Influencer marketing:

(1)

Influencer marketing:

Investigating its effect on consumer’s online purchase

preference

Xuyang Zhang (S2975009)

Supervisors: dr. Hans. Risselada & dr. Abhi Bhattacharya

Faculty of Economics and Business MSc. Marketing Intelligence

(2)

In the field of marketing, advertisement has always been playing an important role in influencing customers’ purchase decisions. With the fast development of internet, online shopping is not a new phenomenon anymore. Along with that, online advertising has experienced explosive growth. We are exposed to all kinds of online advertisements everyday (banners, pop-ups, ads in online videos, etc.). However, it is to be noticed that traditional online advertisements are not always effective these days. “About a third of online advertising campaigns don’t work—they don’t generate awareness or drive any lift in purchase intent,” said by Randall Beard in 2015, president of Nielsen Expanded Verticals. According to in Nielsen’s latest ‘Global Trust in Advertising Survey 2015’, trust on different sources of advertising are reported. Online video ads (48%, no change from 2013), search engine ads (47%, one percentage down), social networks ads (46%, two percentage down) online banner ads (42%, no change). (Nielsen, 2015). Given the low effectiveness of traditional online advertising, the key question for both academics and practitioners is how to engage in consumers by other effective approaches.

(4)

In the domain of eWOM, influencer marketing has recently becoming one of the popular marketing solutions in practice. However, influencer marketing is still a relatively new term in the field of academic marketing research by 2017. In fact, influencer marketing should be seen as one evolution approach of eWOM marketing. In order to generate product-related WOM, prior research introduced seeded marketing campaigns (SMCs), which are also referred to as viral marketing campaigns. An SMC is about selecting certain customers to spread WOW about a focal product. Influencer marketing is concerned as one of the popular approaches of seeding strategy (Chae, Stephen, Bart, & Yao, 2016). A recent paper defined ‘influencers’ as people who built a large network of followers and they are considered as tastemakers in niches to promote their products (De Veirman, Cauberghe, & Hudders, 2017). Therefore, influencer marketing can be seen as a new way for companies to reach and cooperate with these online influencers who own large online networks to spread eWOM and promote their products or services.

(5)

The first research objective of this paper is to find out how different types of influencer marketing (paying vs. non- paying) make an impact on consumer’s online purchase preference. Specifically, this paper conceptualizes influencer marketing in terms of influencers’ recommendations of a product or service, which could directly affect consumer’s purchase preference. There are several types of influencers’ recommendations defined, including non-sponsored recommendation, indirect-monetary recommendation and direct-monetary recommendation. This paper is aimed to find out which way of recommendation is more preferred by online consumers.

The second research objective of this paper is to find out whether the patterns of preference towards different types of recommendation have been influenced by other factors, and how these factors influence consumers preference patterns. Firstly, this paper will look at the moderating effect from national brands vs. private labels. Because influencer marketing can be used for both two brand types, but consumers’ choices also differ due to the nature of different preference of brand types. Secondly, the intrinsic gender difference could also be a moderator that men and women may perceive differently about influencer marketing. Moreover, given the fact that each individual has different level of education, level of income, the difference in individuals’ background could moderate the preference pattern. Besides, since influencers’ networks are built on the Internet, the personal daily use of social networks can also be an important factor to consider. Last but not least, people from different cultural backgrounds may also have different preferences. The moderating effect of culture using Hofstede’s cultural dimension on individualism and collectivism is examined. Overall, by looking at the above-mentioned factors, it is expected to for this paper to get the image of the preference of influencer marketing on an aggregated level.

Beyond the aggregated-level preference, the third research objective of this paper is to explore the heterogeneity of preference on individual level. Based on all the factors influencing the effect of influencer marketing, this paper will further investigate the patterns of individual preference of influencers’ recommendations. As a result, different consumer segments will be classified.

(6)

In addition, Latent Class Analysis is also applied to explore the heterogeneity of preference on individual-level and classify segments accordingly. With respect to the example product category, current study uses fashion (clothes) industry as the focused industry, since brands and prices can be clearly divided into different tiers for manipulation.

In terms of implication, from the academic point of view, this paper sheds light on adding more knowledge to the topic of influencer marketing. This paper described how influencer marketing is related to eWOM marketing. It also distinguishes itself from the traditional eWOM marketing. Since traditional researches are mainly about organic recommendations, while this paper also further discussed sponsored recommendations. And also in terms of the methodology, this paper tests the effect of influencer marketing in a choice based conjoint analysis, which is more accurate than the measurement of rating scores in other literatures.

For practitioners, this paper further illustrates which types of influencer marketing work effectively, and how consumers preference varies in terms of different brand types, demographic groups and cultures. It answers the question of why and how marketers should initiate influencer marketing campaigns.

(7)

2. Literature review

2.1 Conceptual framework

It can be seen in Figure 1 above, the first explanatory variable is price, since price is a major concern when consumers make purchase. The second explanatory variable is influencer recommendation types. In order to see how different types of recommendation are preferred by consumers, four levels are specified including no recommendation, non-sponsored recommendation, indirect-monetary recommendation and direct-monetary recommendation. In addition, the third explanatory variable is brand types, because brand is also an important factor to consider when consumers make purchase, including national brands and private labels. The dependent variable in the conceptual framework is online purchase preference, which represents the actual purchase choice by an online consumer. Apart from the main effect, brand types are also conceptualized to have an interaction effect together with influencer recommendation types. Next, there are also several demographic moderators and one culture

Price (-)

Influencer recommendation types

- No recommendation - Non-sponsored - Indirect-monetary-sponsored - Direct-monetary-sponsored Online purchase preference Brand type: - National brands - Private labels Demographics - Gender - Age - Educational level - Income level Culture:

- Country individualism score -

Social Media time

- Daily time spent on social media

(8)

moderator on the preference of influencer recommendation types added, in order to capture the varying preference in consumers background. Lastly, since a lot of influencer marketing campaigns are carried out on social medias, one’s daily amount of time spend on social medias could also moderate the preference of influencers recommendations.

2.2 Influencer marketing

As mentioned in the introduction section, there is currently very limited research about influencer marketing. First of all, it is necessary to define what is an ‘influencer’. This definition actually should not be completely new. Traditionally, opinion leaders are referred to the influential people who are able to shape ideas of other people. They are frequently asked for opinions and advices by other people (Aleahmad, Karisani, Rahgozar, & Oroumchian, 2015). Market maven has similar definition, which was previously defined as “individuals who have information about many kinds of products, places to shop, and other facets of markets, and initiate discussions with consumers and respond to request from consumers for market information” (Price, Feick, & Higie, 1987). De Veirman et al., (2017) conducted a very recent research about influencer marketing, of which this paper has focused on the impact of number of followers of Instagram influencers on brand attitude. This paper defined ‘influencers’ as people who built a large network of followers and they are considered as tastemakers in niches to promote their products (De Veirman et al., 2017). Therefore, in terms of influencer marketing, the key identifier of an influencer is the ownership of large online networks. Given all the definitions, influencers can be seen as the product-specific or service-specific market mavens who are active on social medias or blogs with many followers, and they are willing to share their opinions, give advices and promote products for consumers. Regarding influencer marketing, it can be defined as the practice of firms to reach and collaborate with those online influencers who own large online networks to help companies to spread eWOM and engage in the promotion of products or services.

