rules of code-switching” (Luna et al., 2005, p. 416), since a high involvement product evaluation could be considered a “highly data-driven mode” (p. 418). This means that people in the high involvement condition are likely to use bottom-up processing, focusing on the data and stimuli to arrive at an evaluation, as opposed to using top-down processing where an existing conceptual understanding aids in driving evaluations. Accordingly, one should see that more attention is paid to the words in the slogan in high involvement conditions where the consumer is more cognitively engaged.
Therefore, under the high involvement condition, the language associations are not predicted to be the driving feature in code-switching’s effectiveness. Instead they are predicted to bypassed, since respondents are more highly motivated to process the message rather than use peripheral cues in forming brand/product evaluations. This means that for high involvement products (e.g. cars), the message itself and thus processing fluency become more important since these products will be approached from a functional rather than emotional perspective. Thus, it is expected that the moderation effect of English proficiency will have a greater impact on the effectiveness of the code-switched slogan in this condition and is thus only shown in the high involvement model.
Following this logic, it would be expected that as the number of English words in the slogan or the extent of code-switching increases, English proficiency will become more important in deriving favourable evaluations of the brand and product advertised.
Hypothesis 2.1: under high involvement, hence central route processing, the effectiveness of
code-switched advertising is moderated by English proficiency for both the one and three English word(s) slogans, the importance of English proficiency increasing as code-switching increases.
the use or lack of code-switching is matched with a person’s global or local identity, respectively, that this leads to higher advertising involvement, and that this in turn allows code-switching to lead to favourable ad and product evaluations. They also suggest (citing Erling, 2007 and Roger, 2010) that global/local identity may be linked to proficiency in the language. This could make sense given that being highly proficient in English would likely be a result of a lot of contact with the language. Therefore, people who are originally from a non-English speaking country but proficient in English, may reasonably be assumed to have a more global than local identity and this will have facilitated their learning of and proficiency in the foreign language. Therefore, given Lin and Wang’s (2016) findings, that advertising that uses code-switching will be met with greater ad involvement and more favourable ad and thus product evaluations by people who are more rather than less proficient in the code-switched language (English), one could assume that bilinguals would have a more favourable attitude toward code-switched advertising and it would thus be more effective for them (i.e. lead to more positive ad and product evaluations).
In addition to being more motivated to process the slogan in a high involvement condition (Petty et al., 1983), those people high in English proficiency, perhaps even fully bilingual, are also more able to engage with the message since it comes closer to being on par with their first language in terms of fluency, meaning they may not need to completely switch over to the code-switched language. Cognitive or data-driven processing thus becomes easier and more likely when confronted with a slogan in that language (Bishop & Peterson, 2015; Luna et al., 2005). Their elaboration likelihood is therefore high and they are more likely to take a central route to persuasion (Petty et al., 1983). This means they would focus more intently on the slogan and on whether it makes sense and through this process be persuaded by the advertisement. Provided then that the slogan meets linguistic expectations, (i.e. grammar does not interfere with the effects), there should not be a material difference in brand and product evaluations between the different code-switching levels, (zero, one, and three English words) at high levels of English proficiency.
of advertising that uses foreign languages to the relevance theory by Sperber and Wilson (1995) (cited in Hornikx et al., 2010). This theory posits that if there is a mismatch between the effort required to understand a message and what is gained from it, this may lead to feelings of frustration.
Therefore, it is hypothesised that the persuasive effect of code switched slogans on brand and product evaluations and through this purchase intention, is moderated by English proficiency. That is, as English proficiency increases (decreases), so should favourable effects code-switched slogans on brand/product evaluations also increase (decrease).
Hypothesis 2.2: Under high involvement, the effect of the number of English words per
slogan (extent of code-switching) on brand and product evaluations and through this purchase intention, will vary positively with English proficiency.
Thus, the following conceptual model is derived for the low involvement condition (Figure 2).