(9)

consumers have more preference towards recommendations in general, no matter which types of influencer recommendation. The following hypothesis is formulated as following:

H1A: When consumers make online purchase, any type of influencers recommendation is more preferred than no recommendation.

Regarding the types of influencer marketing, there are at least two ways for marketers to collaborate with influencers — sponsoring them to generate eWOM or gaining their support organically (Pohal, 2016). Latest research by Lu et al. (2014) studied the impact of sponsored bloggers’ posts on consumers attitudes of sponsored posts and purchase intention. This paper talked about the paid influencer marketing (sponsorship), and they also further distinguished between indirect-monetary vs. direct-monetary sponsorship. Indirect-monetary sponsorship includes many format, such as coupons, discounts, free samples, and exclusive activity attendance. Direct-monetary sponsorship implies the cash compensation to influencers of any amount (Lu et al., 2014). The major topic of their research is about the effect of sponsorship on consumers’ attitudes, and the impact on consumers’ purchase intention by 5-Likert scales.

(10)

The key point that current influencer marketing research will further investigate is what is the pattern of the preference towards different types of recommendations. In other words, this study is interested to know which type of influencer recommendation is more preferred than other types. Lu et al., (2014) showed in their research that no matter bloggers receive direct-monetary or indirect-direct-monetary benefits to generate recommendation posts, consumers attitudes toward the post remain unaffected. However, current paper expects different preference pattern. Firstly, non-sponsored recommendation represents organic support from influencers, their credibility and trustworthiness is the highest compared to the two sponsored types of recommendations, since there is no direct connection between the influencer and marketers. Secondly, inmonetary recommendation is expected to be more acceptable than direct-monetary recommendation. Because indirect-direct-monetary recommendations are mainly in the forms of discount or free samples etc., which are common marketing techniques. While direct-monetary recommendation implies the strong interests binding to firms, consumers will think the information lacks credibility, they will have the least preference towards the recommendation. Therefore, expecting the differences in the preference of influencers’ recommendations. The following hypotheses are formulated:

H1B: When consumers receive influencers recommendation, non-sponsored recommendation is more preferred than indirect-monetary-sponsored recommendation.

H1C: When consumers receive influencers recommendation, indirect-monetary-sponsored

recommendation is more preferred than direct-monetary-sponsored recommendation.

2.3 Brand types

(11)

On the other hand, both national brands and private lables could use influencer marketing to promote their products, the pattern of preferences over influencers recommendation might differ. Lu et. al (2014) indicates that if the consumer has high brand awareness of a particular brand in a blog article, consumers will hold more posisitve attitutes towards the related sponsored recommendation than the recommendations to other brands with relatively lower brand awareness. Given that national brands usually have higher brand awareness than private labels. Therefore, it can be inferred that if a focused brand belongs to national brands, consumers will care less about whether the recommendation from influencers are sponsored or not. Wheras, if the brand is a private label, consumers might become more suspicious and care more about whether the recommendations are non-sponsored or sponsored, especially they care more about wthether the recommendations are directly sponsored with money. Given the arguments in the previours section, non-sponsored recommendation should be the most preferred option. Furthermore, if the sponsorship is direct-monetary, consumers might considers it suspicious compared to indirect-monetary sponsorship. All in all, the following hypotheses are formulated to test the interaction effect of brand types and influencers recommendation on online consumers’ product pruchase preferences.

H2A: If a brand belongs to private labels, the preference of non-sponsored recommendation over indirect-monetary recommendation will be strengthened.

H2B: If a brand belongs to private labels, the preference of indirect-monetary-sponsored recommendation over direct-monetary recommendation will be strengthened.

2.4 Demographic variables

Besides, it is also interesting to investigate how does influencer marketing work in different demographic groups. Therefore, this study also examines the moderating effects of gender, age, educational levels and income levels.

2.4.1 Gender

(12)

consumer reviews on purchase intention is stronger for females than males (Bae & Lee, 2011). Thus, given that women are more susceptible to the opinions of others than men. In terms of the sponsorship to the influencers, firstly, if male’s preference is regarded as reference level, it can be inferred that female gender will weaken the pattern that non-sponsored recommendation is more preferred than monetary-sponsored recommendation. The pattern that indirect-monetary recommendation is more preferred than direct-indirect-monetary recommendation by male will also be weakened. Because females may focus more on the content of recommendation itself and therefore less consideration is paid to the issue of sponsorship. Therefore, current paper regards the preference from males as the benchmark, the following hypotheses describes the moderated preference towards recommendations when the gender of the consumer is female.

H3A: If a consumer is female, the preference of non-sponsored recommendation over indirect-monetary recommendation will be weakened.

H3B: If a consumer is female, the preference of indirect-monetary-sponsored recommendation over direct-monetary recommendation will be weakened.

2.4.2 Age

To be noticed, age is an important moderator regarding the preference of influencers recommendations to consider. Influencer marketing campaigns are gaining steam because they are solutions that align with the behavior and habits of young generations. Some of their defining attributes and behaviors can't be ignored by marketers, one of which is their potential receptiveness to influence marketing. A key feature is that they are relatively easy getting influenced by the opinions of others (Young, 2017). Therefore, as a result of being easily influenced by influencers, it is assumed that the younger the individual, the less consideration he or she will have towards the sponsorship of recommendations. On the contrary, the older the individual, the more consideration that individual will have towards the sponsorship of recommendations. The preference will be more towards non-sponsored recommendation than indirect-monetary recommendation, and they can accept indirect-monetary recommendation more than direct-monetary recommendation. The following hypotheses are formulated.

(13)

H4B: With higher age, the preference of indirect-monetary-sponsored recommendation over direct-monetary recommendation will be strengthened.

2.4.3 Educational level and income level

With respect to educational level, prior literature shows that people with higher educational level tend to engage in greater information gathering and usage before decision making. And they have greater awareness of alternatives. (Cooil, Keiningham, Aksoy, & Hsu, 2007). Influencers recommendations can be considered an important source of information for these consumers before purchase. In terms of sponsorship of influencers recommendations, highly-educated and high-income individual may prefer less about the sponsored recommendations. In a study of trust of sponsored content, Joe (2014) found that as education levels increase, the misgivings about the sponsored content become more pronounced. Also, higher educational level associates with higher income level. Thus, it can be inferred that as the increase of education and income level, the trust of sponsored influencer recommendation will decrease. And, since these highly-educated and high-income individuals engage in greater information search before they make their online purchase. It can be inferred that they will search for more non-sponsored influencers recommendations. As a result, the pattern of preference can be moderated by educational level and income level. The preference of non-sponsored recommendation over indirect-monetary recommendation will be strengthened. And also, higher education and income level could lead to the consideration that indirect-monetary recommendations are more acceptable than direct-monetary recommendations. The following are the hypotheses.

H5A: With higher educational level, the preference of non-sponsored recommendation over indirect-monetary recommendation will be strengthened.

H5B: With higher educational level, the preference of indirect-monetary-sponsored recommendation over direct-monetary recommendation will be strengthened.

H6A: With higher income level, the preference of non-sponsored recommendation over indirect-monetary recommendation will be strengthened.

(14)

2.4.4 Social Media Time

Another interesting factor to be looked at is about social media. Because the popularity of influencer marketing is closely related to the information spread on all kinds of social media. At individual level, the amount of time spent on social media per day could have an effect on the preference of different types of influencers’ recommendations. Since one will encounter influencer recommendations on different platforms, the more time spent online, he or she will be more exposed to all kinds of recommendations. When they realized that those recommendations are compensated, due to the high extent and frequency of exposure, he or she might consider the sponsorship as a normal situation in the online social media world. Given the preference of non-sponsored recommendation is the highest, indirect-monetary recommendation is on the second place, and direct-monetary recommendation is the lest preferred option. In can be inferred that the amount of time spent on social media could weaken the preference pattern regarding sponsored recommendations.

H7A: With higher amount of time spent on social media, the preference of non-sponsored recommendation over indirect-monetary recommendation will be weakened.

H7B: With higher amount of time spent on social media, the preference of indirect-monetary-sponsored recommendation over direct-monetary recommendation will be weakened.

2.4 Culture

(15)

Therefore, it is important to include culture as another moderator of the preference of influencers recommendations. Hofstede’s cultural dimension on individualism vs, collectivism is used to account for the impact of cultural difference. The key issue of this is the degree of interdependence a society maintains among its members. Individualism indicates that everyone takes care of himself or herself and his or her immediate family only; While collectivism indicates, people are born into extended families and cohesive groups for loyalty (Hofstede, 2011). By using Hofstede’s indexes of individualisms, it is assumed people from high individualism scores countries might consider the sponsored influencer recommendation less reliable or trustworthy than people from low individualism scores countries. It is expected to find out how do people behave differently towards the influencer recommendation. The following hypothesis is drawn:

H8A: With higher country individualism score, the preference of non-sponsored recommendation over indirect-monetary recommendation will be strengthened.

(16)

3. Methodology

A conjoint analysis will be used in this thesis to measure the preference of consumers. The first part is a description of purchase preference measurement procedure. After that, the second part is the study design of attributes and levels, followed by the experimental design and data collection. The third part is model specification and estimation procedure.

3.1 Purchase preference measurement

In marketing research, measuring personal preference and utility function evaluation has a long tradition. Conjoint analysis, especially choice based conjoint analysis (CBC), is often used for such measurements (Halme & Kallio, 2014). It is to be noticed that another traditional way to measure is rating-based conjoint (RBC) analysis. Both CBC and RBC are popular approaches in history. However, it was also found that CBC has higher individual-level validation and more accurate of individual-level prediction (Karniouchina, Moore, van der Rhee, & Verma, 2009). Moreover, CBC is also more effective than RB, since choices are integrated in people’s daily life. While RBC ratings cannot represent the actual behavior of a personal purchase decision, the meaning interpretation of the rating scale is questionable and open to discussion. Importantly, as introduced in the earlier part of this paper, cross-culture effect will be taken into account. RBC are problematic since different cultures have different reactions to rating-scales (Eggers & Sattler, 2011). Given all the facts, this current paper will use CBC to analyse consumer’s online purchase preference.

3.2 Study design

3.2.1 Influencer Marketing: choice of example category

(17)

3.2.2 Attributes and levels

In a conjoint analysis, participants jointly evaluate the attributes and levels of the product. The preference in product attributes is estimated to be decomposed in different levels. As mentioned in the conceptual part of this study, three main attributes are determined on different levels, including ‘Price level’, ‘Influencer recommendation type’ and ‘Brand type’. ‘Price level’ and ‘Influencer recommendation’ have four levels respectively, ‘Brand type’ has two levels’.

‘Price level’ is divided into ‘lower than €200’, ‘€200 to €500’, ‘€500 to €800’ and ‘Higher than €800’. These levels are determined by the actual representation of the marketplace price information of winter jacket in Europe. Taking the famous European fashion shopping website Zalando for example, majority of the prices locate in the range of ‘lower than €200’ (nearly 1400 items), followed by the second largest range ‘‘€200 to €500’ (nearly 720 items). The other two ranges have less numbers of items, of which ‘€500 to €800’ contains 60 items ‘Higher than €800’ contains 18 items (Retrived in November 2017).

In line with the conceptual part of current study, ‘Brand type’ is distinguished by ‘National Brand’ and ‘Private Labels’. ‘National Brand’ represents well-known brands in different tiers such as ZARA, ADDIDAS, G-STAR RAW, BURBERRY, Supreme, Canada Goose, etc. ‘Private Label’, on the other hand, stands for store brands in different tiers such as Zalando Essentials, Pier One, YOURTURN, Brooklyn’s Own, etc.

Lastly, ‘Influencer recommendation’ is consisted of four levels, according to the previous theory part in current paper. The four levels are ‘No recommendation’; ‘Non-sponsored recommendation’: organic (voluntary) support from online influencers; ‘Indirect-money recommendation’: coupons, gifts or discount rights to influencers; ‘Direct-money recommendation’: monetary compensation to influencers.

3.2.3 Moderators

(18)

individual’s daily amount of time spent on social media is also asked. The answers for the multiple choices will be used to test the moderation effects respectively. Lastly, the moderator of individualism score based on respondent’s nationality background uses the score provided by Hofstede, which is available online. After the completion of data collection, the multiple-choice data and conjoint multiple-choice data will be merged to analyze the potential moderation effects

3.2.4 Experimental design

In the experimental design of current study, a fractional design was applied instead of full factorial design. Taking all the variables and sub levels into account, there will be 32 (4*4*2) stimuli. Although full factorial design is the most effective approach, respondents cannot easily handle all the possible combinations in a limited survey time. According to Eggers & Sattler (2011), respondents tend to get bored when answering repeated and monotonous choice sets. In order to avoid such behavior, a survey design should include only about 12-15 choice sets or should motivate the respondents throughout the survey (Eggers & Sattler, 2011). Therefore, current study decided to use a fractional factorial design, which represents a subset of all the stimuli. In order to be as efficient as possible, the actual choice design is a random selection of all stimulus with 4 alternatives shown per choice set, which are maximally different from each other. And it is decided to use 12 choice sets in total. In addition, respondents are provided with short textual and graphic explanations to help complete the survey.

3.2.5 Data Collection

Because current paper is a master thesis conducted in University of Groningen, taking into account of the limitations of time and space, the questionnaire in this paper was mainly distributed in the Netherlands. While the background of the participants are not only Dutch citizens, but also other individuals from different countries and regions, in order to test influence of different culture background. The format of the questionnaire is digital, which helps the ease of spreading. Respondents can participate easily by clicking a link. To make sure the survey is designed in an understandable way, a pre-test of the survey was firstly performed. After the feedback from a small group of respondents, the set-up of textual explanation was improved, which make the survey clearer.

(19)

the minimum of 200 completed surveys (SurveyAnalytics, 2018). In order to get sufficient amount of response, besides the respondents from the personal network of the author, they are also encouraged to help sent out the questionnaires to their own network, which makes it a snowballing process.