High Involvement – Conceptual Model 2 (H2, H2.1, H2.2)
Figure 2
2.3 Attitudes toward and influence of English in Germany
et al., 2007), even though comprehension of the language and attitudes toward English and code-switched advertising are mixed (Stefanowitsch, 2002; Hilgendorf, 2007; Zhiganova, 2016). This makes Germany an interesting domain for this study and a legitimate field in which to examine whether code-switching is justified for a German audience.
German brands such as Audi use the occasional English word or a German adaptation of an English word, in their advertising. For instance, a 2017 campaign run by Audi for its ‘Black Edition’ cars uses these exact words along with the words “Welcome to Black” underneath otherwise German slogans (Breyer, 2017). And one of its advertisements even made use of code-switching in the slogan, “Understatement ist überbewertet”. Aldi, the German low-cost supermarket chain, used the word “snacken”, a modified version of the English verb ‘to snack’ in its 2018 advertising as part of the slogan “Natürlich snacken zum ALDI Preis” to promote healthier, more natural snack alternatives at lower prices (Aldi-Nord, 2018).
These product categories (car versus snack) and brands are very different from each other and could be assumed to reach a wide range of demographics, perhaps with some overlap between them, as well as elicit different levels of involvement. This seems to suggest that advertisers in various domains in the German market deem English use through code-switching to be an effective way to communicate to their consumers, and clearly assume that it leads to more favourable associations or sounds more attractive than using a German alternative. Either that, or perhaps the situation in Germany is such that the use of certain English words or hybrids may be so pervasive that it is not even noticed as a foreign language and thus not distinguishable from German by the average consumer. This could suggest that the amount of borrowed English words in German language is growing. Or, as this study hopes to discover, there may be differential effects arising through the specific type of product advertised and thus the need for advertisers to consider different factors depending on product type, when deciding to use English in their advertisements (e.g. English proficiency).
These factors can obviously vary between individuals, and the variation of interest here is in whether different levels of English proficiency have different (or no) effects depending on how much English is used in a slogan (extent of code-switching), and how involved the person is with the product being advertised.
3. Method
3.1 Participants
Responses collected totalled 186, however, after removing participants who finished an inadequate amount of questions or no questions, the final sample consisted of 152 people. Most of the participants were born in Germany (the mode for that yes/no question was 1, equating to “yes”). Most participants also indicated that they speak German at home, do not speak English at home, and do consider German and not English their native language. English proficiency for the sample was relatively high (M = 4.11, SD = 0.94) according to participants’ self-reports. This does not however necessarily mean that there isn’t a range in actual proficiency or comprehension in the sample and therefore there may still be differential effects of English proficiency. The average age group was group two, out of four possible age brackets (M = 2.03 SD = 1.20). Thus, the average age of the sample fell between 31 and 46. The average education level was between groups two and three, out of six possible answer choices (six being N/A) (M = 2.51 SD = 1.31). Thus, the average education level of the sample fell between technical college and Bachelor’s degree.
3.2 Design
two subgroups that would be viewing the two advertisements in a different order: first the high involvement (car) ad, then the low involvement (ice cream) ad, or vice versa. So, each respondent viewed two ads, but only ever in one order and only for one code-switched condition.
3.3 Materials
Scales that were not used in subsequent hypothesis tests (for several reasons elaborated on in Appendix 4) but included in the survey can be found in Appendix 4. Results of the reliability analysis of the scales used are shown according to code-switching condition and involvement condition. All were above the 0.6 threshold to allow the items of these scales to be summed and average scores derived for each respondent.
Manipulation of Involvement
The manipulation of involvement was modeled off a similar study performed by van Meurs, Korzilius, and Bergevoet (2015), who followed Petty and Cacioppo’s (1983) research. Participants in the high involvement condition were asked to carefully and critically evaluate advertisements shown to them and told that their evaluations were important for the study.
The product advertised in the high involvement condition was a car since purchasing a car could reasonably be considered to be a highly motivated and involved purchase. However, given that respondents are not actually considering purchasing a car or the car in the ad (this being very unlikely), the high involvement condition included verbal cues to induce involvement (as explained above). The low involvement product was an ice cream, with only a short, simple request to evaluate it but no verbal cues were added to further manipulate involvement. The fact that the brand would likely be unknown to most people could have added to a lower involvement.
the same brand evaluation scale as will be explained below. Nonetheless, given that Kāpiti is a New Zealand brand, it is unlikely that many of the German respondents will know it and have an existing evaluation of it.