3.3 Model specification

The statistical basis of current paper is random utility theory (RUT). The fundamental assumption of the model specification is that consumers choose the option with the highest utilities (U). It is also argued that the utility (U) is a potentially undetectable latent class for different attributes of a product or service. These constructs consist of systematic components (V) and error terms (ε), The error terms represent all the rest of the effect in the model, of which the model does not consider for (Eggers & Eggers, 2011).

U = V + ε

For all the attributes measured in this survey, there are three models established in current paper. The first model only captures the main effects from the three main attributes. The model specification is listed in Equation 1. The second model takes the moderating (interaction) effect of ‘Brand type’ on the ‘Influencer recommendation’ into account. The third model further extends and incorporates the moderating effects of other socio-demographic variables and cultural variable.

(i represents the indication of alternatives in one choice set; j represents the four levels of prices;

k represents the two brand types; l represents the four types of recommendations) Equation 1: Ui_{= β0 + β1 *price}j + β2 *brandk + β3*recommendation typel + ε

Equation 2: Ui= β0 + β1 *pricej + β2 *brandk + β3*recommendation typel + β4*brandj*recommendation typel + ε

Equation 3: Ui= β0 + β1 *pricej + β2 *brandk + β3*recommendation typel +

β4*brandj*recommendation typel + β5*gender*recommendation typel + β6*

gender*recommendation typel + β7* gender*recommendation typel + β8*

gender*recommendation typel + β9* gender*recommendation typel +β10*

(20)

It can be seen that the last moderation model adds is very large. It is decided to adopt a step-wise procedure, which test each moderator one by one in the model. After that, in the final moderation model, only the significant moderators will be used. On the one hand, this reduces the complexity of the model, on the other hand, it also helps to avoid multicollinearity.

According to Eggers & Eggers (2011), when dependent variable is an option that is any alternative from choice set J, the preference of alternative i can be translated in to probabilities (P) by Multinomial Nominal Logit model (MNL). The likelihood to choose choice i from choice set J can be presented as the model below:

(α represents the indication of attributes; k represents the indication of levels in each attributes)

𝑃(𝑖|𝐽) = exp⁡(𝛽𝑘 ∗ 𝑋𝛼) ∑𝑗_𝑗=1exp⁡(𝛽𝑘 ∗ 𝑋𝑗)

The above described CBC using MNL will aggregate the preference of a particular choice of all participants on an aggregated level, of which there is no preference heterogeneity across consumers. Therefore, a followed Latent Class Analysis will also be carried out in order to account for consumer’s heterogeneity and find out in what segments do different consumers belong to. Thus, to present individual preference, the likelihood that respondent a will choose alternative i out of a set of J alternatives is:

(21)

4. Results

4.1 Descriptive sample

The data collection lasted nearly one month. The total number of questionnaires completed is 182. Although slightly lower, the total number is close the minimum requirement of 200 respondents, which is sufficient for the conjoint analysis aggregated level and the latent class analysis on individual level.

The questionnaire of this paper covered six social population questions. First of all, there were slightly more male than female in the distribution of respondents. Men accounted for 54.40% of the total, while women accounted for 45.60%. Then, there was the distribution of nationality of the participants. The first question is about continents. In general, participants are mostly from Asia or Europe, and a total of single digit participants in other continents. Considering about the author's personal nationality, China has the largest number of participants in this survey (39.6%), followed by the Netherlands (23.1%) and Germany (15.9%).

(22)

aged 25 to 34. After that, it was the age group 35 to 44 years old, which accounted for 8.20% of the population. Regarding the distribution of educational level, the majority of respondents in the questionnaires indicated Master (56.6%), followed by Bachelor (30.8%), and a small number of participants indicated PhD (10.4%). Furthermore, in terms of income levels of respondents, majority of the respondents earn less than €1000, which is logical since a lot of participants are students. After that, the next two income levels (Between €1001 and €2000, Between €2001 and €3000) take large portion of the distribution, with 26.4% and 20.9% respectively. The last is about the social software daily use of time, the vast majority of participants spend 1 to 2 hours a day, secondly, 33.5% of participants spend 2.5 to 4 hours, 18.7% of participants spend less than an hour of time.

Figure 2 Descriptive Statistics

4.2 Conjoint Analysis

(23)

order to get reliable results, the moderators will be added to model one by one. A final model with all the significant moderators will be estimated as a final model.

Step-wise moderation effects

Table 1 and Table 2 below show the results of two insignificant moderators. Among the step-wise estimates of those moderators, for ‘Age’, it is not surprising to have all levels with insignificant moderation effects, since the majority of the participants are young people, leading to insufficient variation of age in the sample. Regarding ‘Brand types’, there was also no significant effects found. Therefore, the related Hypotheses 2A, 2B, 4A, 4B in current research are not supported. Since the significant moderation effects will be measured again in the final model, for the significant moderators in step-wise procedure, the results can be found in Appendix.

Table 1 Step-wise moderation model: Brand

Attributes Utility Std. Err Wald p-value

Price Levels Below €200 0.8285 0.0382 604.9782 8.4e-131**

€200 to €500 0.4627 0.0396 €500 to €800 -0.3671 0.0486 Higher than €800 -0.9241 0.0583 Influencer Recommendation Type No recommendation 0.1462 0.1523 43.1718 2.3e-9**

Non- sponsored recommendation 0.8826 0.1482

Indirect-monetary recommendation -0.2632 0.1535 Direct-monetary recommendation -0.7656 0.1753

Brand types National brands 0.3518 0.0265 176.1411 3.4e-40**

Private labels -0.3518 0.0265 Brand types*influencer recommendation types Brand*No recommendation -0.0390 0.1034 0.1421 0.71 Brand*Non-sponsored recommendation -0.0723 0.0991 0.5325 0.47 Brand*Indirect-monetary recommendation 0.0990 0.1059 0.8742 0.35 Brand*Direct-monetary recommendation 0.0123 0.0102 0.92

Table 2 Step-wise moderation model: Age

Price Levels Below €200 0.8285 0.0382 606.7640 3.4e-131**

€200 to €500 0.4601 0.0395

(24)

Higher than €800 -0.9218 0.0581 Influencer Recommendation

Type

No recommendation -0.0039 0.1266 48.1507 2.0e-10**

Non- sponsored recommendation 0.7510 0.1097

Indirect-monetary recommendation -0.2586 0.1306 Direct-monetary recommendation -0.4886 0.1571

Brand types National brands 0.3572 0.0250 203.9968 2.8e-46**

Private labels -0.35172 0.0250 Brand types*influencer recommendation types Age*No recommendation 0.0544 0.0665 0.6709 0.41 Age*Non-sponsored recommendation 0.0159 0.0576 0.0766 0.78 Age*Indirect-monetary recommendation 0.0748 0.0682 1.2006 0.27 Age*Direct-monetary recommendation -0.1452 0.0844 2.9600 0.085 4.2.1 Model fits Model fits

A comparison of model fits of main effects model and null model was performed. As indicated in the Table 3 below, Model 1 is the main effects model. It can be found that Model 1 is significantly better than the null model according to the Likelihood Ratio Test p(Chisq=1355.235, df=7)<0.05.

(25)

Table 3 Model fits comparison

Model 1: Main effects Model 2: Main effects + Significant moderators

LL(0) -3027.667 -3027.667

LL(beta) -2350.0493 -2224.3998

Npar 7 22

Likelihood Ratio Test compared to null model

p(Chisq=1355.235, df=7)<0.05 p(Chisq=1606.534, df=22)<0.05

Likelihood Ratio Test p(Chisq=251.299, df=15)<0.05

BIC 3789.3985 4563.2878 AIC 4350.0987 4492.7996 AIC3 4175.0987 4514.7996 CAIC 3614.3975 4585.2878 McFadden R-sq 0.2238 0.2653 McFadden R-sq adj. 0.2212 0.2590r Hit-rate 53.3% 55.9%

4.2.2 Main effects model

In Table 4 below, the results of main effects are provided. To be noticed, the results are based on the default effect-coding of the main attributes. With regard to the importance of the attributes, it is found that ‘Price level’ is the most important attribute (43.86%). The focused attribute ‘Influencer Recommendation Type’ in this research, also has high importance (38.25%). Lastly, ‘Brand Type’ has relatively low importance (17.89%). To double check and confirm the results of the focused variable ‘Influencer recommendation type’, a dummy coding method is also applied to transform all the conjoint data collected. The dummy coding procedure regarded no recommendation as the reference level. The results are shown in Table 5. It can be found that the estimates are similar to the effects coding.

Table 4 Main Effects Model (effect-coded)

Attributes Utility Std. Err Wald p-value Importance

Price Levels Below €200 0.8274 0.0381 606.4609 4.0e-131 43.86%

€200 to €500 0.4607 0.0395

€500 to €800 -0.3672 0.0486

Higher than €800 -0.9210 0.0581

Influencer Recommendation Type No recommendation 0.0929 0.0419 467.5874 5.0e-101 38.25%

Non- sponsored recommendation 0.7784 0.0378 Indirect-monetary recommendation -0.1247 0.0436 Direct-monetary recommendation -0.7466 0.0529

Brand types National brands 0.3567 0.0250 203.4785 3.6e-46 17.89%

(26)

Table 5 Main Effects Model (dummy-coded)

Influencer Recommendation Type No recommendation 0.1741 0.0805 3953.3385 3.9e-25

Looking at p-values in Table 4, it can be seen that all attributes are significant (p <0.05). The focused attribute in current study is ‘Influencer Recommendation Type’. In the earlier conceptual part of this study, Hypothesis 1A indicated that any format of recommendation is more preferred than no recommendation. The main results here verified whether Hypothesis 1A is supported. Firstly, though ‘Influencer Recommendation Type’ is a significant attribute, in order to find out whether the estimate of these levels of recommendation are significantly different from ‘No recommendation’, it is decided to get t-statistics. A t-statistic is calculated by first subtracting the utilities of two attribute levels. After that, these differences are divided by the pooled standard errors of these attribute levels. According to the t-statistics, it can be found that the difference between the utilities of ‘Non-sponsored recommendation’ and ‘No recommendation’ (t(175)= 7.720387, p < 0.05) is insignificant. Similar findings are also obtained for ‘Indirectmonetary recommendation’ and ‘No recommendation’ (t(175)= 2.450702, p <0.05), ‘Directmonetary recommendation’ and ‘No recommendation’ (t(175)= -9.004303, p, <0.05). To confirm the results, same calculation is conducted also for the dummy-coded version. The results show the same findings and confirmed the difference (‘Non-sponsored recommendation’ vs. ‘No recommendation’: t(175)= 2.309989, p<0.05; ‘Indirect-monetary recommendation’ vs. ‘No recommendation’: t(175)= -1.841691, p<0.05; ‘Direct-monetary recommendation’ and ‘No recommendation’: t(175) = -3.742624, p<0.05).

Furthermore, in both effect-coded and dummy-coded versions, with regard to the estimates of utilities, it can be seen that only ‘Non-sponsored recommendation’ has highest utility, followed by ‘No recommendation’. While the other two types of recommendation have negative utilities, which are even not as preferred than ‘No recommendation’. Therefore, Hypothesis 1A is only partly supported.

(27)

recommendation’ is more preferred than ‘Direcmonetary recommendation’. Again, t-statistics are calculated between each other. For ‘Non-sponsored recommendation’ and ‘Indirect-monetary recommendation’, t(175)=10.17109, p<0.05. For ‘Indirect-monetary recommendation’ and ‘Direct-monetary recommendation’, t(175)= 7.004098, p<0.05. Also, similar procedure is applied with the dummy-coded version. (‘Non-sponsored recommendation’ vs. ‘Indirect-monetary recommendation’: t(175) = 4.15168, p<0.05); ‘Indirect-monetary recommendation’ and ‘Direct-monetary recommendation’, t(175) = 1.900933, p<0.05). Thus, Hypothesis 1B and Hypothesis 1C are supported.

4.2.3 Moderating effects model

The moderating effects model is based on effect-coding, which includes the main attributes plus the significant moderators to as the final model, including ‘Income’, ‘Education’, ‘Social-Media-Time’ and ‘Culture’. The results are shown in Table 6 below.

Table 6 Moderation Model

Price Levels Below €200 0.8525 0.0395 606.4609 3.3e-131

€200 to €500 0.4697 0.0406

€500 to €800 -0.3741 0.0494

Higher than €800 -0.9481 0.0588

Influencer Recommendation type No recommendation 0.1439 0.0831 39.1036 1.7e-08

Brand types National brands 0.3780 0.0258 214.1355 3.6e-46

Private labels -0.3780 0.0258

Moderation Utility Str. Err Wald p-value

1. Gender*No recommendation -0.6335 0.0950 44.4665 2.6e-11**

Gender*Non-sponsored recommendation 0.1756 0.0782 5.0496 0.025** Gender*Indirect-monetary

recommendation

0.4128 0.0930 19.6978 9.1e-6**

Gender*Direct-monetary recommendation 0.0451 0.1133 0.1577 0.0450*

2. Education*No recommendation 0.2817 0.0718 15.3802 8.8e-5**

(28)

Income *Indirect-monetary recommendation 0.0224 0.0460 0.2368 0.63 Income *Direct-monetary recommendation -0.0061 0.0546 0.0134 0.91

4. Social Media Time*No recommendation -0.1449 0.0560 6.6979 0.0097**

Social Media Time*Non-sponsored recommendation

-0.0452 0.0454 0.9912 0.32

Social Media Time *Indirect-monetary recommendation

0.1954 0.0534 13.3948 0.00025**

Social Media Time *Direct-monetary recommendation 0.0944 0.0641 0.0068 0.93 5. Culture*No recommendation 0.0139 0.0017 65.0222 0.0095** Culture*Non-sponsored recommendation 0.0012 0.0014 0.6781 0.34 Culture *Indirect-monetary recommendation -0.0005 0.0017 0.0724 0.00028** Culture *Direct-monetary recommendation -0.0146 0.0022 44.7064 2.3e-11*

First of all, in the moderation model, the main effects are checked. It is found that all main attributes are significant. Regarding the focused influencer recommendation type variable, although the values of utilities changed. The findings of the main effects only model still hold. Non-sponsored recommendation is the most preferred option (β= 0.7926, p<0.05). No recommendation is more preferred than the two types of sponsored recommendation (β= 0.1439, p<0.05). Indirect-monetary recommendation is (β= -0.2862, p<0.05) less negatively preferred than direct-monetary recommendation (β= -0.6543, p<0.05).