The reason existing brands were used was to give the advertisements as much of an authentic feel as possible. In creating the slogans, the Matrix Language Frame model theory (Myers-Scotton, 2005 cited in Luna, Lerman & Peracchio, 2005) was followed. German was the matrix language, therefore, where possible the linguistic structure followed German rules, with English simply swapped in as the embedded language. This was done to ensure as much linguistic correctness as possible, so that this would not to interfere with the effects being studied, given Luna, Lerman & Peracchio’s (2005) findings that following the “linguistic rules of code-switching” is of greater importance in situations such as a high involvement condition.
Where possible, the slogans were constructed so their structure would be similar and their message would have a similar feeling and evoke a similar response, in terms of the emotion contained in them, i.e. both advertisements and messages tended toward experiential consumption, and allowing oneself the luxury of the Kāpiti ice cream, or drawing attention to the feeling of driving an Audi. This was done so that an unrelated emotion wouldn’t interfere with the effect of English in the slogan. The three different slogans for each advertisement can be found in Appendix 1 as well as in the survey Appendix 3 as part of the full advertisement.
Both products could be considered to speak to those in the middle-high income range, given their depiction in the advertisements, and their brands. However, Audi could also be considered a somewhat standard car in Germany, given its status as a first world country, with a largely affluent population, although it would obviously depend on the model, the one depicted in the ad used here is not a particularly high-end one. Ice cream, while here depicted as ‘designer/luxury’ ice cream, does not require the same financial investment as a car, hence its categorisation as a low involvement product in this study. Therefore, it would be expected that everyone taking part in this study could envision purchasing such a product/brand. Furthermore, according to Pillar (2001), bilingual or multilingual advertising is commonly associated with ideas that identify its inclusion in a “prestige system” (p. 173). Therefore, its use for products that also incorporate similar prestigious ideas (“internationalism, future orientation, success and elitism, sophistication, fun, youth, and maleness” p. 173) could be considered a congruent or fitting use of English.
English proficiency – The measure that was used in subsequent analyses was based on
participants consumed music, movies, newspapers/magazines, and TV; what language (English or German) participants used at home, at school, and with friends, and how proficient they were in English and German, at reading, writing, and speaking. Only the last question explicitly asking about proficiency was used to arrive at an average score (1-5) for each respondent for their evaluation of their English proficiency (Cronbach’s Alpha (a) = 0.93, M = 4.11 SD = 0.94). This was because it was considered a more direct measure than the other questions. Of the other questions mentioned, only those considered the most relevant were evaluated to form part of the demographic information described in the participants section above.
Language favourability – in order to get a clearer measure of participants’ evaluations
of the use of English (German in control), respondents were asked to indicate their overall feelings toward the use of English in the slogan/advertisement on two items, on a 7-point semantic scale ranging from “extremely unfavourable” to “extremely favourable” and “extremely negative” to “extremely positive” (taken from research by Krishna & Ahluwalia, 2008) (M = 4.36, SD = 1.56). This was then used as proxy for the valence of the associations made with English use in the slogan. (A: a_low = .96, a_high = .97; B: a_low = .94, a_high = .98; C: a_low = .92, a_high = .97).
Product evaluations – this question was modelled off research by Luna & Peracchio
(2005a), using six 5-point scaled items labelled “poor quality/high quality,” “not appealing at all/very appealing,” “I would not buy it/I might buy it,” “I would not recommend it to a friend/I would recommend it to a friend,” “mediocre/exceptional,” and “very bad/very good” (low: M = 3.13, SD = 1.05; high: M = 3.61, SD = .92) (A: a_low = .94, a_high = .93 B: a_low = .92, a_high = .95; C: a_low = .94, a_high = .93).
Brand evaluations – this was measured by adapting a scale from a study by Petrevu and
Lord (1994) (cited in Noriega & Blair, 2008). Four items were measured on a 7-point Likert scale (1 = “Strongly disagree”; 7 = “Strongly agree”). The items were “Buying an Audi/Kāpiti is a good decision”, “I think Audi/Kāpiti is a satisfactory brand”, “I think Audi/Kāpiti has a lot of beneficial characteristics”, “I have a favourable opinion of Audi/Kāpiti” (Low: M = 4.18,
SD = 1.24; high: M = 5.40, SD = 1.17) (A: a_low = .95, a_high = .92; B: a_low = .92, a_high = .95; C: a_low = .94, a_high = .93). Initial brand evaluations of both Audi and Kāpiti were also measured on the same scale.
Purchase Intention – this was measured with three items adapted from a study by
items were “It is very likely that I would buy the car/ice cream”, “I would purchase Audi/Kāpiti the next time I need a car/ice cream”, I would definitely try Audi/Kāpiti” (low: M = 3.85, SD = 1.67; high: M = 4.07, SD = 1.67) (A: a_low = .95, a_high = .94; B: a_low = .87, a_high = .92; C: a_low = .92, a_high = .94).
Manipulation check questions
Consumer Involvement Profile (“CIP”) – the modified CIP (Bearden, Netemeyer, & Haws,
2011) was used to measure the expectation regarding involvement with the two different products. The CIP consisted of two, three item 7-point Likert scales: Product class involvement and purchase decision involvement. The questions for product class involvement were: “cars/ice creams are important to me”, “for me, cars/ice cream do/does not matter”, “cars/ice cream are/is an important part of my life”. Purchase decision involvement: “I choose cars/ice cream very carefully”, “which car/ice cream I buy matters to me a lot”, “choosing a car/ice cream is an important decision for me.” After recoding one reverse scored item, these scores were combined and averaged to arrive at an average score for Consumer Involvement Profile for each participant (low: M = 4.30, SD = 1.34; high: M = 5.12, SD = 1.28) (product class A: a_high = 0.852, a_low = 0.862; B: a_high = 0.933, a_low = 0.858; C: a_high = 0.873, a_low = 0.841) (purchase decision A: a_high = 0.911, a_low = 0.867; B: a_high = 0.953, a_low = 0.919; C: a_high = 0.940, a_low = 0.882).
Advertisement involvement – to check whether the manipulation of involvement
worked and whether people were actually engaged in the task, a further involvement check was included. A three item, 7-point Likert scale (1 = “strongly disagree”, 7 = “strongly agree”) was used from research by Lin and Wang (2016). The items were, “I was involved with the advertisement content”, “I concentrated on the advertisement content”, “I paid attention to the advertisement content when viewing the advertisement” (low: M = 4.55, SD = 1.34; high: M = 4.71, SD = 1.28) (A: a_high = 0.780, a_low = 0.854; B: a_high = 0.798, a_low = 0.681; C: a_high = 0.847, a_low = 0.816).
3.4 Procedure
Respondents were sought through use of a snowballing procedure, through friends and family as well as relevant groups on social media.
wouldn’t interfere with the experiment. The survey was first produced in English and then translated to German, and edited by a second party to ensure accuracy in context. German Likert scale labels were based off research on surveys and typical labels (Villar, 2009), but adapted to ensure they sounded correct in every instance.
The survey started by asking demographic questions, age, and level of education as well as the questions on a person’s affiliation with the English language and whether or not they were born in Germany. Following this, they were asked a number of questions that aimed at ascertaining their self-assessed level of English proficiency.
Respondents were informed that they would be evaluating two advertisements for existing brands and they were thus asked about their existing attitudes toward these brands. If people indicated they did not know the brand, these questions were skipped. After this, respondents were randomly placed in a code-switching group, either the control condition (no code-switching, full German slogan (A), the one English word condition (B), or the three English words condition (C). There were roughly 50 people per condition, given the 150 people in total. Within these groups, respondents were once again randomly split, to get two subgroups that would be viewing the two advertisements in a different order: first the high involvement (car) ad, then the low involvement (ice cream) ad, or vice versa. So, each respondent viewed two ads, but only ever in one order and only for one code-switched/control condition.