Secondly, to interpret the moderation effect, it is achieved by adding the utilities of main effect plus the observed significant moderation effects on related sub levels of the influencer recommendation attribute. Regarding gender, male is coded as the 0, female is coded as 1. Firstly, results show that female are less likely to choose ‘No recommendation’ than male (β= -0.6335, p<0.05). They even will have negative preference of no recommendation. Secondly, it is found that gender has a significant moderating effect for all sub levels of influencer recommendation type. Hypothesis 3A assumed that the preference of ‘Non-sponsored recommendation’ over ‘Indirect-monetary recommendation’ will be weakened if the participant is female. To give clarity, the following calculation is shown:

Assume a participant is female, the overall utilities of influencer recommendation type become: - No recommendation: 0.1439 - 0.6335 = -0.4869

(29)

- Indirect-monetary recommendation: -0.2862+0.4182 = 0.0132 - Direct-monetary recommendation: -0.6543 + 0.0451= -0.6092

If calculate the difference of utilities between ‘Non-sponsored recommendation’ and ‘Indirect-monetary recommendation’, the range for male is (0.7926-(-0.2862)=1.0788). For female, the range become (0.9682-0.0132 = 0.955). Thus, the preference difference of ‘Non-sponsored recommendation’ over ‘Indirect-monetary recommendation’ become smaller, which represents a weakened preference pattern. Hypothesis 3A is supported. Next, Hypothesis 3B indicated that the preference of ‘Indirect-sponsored recommendation’ over ‘Direct-monetary recommendation’ will be weakened if the participant is female. The range of the two levels for male is (-0.2862-(-0.6543) = 0.3681), while the range for female is (0.0132-(-0.6092) = 0.6244) Therefore, since the utilities difference become larger, Hypothesis 3B is not supported.

Looking at educational level, the only significant moderating effect is found on ‘No recommendation’. Thus, it can be inferred that consumers with higher level of education are more likely to choose ‘No recommendation’ (β= 0.2817, p<0.05). For the three types of recommendation, the only significant moderation effect was observed on ‘Direct-monetary recommendation (β= -0.3454, p<0.05). Therefore, the assumption of Hypothesis 5A that higher level of education will strengthen the preference of ‘Non-sponsored recommendation’ over ‘Indirect-monetary recommendation’ is not supported. With the only significant negative moderation effect for ‘Direct-monetary recommendation’, the overall utilities with one level increase of education are as follows:

- No recommendation: 0.1439 + 0.2817 = 0.4256 - Non-sponsored recommendation: 0.7926 - Indirect-monetary recommendation: -0.2862

- Direct-monetary recommendation: -0.6543 -0.3454= -0.9997

(30)

Concerning income as a moderator, it is to be noticed that it the moderating effect on ‘No recommendation’ used to be the only significant in the step-wise procedure. However, in the final model here, none of the estimates is significant. The possible reason is that there is a significant correlation between educational level and income level. Because as also explained in the earlier conceptual part of this study, which mentioned that higher level of income associates with higher level of education. By conducting a correlation test between the two variables, a significant correlation was found (ρ = 0.315, p <0.05). Thus, this could be the reason that no significant moderating effect from income level was observed. Therefore, Hypothesis 6A and Hypothesis 6B are also not supported.

In terms of the amount of time spent on social media, it can be found that consumers with higher amount of time spent on social medial tend to have lower likelihood to choose no-recommendation (β= -0.1449, p<0.05). Secondly, Hypothesis 7A assumed that higher amount of time spent on social media could weaken the preference of ‘Non-sponsored recommendation’ over ‘Indirect-monetary recommendation. The moderating effect on ‘Non-sponsored recommendation is insignificant, while that for ‘Indirect-monetary recommendation’ is significant (β= 0.1954. p<0.05). Here are the overall utilities change with one level of time spent on social media increase:

- No recommendation: 0.1439 -0.1449 = -0.0001 - Non-sponsored recommendation: 0.7926

- Indirect-monetary recommendation: -0.2862 +0.1954 = -0.0908 - Direct-monetary recommendation: -0.6543

(31)

Looking at the last moderator culture, a significant moderating is also found for ‘No recommendation’ (β= 0.0139, p<0.05). It implies that citizens from more individualism-oriented country are more likely to choose ‘No recommendation’. Regarding Hypothesis 8A, it is assumed that a self-reported higher individualism score will strengthen the preference of ‘Non-sponsored recommendation’ over ‘Indirect-monetary recommendation’. The moderating effect on ‘Non-sponsored recommendation’ is not significant, while the moderating effect on ‘Indirect-monetary recommendation’ is negative and significant (β= -0.0005, p<0.05). The utilities change is as follows:

- No recommendation: 0.1439 +0.0139 = 0.1578 - Non-sponsored recommendation: 0.7926

- Indirect-monetary recommendation: -0.2862 -0.0005 = -0.2867 - Direct-monetary recommendation: -0.6543 -0.0146 = -0.6689

The results imply that the ‘Indirect-monetary recommendation’ will be further less preferred than ‘Non-sponsored recommendation’ as the country individualism score increases. Thus, Hypothesis 8A is supported. For Hypothesis 8B, it is assumed that higher individualism score will strengthen the preference of ‘Indirect-monetary recommendation’ over ‘Direct-monetary recommendation’. Firstly, the moderating effect on ‘Direct-monetary recommendation’ is also significant (β= -0.0146, p<0.05). Comparing the ranges change, it can be found that ‘Direct-monetary recommendation’ will be even further less preferred than ‘Indirect-‘Direct-monetary recommendation’ (-0.2862-(-0.6543) = 0.3681) vs. (-0.2687-(-0.6689)=0.4002). Thus, Hypothesis 8B is supported.

4.3 Preference-based segmentation: Latent Class Analysis

(32)

segments that differ in preferences can be found, and how many segments will be optimal accounting for the heterogeneity. The LCA applied in this study is a mixture model.

4.3.1 Optimal number of segments

Similar to the aggregated-level conjoint analysis, all the moderators are added in the model. The difference is that these moderators are added in the model as covariates. The estimation procedure was repeated multiple times with specification of 2 to 15 classes to see which solution is stable. The Table 4 below shows the scores of information criteria and classification errors. Firstly, several information criteria standards are applied to compare model fits. It is found that the lowest score on AIC (3123.919) is observed on 13-Class Choice model, and AIC3 (3325.236) are observed on the 8-Class Choice model. However, AIC and AIC3 are criticized for the small penalty of adding more parameters. On the other hand, BIC and CAIC are more preferred for large samples sizes. In terms of BIC, 7-Class Choice Model has the lowest score (3655.003). Looking at CAIC, 6-Class Choice model has the lowest score (3776.311). The difference between the 7-Class Choice and 6-Class Choice model is small. Secondly, all the classification errors are smaller than five percent, which indicates that all the participants in the survey are well allocated. The classification error of 0.0144 of the 6-Class Choice model is the lowest, which implies that with a probability of 1.44%, the participants in this survey are not assigned to the most probable segment. Therefore, it is decided to use the 6-Class Choice model as the focused model. In addition, to get a straightforward indication, a scree-plot is drawn. Table 7 shows that the minimum point is at 6 classes for CAIC. Therefore, 6 segments are considered as the most reasonable approach.