After viewing the slogan and advertisement, people were asked to give their evaluation of the slogan just viewed. They were then asked to write down any associations the German/English language in the slogan may have evoked for them. After this they were asked to indicate their overall feelings toward the use of English/German in the advertisement on a semantic differential scale.
The final set of questions were the manipulation check questions, first the product and purchase CIP questions, and then the question asking specifically about their level of involvement with the advertisement. Participants were then thanked for their participation and told that the advertisements shown were not endorsed by the depicted brands and associated companies, and were used only for the purposes of the study.
4. Data Analysis Plan
In the following section, the previously discussed high and low involvement models will be analysed. Analyses of the full models rather than individual analyses of the component parts will be carried out. However, these contain sub-analyses of all the sections that make up the model, which will be looked at in detail. Therefore, the analysis plan will progress as follows:
1. Order effects check: this will look into any effects the order of viewing of the ads
may have had on subsequent evaluations.
2. Manipulation check: this will look into whether the manipulation of the two
involvement conditions, high and low, was effective.
3. Regression assumptions check: to test whether assumptions have been met to allow
for meaningful analysis of the results of the subsequent hypothesis tests.
4. Hypothesis testing:
a) Low involvement model – Mediated Mediation
i) Brand evaluations as second mediator ii) Product evaluations as second mediator
b) High involvement model – Moderated Mediation
i) Brand evaluations as mediator ii) Product evaluations as mediator
Relevant descriptive statistics will be included in each of the above sections. Summary tables of the figures mentioned in text can be found after each section or between the text as
appropriate, unless otherwise specified.
5. Results
of interest. After this, a manipulation check will also be done to test whether there were differences in involvement between the two conditions, high and low. In addition, a brief summary of the outcomes of the assumptions tests for multiple regression analysis will be given.
5.1 Order effects check
Since the involvement part of the experimental design was a within-subjects design, the order in which people saw the high and low involvement product ads may have affected their evaluations of the language used, their brand and product evaluations and their purchase intention, i.e. the effects of seeing and evaluating one ad may have affected perception and thus evaluations of the second ad.
In order to account for any potential effects the order in which the advertisements (either high involvement/car or low involvement/ice cream) may have had on subsequent evaluations, an independent samples t-test was conducted, with a grouping variable (High_first) as the independent variable, indicating whether the participant first saw the car advertisement (high = 1), or saw this second, i.e. they saw the ice cream ad first (low = 0). The dependent variables that constitute the two proposed models were included in the analysis in order to see if there were differences on any of these that may affect analysis of the models.
The null hypothesis for Levene’s test of equality of variances is that the variances are equal between the groups. The null hypothesis of equal variances could not be rejected for all but two of the dependent variables, language favourability for the ice cream ad (F = 1.83, p = .01) and brand evaluations for the car ad (F = 3.69, p = .06), although the latter tended toward marginal significance. Nonetheless, for these variables equality of variances was not assumed. For all other dependent variables tested, equality of variances was assumed.
The t-test for equality of means was only significant for brand evaluations for the car ad (t(116.2) = 1.98, p = .05). Therefore, the mean brand evaluation score between the group that saw the car ad first (N = 63, M = 5.17, SD = 1.30) and the group who saw it second (N = 79, M = 5.57, SD = 1.03) is significantly different, suggesting that there were some carryover effects on brand evaluations.
splitting the file in two for high first/low first analyses, since this would result in a sample that would be too small for analysis.
5.2 Manipulation Check
A manipulation check was performed to test if the manipulation of product involvement was effective. Table 1 displays the sample sizes, means, and standard deviations for the variables that were used in the manipulation check. The sample sizes varied slightly, given that not all people completed all parts of the survey. However, these participants were deliberately kept in so as not to lose valuable information provided by these respondents for other questions that they did complete. It is not expected that this will materially affect the results. In the hypothesis tests, the programme excludes cases where there is missing data, as such there may be some discrepancies between the sample used in descriptive statistics and the sample actually analysed for the hypothesis tests. However, again this is not expected to be material.
Table 1
Sample sizes, means, standard deviations and Pearson correlations for the order check variables, in terms of the variables in the same involvement condition only
N M SD Correlations
CIPtot_car 142 5.12 1.28 with Adinv_car = .28*
CIPtot_ice 144 4.30 1.34 with Adinv_ice = .16
Adinv_car 140 4.17 1.28
Adinv_ice 144 4.55 1.34
* p<.01
The null hypothesis for this test is that the means for both groups, for both high (car) and low (ice cream) involvement product are the same. Since this was a within-subjects design, hence the need for repeated measures ANOVA, the assumption of independence between high and low involvement groups is not met. Thus, the test of sphericity was analysed. This was not met (p < .001), thus degrees of freedom were adjusted and Greenhouse-Geisser and Huynh-Feldt results analysed in the ‘Within-Subjects Effects’ table. This showed that the manipulation of involvement conditions was not statistically significant and thus did not work as desired, since there was no difference in ad involvement scores, i.e. people’s involvement with the ad, between the conditions, F(1, 135) = 1.60, p = .21.
This may be due to many factors, but it could also be that people disliked the picture or slogan, or as indicated by some in the car/high involvement condition, there was a perceived lack of fit between the slogan and the picture. This will be elaborated on in the discussion. Given this, a manipulation check using ad involvement as the criterion may not give a good enough indication on how people would respond to these two different products. And since a car and an ice cream can be objectively deemed to be quite different when it comes to not only their physical attributes but also the decisions involved in purchasing them, one might reasonably expect there to be differences in the involvement people have with these two products, which could also be reflected in the way they view the associated advertisements (provided nothing else might interfere with their evaluation or appreciation of the ad, something that may have happened here).
Therefore, while the above analysis using ad involvement cannot be disregarded, the same analysis was run with ‘Consumer Involvement Profile’ added instead of ‘Ad involvement’ in order to see if there was an interaction with general involvement with the product class and purchase.
in this study, their involvement with the advertisement may mean that the manipulation and thus the effect of products depicted in the advertisements did not come through as hoped. This needs to be kept in mind when analysing the hypothesis tests and making conclusions on these, since we cannot be sure that any differences observed between the two conditions are a result of the two conditions (i.e. a result of the manipulation of involvement) and not despite them.
5.3 Regression Assumptions Check
A few tests were run using a random variable procedure to test for outliers as well as the assumptions of regression in order to be able to carry out and meaningfully interpret the subsequent analyses. Less than five respondents were just over the chi-squared threshold values (df = 4) at p = .001, across the conditions, to be deemed outliers. However, since they were only just over the threshold levels and when checked their answers seemed legitimate, it was decided not to remove them from the sample since their effects were considered not to be material.
Low involvement (N = 144)
Linearity of the standardized residuals was considered fine. The residuals were also relatively normally distributed, but with a slight right skew. Even if normality is not entirely met, bootstrapping confidence intervals at 95% with a sample of 5000, address this issue (Hayes, 2017, p. 172). There does not seem to be an overly concerning amount of heteroscedasticity in the plot of the standardized residuals, nonetheless, taking a conservative approach heteroscedasticity-consistent standard errors were run for all analyses, by selecting HC3 in the PROCESS options which calculated these according to Davidson-MacKinnon.
High involvement (N= 142)
5.4 Hypothesis Testing
The following analyses were all run with heteroscedasticity-consistent standard errors (HC3), mean-centering for products (which only came into effect for interaction effects in the moderated mediation), and 95% confidence interval for bootstrapping with a sample of 5000.
5.4.1 Mediated Mediation Analysis - Low Involvement Model (ice cream)
In order to test if code-switched advertising was processed under a peripheral route for the low involvement product, as hypothesised in Model 1, a mediated mediation was run. This analysis was conducted with code-switching condition, an indicator-coded multi-categorical variable as the independent variable. Language favourability, that is, the evaluation of the use of English in the slogan, was used as the first mediator. This variable is a proxy for the valence of associations derived from code-switching use in the slogan. Brand or product evaluations were the second mediator, leading to purchase intention as the dependent variable. Table 2 contains the relevant descriptive statistics for the variables of the low involvement model.