Table 7 Information Criteria & Classification Errors

AIC(LL) AIC3(LL) BIC(LL) CAIC(LL) Npar p-value Class.Err.

2-Class Choice 4188,2441 4217,2441 4281,1603 4310,1603 29 1,2e-757 0,0226 3-Class Choice 3877,3064 3928,3064 4040,7107 4091,7107 51 3,2e-699 0,0209 4-Class Choice 3646,9056 3719,9056 3880,7981 3953,7981 73 6,5e-658 0,0167 5-Class Choice 3432,4796 3527,4796 3736,8602 3831,8602 95 2,4e-620 0,0198 6-Class Choice 3284,4426 3401,4426 3659,3114 3776,3114 117 1,7e-597 0,0144

(33)

14-Class Choice 3132,9184 3425,9184 4071,6923 4364,6923 293 . 0,0222 15-Class Choice 3144,721 3459,721 4153,9831 4468,9831 315 . 0,023

Figure 3 Scree-plot: number of classes

4.3.2 Segments interpretation

The results of all the six segments from the LCA are shown in the Table 8 below. There are 6 segments classified. These segments differ in preferences.

Table 8 Segments interpretation

Importance Class 1 Class 2 Class 3 Class4 Class 5 Class 6

Price Levels 53.81% 41.93% 13.92% 85.08% 21.71% 57.21%

Influencer Recommendation Types 37.50% 20.65% 57.78% 8.20% 61.53% 26.50%

Brand Types 8.69% 37.41% 28.30% 6.73% 16.76% 16.30%

Segment size Class 1 Class 2 Class 3 Class4 Class 5 Class 6

44 35 32 31 25 15

Attributes Class 1 Class 2 Class 3 Class 4 Class 5 Class 6 p-value

Price Levels

Below €200 1.2904 0.8455 0.3246 4.3189 0.8198 -1.6478 4.4e-121

€200 to €500 1.8883 0.6846 0.5539 1.395 0.2188 0.3522

€500 to €800 -1.0211 -0.5366 -0.0776 -0.6992 -0.184 1.1076

Higher than €800 _-2.1576 _-0.9935 _-0.8009 _-5.0147 _-0.8545 _0.188

Influencer Recommendation Type

No

(34)

Non- sponsored recommendation 1.1783 0.2673 3.7827 0.1981 1.6887 0.5097 Indirect-monetary recommendation 0.2092 0.2912 -0.276 -0.2235 -1.7732 0.231 Direct-monetary recommendation -1.6413 0.056 -1.6674 -0.4369 -2.3303 -0.7666 Brand types

National brands 0.3268 0.8203 1.3768 -0.369 0.6464 0.3925 1.8e-51

Private labels -0.3268 -0.8203 -1.3768 0.369 0.6464 -0.3925 (Covariates) Gender Male 0.0822 -0.6978 -0.4084 -0.0117 1.4801 -0.4442 0.011 Female -0.0822 0.6978 0.4084 0.0117 -1.4801 0.4442 Educational Level High School 1.8636 -0.0266 0.8894 -2.6896 -5.8814 5.8446 0.72 Bachelor 2.3107 3.6087 -0.4233 3.26 -5.4144 -3.3418 Master 1.44 1.3351 -2.4913 1.9981 -1.1914 -1.0904 PhD -4.2288 -1.5711 1.8573 -1.9532 4.5269 1.3688 Others -1.3855 -3.3461 0.1678 -0.6154 7.9603 -2.7812 Income Level Below €1000 -0.0268 0.8607 0.2986 0.9374 3.8007 -5.8707 0.044 €1001 to €2000 0.4707 1.0436 1.4463 0.1347 -2.7894 -0.3059 €2001 to €3000 -0.3411 -1.1181 0.3789 -0.6437 1.5292 0.1948 €3001 to €4000 -4.9369 -1.9306 1.2924 -3.8726 2.1822 3,4042 Above €4000 -4.8341 -2.7169 -3.4163 -3.4442 -4.7226 2.5775

Social Media Time

Less than 1 hour -4.9369 1.9306 1.2924 -3.8726 2.1822 3.4042 0.44

1 to 2 hours 4.8341 -2.7169 -3.4163 3.4442 -4.7226 2.5775

2.5 to 4 hours -0.9722 -0.2314 1.0187 0.4983 0.55 -0.8635

4.5 to 6 hours -0.405 -0.7465 0.8754 -0.1657 0.2577 0.1841

Over 6 hours 0.6644 -0.0826 0.9293 0.695 -0.9335 -1.2727

Culture 0,0299 -0,0683 -0,0192 0,0097 0,0672 -0,0193 31,5652 7.2e-06

Segment 1: Male main-stream non-sponsored recommendation seekers

(35)

Segment 2: Female price-sensitive indirect-monetary recommendation seekers

The second segment consists of 35 participants. The attributes importance is relatively even. Similar to Segment 1, price is also the most important factor for them (41.93%). Brand types is the second important factor (37.41%), and influencer recommendation types takes up 20.65% of importance. This Segment are mainly female participants, also at the income level between €1001 to €2000. They are price-sensitive, with the most preferred price level at below €200. National brand is their primary choice. And they also accept indirect-monetary recommendation.

Segment 3: Female non-sponsored recommendation lovers

32 participants are included in the third segment. Regarding importance of attributes, influencer recommendation types is the most important factor (57.78%), and brand types is the second important one (28.30%). Price level for this group is the least important factor (13.92%). This group is also a female segment with income level between €1001 to €2000. Although price is not the most important factor, like Segment 1, they consider that €200 to €500 is most preferred. In terms of recommendation, it is important for them to have non-sponsored recommendation as their wanted type of recommendation. Also, national brand is the most preferred brand type.

Segment 4: Female extremely price-sensitive consumers

Segment 4 has 31 participants. This segment is the price-sensitive group. Looking at attributes importance, it can be found that price is extremely important for them (85.08%), the other two attributes only take up 8.20% and 7.73% respectively. This segment is the female group with below €1000 income. Since they are price sensitive, price level of below €200 is their first choice. The recommendation from influencers are not very important to them, no recommendation is considered as their common option. With regard to brand types, it is not surprising that private labels are preferred.

Segment 5: Male relatively price-sensitive independent consumers

(36)

Segment 6: Female high-income non-sponsored recommendation seekers

The last segment contains 15 participants. Participants in this segment consider price is the most important factor (57.21%), followed by influencer recommendation types (26.50%). Brand types takes 16.30% of the importance. This segment is also mainly consisted of female participants. Most of them have income between €3000 to €4000. It is noted that they value the non-sponsored recommendation most. And their most preferred price option is between €500 to €800. National brands are most preferred by them.