Since there were no statistically significant carry-over effects, the whole sample was used rather than splitting analysis for those who saw the high (low) involvement ad second and those who saw the high (low) involvement ad first. However, even if there were carryover effects, the sample size did not permit splitting, since this would yield sample sizes per condition that would be too small.
The analysis was run twice to test the model once with brand evaluations and once with product evaluations as the second mediator. The analysis with brand evaluations as the second mediator will be presented first, followed by that with product evaluations as second mediator.
Table 2
Sample size, means, standard deviations for the variables of the low involvement model
5.4.1.1 Brand Evaluations (mediator 2)
The following analysis concerns the results produced when testing the model (see
Figure 3) with brand evaluations as the second mediator. All numbers discussed in the
following analysis can be found summarised at the end of this analysis in Table 3.
Figure 3
Note. No code-switching group N = 50. 1-word English code-switching group N = 47.
3-words English code-switching group N = 47. *p</=.05 **p<.01 ***p<.001
Predicting language favourability evaluations (M1) (model 1 in Table 3)
Evaluating path a1 of the model, i.e. the effects of code-switching condition on the first
mediator, shows that the one-word code-switching condition on language favourability was significant, B = -.58, t(141) = -1.98, p = .05, as was the effect of the three-word code-switching condition on language favourability, B = -.99, t(141) = -3.07, p < .01. This means both the slogan with a low amount of English (B = -.58) and that with a higher amount of English (B = -.99) lead to significant negative evaluations of the English language used in them, more so for the slogan that used more English, relative to the no code-switching condition.
Predicting brand evaluations (M2) (model 2 in Table 3)
Looking at path a2, the effects of code-switching condition on the second mediator,
shows that the one-word English code-switching condition did not significantly predict brand evaluations, B = .20, t(140) = 0.80, p = .43, and nor did the three word English code-switching condition, B = .33, t(140) = 1.37, p = .17. However, for path d21 of the model, language
4.30, p < .001, meaning that as language favourability increases, or decreases, brand evaluations move in the same direction, providing support for the first part of Hypothesis 1.1 which posited that the valence of the associations would determine the valence or direction of brand evaluations and purchase intention. This also suggests that there could be an indirect effect at work.
Predicting purchase intention (Y) (model 3 in Table 3)
In terms of the direct effects on purchase intention (path c’), the one-word code-switching condition did not significantly predict purchase intention, B = -.05, t(139) = -0.24, p = .81, and nor did the three-word code-switching condition, B = -.14, t(139) = -0.63, p = .53. However, when controlling for code-switching condition, both mediators were significant predictors of purchase intention: M1/language favourability, B = .15, t(139) = 2.47, p < .05, and M2/brand evaluation, B = 1.01, t(139) = 18.02, p < .001. Thus, for two people in the same code-switching condition, the person whose language favourability was one unit higher, showed a .15 increase in purchase intention. In terms of M2/brand evaluation, for two people in the same code-switching condition, the person whose brand evaluation was one unit higher, showed a 1.01 unit increase in purchase intention for the ice cream. This provides support for the second part of Hypothesis 1.1. Therefore, the mediators’ separate effects on purchase intention both show a positive relationship, meaning they move in the same direction.
Given that code-switching condition is a multi-categorical variable, and that this model predicts two mediators operating in serial, one also has to look at the relative direct and indirect effects to get a clearer picture.
Total effects of code-switching condition (X) on purchase intention (Y) (model 4 in Table 3)
“…investigators routinely begin the causal steps procedure by first testing whether X affects Y by conducting a hypothesis test for "!, the total effect of X. A failure to reject the null hypothesis that "! = 0 means that the remaining criteria to establish M as a mediator are irrelevant, so the causal steps procedure stops in its tracks. This logic is predicated on the belief that an effect that doesn’t exist can’t be mediated, so there is no point in trying to explain the mechanism generating such a noneffect. But this logic is flawed. It is possible for X to exert an effect on Y indirectly through M even if one cannot establish through a hypothesis test that the total effect is different from zero. Although this seems counterintuitive, that doesn’t make it not true. The size of the total effect does not constrain or determine the size of the indirect effect. An indirect effect can be different from zero even when the total effect is not”.
Thus, given the above, indirect effects will be analysed here and assertions about mediation will be based on an assessment of the 95% bootstrapped confidence intervals of these effects. According to Hayes’ (2017) recommendations, an “inference about indirect effect(s) serves as a statistical test of mediation” (p. 187) and that if “at least one of the g-1 [reference groups] relative indirect effects is different from zero” (p. 192) the effect of X on Y, here the effect of code-switching condition on purchase intention, can be said to be mediated.
Relative total, direct and indirect effects of code-switching condition (X) on purchase intention (Table 3)
Looking at the relative total and direct effects of code-switching condition on purchase intention and their associated Omnibus test, one can see again that both the total (c) and direct (c’) effects for both code-switching conditions, one word (X1) and three word (X2), are not significant or substantially different from zero, since their confidence intervals include zero
significant, there does seem to be mediated mediation effect in operation here given that at least one of the indirect effects is different from zero.
Since the confidence intervals are both negative, this means that the indirect effect of code-switching (one or three words) on purchase intention through both mediators, relative to not code-switching, is negative, and more so for the three-word (a2d21b2 = -.27) than the
one-word (a1d21b2 = -.16) switching condition. Both the one word and three-word
code-switching conditions, relative to the no code-code-switching condition, lead to negative evaluations of the English language use (negative a1). And this negative evaluation then negatively affected
brand evaluations (because of the positive d21 i.e. it moves in the same direction). Given the
positive relationship of brand evaluations with purchase intention (positive b2), the decreasing
brand evaluations then also lowered purchase intention for both code-switching conditions. For the other partial relative indirect effects, only the three-word code-switching condition showed a significant effect with a negative confidence interval of between -.3136 and -.0199, for its effect on purchase intention through language favourability only. Thus, increasing the extent of English in the slogan by one code-switching unit (in this case from 1 to 3 words), starts to show a significant negative effect on people’s evaluation of the language use, and given language favourability’s positive relationship with purchase intention (since b1
is positive) this then also causes purchase intention to move in the same direction and decrease. Since both these significant indirect effects showed a higher negative effect for the three-word code-switching condition than the one-word code-switching condition, relative to the no code-switching condition, Hypothesis 1.2 is not supported, given that it was hypothesised that there would be no differences based on the different amounts of English in the slogans, due to a peripheral route being taken. There could be several reasons for this, which will be looked into in the discussion.
involvement and thus it is not certain that purely peripheral processing was engaged. The unstandardized coefficients of the effects of the model paths can be seen in Figure 3.
Table 3
Low involvement model results – brand evaluations (mediator 2)
B t p Model fit Model 1 à predicting M1 R2 = .07, F(HC3) = 4.97, df1 = 2, df2 = 141, p = .01 X1 (a1) -.58 -1.98 .05 X2 (a1) -.99 -3.07 .00 Model 2 à predicting M2 R2 = .11, F(HC3) = 6.36, df1 = 3, df2 = 140, p = .00 X1 (a2) .20 .78 .43 X2 (a2) .33 1.37 .17 M1 (d21) .27 4.30 .00 Model 3 à all predicting Y (direct effects c’) R2 = .65, F(HC3) = 107.37, df1 = 4, df2 = 139, p = .00 X1 (c’) -.05 -.24 .81 X2 (c’) -.14 -.63 .53 M1 (b1) .15 2.47 .01 M2 (b2) 1.01 18.02 .00 Model 4 à X predicting Y (total effect c) R2 = .00, F(HC3) = .20, df1 = 2, df2 = 141, p = .82 X1 (c) -.10 -.28 .78 X2 (c) -.22 -.62 .53