5. Conclusions and discussion

This study investigated the very recent phenomenon of influencer marketing in a choice based conjoint analysis approach and latent class analysist. The key objective of current study is to find out how does influencer marketing make an impact on consumers preferences. Besides, it is also the interest of this paper to look at what factors influence the pattern of the preference. In addition, the preference patterns on induvial levels are segmented. The following Table 9 is a summary of all the hypotheses in this paper.

Table 9 Hypotheses overview

Hypotheses Supported

1A When consumers make online purchase, any type of influencers recommendation is more preferred than no recommendation.

Partly-supported 1B When consumers receive influencers recommendation, non-sponsored recommendation is more

preferred than indirect-monetary-sponsored recommendation.

Yes 1C When consumers receive influencers recommendation, indirect-monetary-sponsored

recommendation is more preferred than direct-monetary-sponsored recommendation.

Yes 2A If a brand belongs to private labels, the preference of non-sponsored recommendation over

indirect-monetary recommendation will be strengthened.

No 2B If a brand belongs to private labels, the preference of indirect-monetary-sponsored

recommendation over direct-monetary recommendation will be strengthened.

No 3A If a consumer is female, the preference of non-sponsored recommendation over

indirect-monetary recommendation will be weakened.

Yes 3B If a consumer is female, the preference of indirect-monetary-sponsored recommendation over

direct-monetary recommendation will be weakened.

No 4A With higher age, the preference of non-sponsored recommendation over indirect-monetary

recommendation will be strengthened.

No 4B With higher age, the preference of inmonetary-sponsored recommendation over

direct-monetary recommendation will be strengthened.

No 5A With higher educational level, the preference of non-sponsored recommendation over

(37)

5B With higher educational level, the preference of indirect-monetary-sponsored recommendation over direct-monetary recommendation will be strengthened.

Yes 6A With higher income level, the preference of non-sponsored recommendation over

No 6B With higher income level, the preference of indirect-monetary-sponsored recommendation over

direct-monetary recommendation will be strengthened.

No 7A With higher amount of time spent on social media, the preference of non-sponsored

recommendation over indirect-monetary recommendation will be weakened.

Yes 7B With higher amount of time spent on social media, the preference of

indirect-monetary-sponsored recommendation over direct-monetary recommendation will be weakened.

No 8A With higher country individualism score, the preference of non-sponsored recommendation

over indirect-monetary recommendation will be strengthened.

Yes 8B With higher country individualism score, the preference of indirect-monetary-sponsored

recommendation over direct-monetary recommendation will be strengthened.

Yes

Main effects: Preference of influencer recommendation

To conclude the first research objective, non-paid influencer marketing is the most effective approach to generate positive impact. While for the paid influencer marketing, the effect in general is counterproductive, as sponsorship triggesr the negative preference by consumers. This finding is contradictory to the findings by Lu, Chang, & Chang (2014). They concluded in their research that whether bloggers receive direct-monetary or indirect-monetary, consumers attitudes remain unaffected. The reason behind is expected in the conceptual part of current study, when consumers realize the sponsorship to online influencers, they think the credibility and trustworthiness of sponsored recommendations decrease.

Moderating effects:

To conclude the second objective of current study, gender, educational level, daily social media time spending and country cultural background indeed have impact on the preference patterns of recommendations. While brand types, income level and age did not show significant impact in current research.

(38)

types of influencer recommendations. Secondly, age as a moderator, is also insignificant. The reason behind is because of the limitation of current study. The main participants are young people in this survey, the variation across age is not large.

Next, regarding gender, in an earlier research, it was found that women tend to use internet to receive social support and their e-commerce transactions are more emotional, while men’s style is more pragmatic (Teso, Olmedilla, Martínez-Torres, & Toral, 2018). Furthermore, as expected in the earlier part of this study, females indeed are observed with much smaller preference difference between non-sponsored recommendation and indirect-monetary-sponsored recommendation than males. Given that women tend to get more social support, indirect-monetary-sponsored recommendation by influencers can be seen as a way to search for more information and social support before purchase, apart from the most preferred non-sponsored recommendation.

In terms of education, looking at the findings of significant income level variable. The more that people are educated, the even less people will prefer direct-monetary recommendation. On the one hand, the negative preference is because of the low credibility and trust of the sponsored recommendation. On the other hand, highly-educated individuals have the ability to engage in more information search, as there are many alternatives to direct-monetary recommendations. Thus, educational level indeed makes an impact of a consumer’s preference pattern.

Regarding the moderator social media time, it is confirmed that people are less likely to follow no recommendation with higher amount of time spent on social media per day. They consider that indirect-monetary recommendation is more acceptable than direct-monetary recommendation. The possible reason is that indirect-monetary recommendation has become a common scenario for many influencers active on online social media. Those who are exposed to social media for longer time per day have the willingness to view certain influencers’ recommendations, their dislike of indirect-monetary recommendation may decrease when they are exposed to that frequently.

(39)

recommendation and direct-monetary recommendation than those from low individualism-oriented countries. Specifically, the high individualism-individualism-oriented countries in current study are European countries Netherlands and Germany, and the low individualism-oriented country is China. The possible explanation of the results can be inherent in the culture. According to the Nielsen Global Trust in Advertising (2015), there is a relevant dimension which provide useful insight. With regard to ‘the trust or somewhat trust towards consumer opinions online’, European participants has the lower percentage score (60%) than Asian participants (70%) (Nielsen, 2015). Therefore, it can be inferred that for high-individualism-oriented countries, the preference for non-sponsored recommendation over sponsored recommendation is much higher than low individualism-oriented countries.

Latent class individual-level heterogeneity

Finally, to conclude the last objective of this study, the individual-level heterogeneity is revealed by the results of Latent Class Analysis. Similar to the conjoint analysis, the major finding is that non-sponsored recommendation is the first choice for the majority of the segments. There are also two other interesting segments. Firstly, females from low-individualism-oriented countries tend to value indirect-monetary-sponsored recommendation as also their first choice. Secondly, for the price-sensitive segments, they do not want any type of recommendation, and they feature for the low-income level.

6. Limitations and implications Limitations

(40)

Implications

Firstly, for academic research, this paper intents to fill in the gap of influencer marketing topic. As mentioned in the earlier text, influencer marketing can be seen as a novel evolution approach of eWOM marketing. The difference between the traditional eWOW marketing and influencer marketing is that the former focuses on the spread of positive content of a brand, while the later also use extrinsic incentives (indirect-monetary vs. direct-monetary) to simulate the information flow. Current paper is the first research which considers different types of influencers’ recommendation as an attribute in Choice Based Conjoint method. The Choice Based Conjoint method for investigating consumers preference of influencer marketing has not been used before. By quantifying the effect of sponsoring vs. non-sponsoring of influencer marketing on consumer’s purchase preference, this paper contributes for researchers to advance the understanding of how does influencer marketing actually work and how consumers response to influencer marketing in their actual purchase choice.

Influencer marketing: