Customer engagement on Instagram for luxury fashion brands: An empirical comparative analysis

Abstract This study explores the engagement generated by the posts that luxury brands publish on Instagram, based on the different types of content, communication strategies and other aspects of those posts. Its main intention is to clarify which items lead to a greater engagement of luxury fashion brands with their Instagram audience. A content analysis method was applied to analyse a random sample of 598 Instagram posts published by two of the world’s most important luxury brands, Dior and Chanel. The analysed period was one year with the intention of having both a large sample and overriding seasonal effects. The results reveal which items studied generate more engagement, and which generate less. It is empirically confirmed that posts displaying products generate more engagement, especially if the products are fashion accessories. Moreover, content related to social responsibility also increases engagement. Finally, and as a highly relevant factor, we observe that posts in video format generate only limited engagement among followers. The results help companies gain a clearer idea of the potential applications of social media platforms, thus improving their understanding of the impact of social media. These findings are of interest to researchers and, especially, marketing and social media professionals, as they provide insights into what to publish in order to gain higher engagement rates.


Introduction
Today, the Internet is an essential part of organizations' communication strategies and brand management (Felix et al., 2017;Lorenzo-Romero et al., 2021;Vermesan et al., 2022). Brands need to expand their brand presence and online services to offer their consumers more experiences (Cho et al., 2018). Now traditional marketing tools coexist with digital platforms through the use of social networks (Guercini et al., 2018).
Consumers value information provided by other consumers, such as reviews and recommendations, more highly than corporate information (Hussain et al., 2018). Therefore, these interactions can influence consumption intention (García Medina et al., 2018) and contribute to building brand identity. As consumer interactions are less controlled and can spread rapidly, these interactions pose challenges that must be constantly monitored (Yadav & Pavlou, 2014).
This monitoring process provides information on the interaction with consumers (Beverungen et al., 2019), and brands tend to use posts that generate the most interaction (Wagner et al., 2017).
Numerous studies have addressed the analysis of various social networks (Hsiao et al., 2019;Kouokam & Dirik, 2019;Kursuncu et al., 2019;Na & Kim, 2019). In contrast, academic research examining the role of social media in the luxury sector remains largely unexplored (Arrigo, 2018).
Among the different social networks, Instagram is gaining more followers every day among consumers interested in the world of fashion and luxury. As of January 2022 it boasted more than 1,478 million active users per month (Statista, 2022).
In view of this reality, this study analyses the levels of engagement generated by the luxury brands Dior and Chanel on Instagram based on the presence of certain variables in their posts. These two brands were selected according to the data from the latest Interbrand ranking (2022), which ranks them as two benchmark brands in the luxury sector. The data was collected by monitoring the activity of these brands on Instagram, the characteristics of their posts and the interactions received.

Social networks in the fashion sector
Clothing and fashion accessories constitute a powerful social symbol that is used to communicate personal and group identities (Sörum & Gianneschi, 2022). Individuals express themselves through consumption in many ways, and in this context, the product and brands have the ability to convey messages to others (Son et al., 2022). Being able to decide what information to share and with whom conditions consumers' social identity (Wolny & Mueller, 2013).
Social networks allow fashion brands to facilitate their recognition and share trends with users dynamically and directly. Currently, most fashion brands are present on social networks and use them to communicate with their followers (Lim, 2020;Nash, 2019). They frequently use these channels to follow and share trends. Their impact on society is so great that it allows the individuals making up that society to be segmented based on their use. This symbiosis between consumers' engagement with fashion and the interaction with brands on social networks provides the ideal conditions for the analysis of these interactions, providing information that is relevant for the digital strategy of brands on the social networks on which they are present (Schultz, 2016).

The specific case of luxury brands
In recent years, global markets such as India, China, Brazil and South Africa have experienced rapid growth driven by accelerated industrialization through potent production and marketing strategies, an influx of direct foreign investment overseen by liberal government policies, technological innovations, the development of affordable and accessible telecommunications, infrastructure expansion and improved global connectivity that facilitates international trade (Cavusgil, 2021;Srivastava et al., 2020). This accelerated growth has led to an increase in living standards that is reflected in the consumption of luxury brands (Shahid & Paul, 2021;Singh et al., 2022).
The luxury market has witnessed a profound resurgence following the gloomy period of the pandemic, with the United States, China, Europe and South Korea showing strong momentum, with a greater focus on growing trends such as social media, the metaverse and sustainability initiatives .
The so-called relationship equity expresses the tendency of customers to maintain a relationship with a brand, going beyond its objective and subjective evaluations. Marketing activities of luxury fashion brands on social media have been shown to be positively correlated with purchase intent (Kim & Ko, 2012). This also affects the willingness to pay a premium price for brand loyalty.
Luxury is often centred on rarity, scarcity, and discretion. However, the easy availability of and accessibility to the contents of the digital environment seem to contrast with the exclusivity of luxury brands (Schmitt et al., 2022). However, with the convergence of digital apps and new media, such as mobile technology, luxury companies have realised the inevitability of developing an online presence (Arrigo, 2018). Platforms like Instagram have made it easy for luxury brand engagement due to its visual storytelling, which complements the aesthetic standards of many of the most sought-after brands (Bachmann et al., 2018;Dahlhoff, 2016) and have demonstrated that these marketing activities positively correlate with purchasing intent (Kim & Ko, 2012) and a willingness to pay a higher price for brand loyalty (Khan,).

User engagement on fashion social networks
User engagement is one of the benefits of the presence of brands on social networks, which various authors have made reference to. According to Helme-Guizon and Magnoni (2019), consumer engagement is the key element to generating brand loyalty. Although some authors have described this engagement as a multidimensional construct consisting of three dimensions (cognitive, affective, and behavioural), others have chosen to maintain two dimensions: onedimensional and two-dimensional. There are also those who have added the social dimension to the latter when talking about engagement in the social context (Michael, 2022).
Engagement encompasses various activities, including interacting with a brand and searching for and sharing brand information. In today's consumer-centric market, brand engagement helps consumers make informed purchasing decisions (Cheung et al., 2021). Therefore, marketers strive to engage consumers through advertising on social media, including on Instagram, in order to strengthen the relationship between the two and to ultimately increase commercial profitability (Cuevas-Molano et al., 2021).
Tools that monitor social networks measure engagement levels based on the various types of follower interaction (likes, comments and shares). These interactions make it possible to quantify user engagement and establish comparisons between different social networks.
Although multimedia content is relevant on all social networks, it is even more so on Instagram, since it is the main element of Instagram posts and, therefore, has a decisive impact on the engagement of brands with users. Relevant studies have addressed this issue, including those by Akram et al. (2022), Na and Kim (2019), and D. Lee et al. (2018) or Swinyard and Smith (2003), who observed that the amount of time dedicated to searching for information online positively influences the actual online purchasing behaviour. Others have shown that actions such as likes, positive comments or following the brand reflect consumer interest in the product, service or brand (J. Lee & Hong, 2016), while others have analysed the content published by fashion brands to study their marketing strategies (Chen & Luo, 2017), and others have referred to the attractiveness of a post generated through media richness (for example, the inclusion of images or videos) to obtain a positive effect on user behaviour (Schreiner et al., 2019). Ha et al. (2017) analyse the framing of images in posts (for example, selfies, body snaps, etc.) and their relationship with Instagram users' perceptions. Along this line, a study by Kusumasondjaja (2019) explores the role of visual aesthetics and the communication of luxury fashion brands on Instagram.
Regarding the presence of people in the posts, Yan et al. (2022) highlight that the face generates a greater degree of reaction; in contrast, Hu et al. (2014) point out that images that include only the product are less favourable for the Instagram audience. Another study by Valentini states that when the product appears in the foreground of the image, the audience is more engaged, while less interaction is generated when the product is in the background.
Based on these findings, the present study aims to answer the following question: what information can be obtained by studying the interactions between consumers and luxury fashion brands? This invites us to analyse not only the content of the posts but also to study how this content influences the engagement generated, being able to also compare it between the two brands analysed.

Objectives and hypotheses
The research framework aims to analyse interaction based on post contents, the communication strategy used and, finally, product elements and other formal aspects of the posts. Çukul (2015) defines eight categories to analyse the contents of posts: product, promotion, advertising, special days, social responsibility, workplace/workers, public relations and content provided by the consumer. Considering the Çukul system of categories, we can formulate the following hypothesis:

Variables referring to the contents of posts
H1. The presence of content variables in posts affects the engagement generated.

Communication strategy variables
Another issue that should be analysed is the ability of the different communication strategies established by Goor (2012) to generate engagement, who proposes six categories: persuasion, emotion, relationship, sales response, self-efficacy and symbolism ( Figure 1).
This system of categories has been used before to analyse fashion brands on Instagram. The analysis of these strategies and their ability to generate engagement on luxury fashion brand Instagram accounts allows us to formulate the following hypothesis: H2. Different communication strategies generate different levels of engagement.

Variables referring to the product and formal elements of the post
In addition to the category systems of Çukul (2015) and Goor (2012), the set of posts analysed were categorized based on the authors' own model, with 16 variables defining aspects of the product presented, as well as other formal elements of the post: people, no product/evocative, text in image, people's faces, studio, location, username mentioned, series of images, video, series of posts, Insta shopping, celebrities, clothes, make-up, accessories and perfumes. This system of categories was developed to be faithful to the required values of completeness, objectivity, mutual exclusion, fidelity and relevance (Krippendorff, 2004).
Considering the product elements and the formal elements of the post defined in this model, the following hypotheses are proposed: H3. The format of the posts affects their ability to generate engagement.

H4.
The presence of product variables in the posts affects the generation of engagement.
H5a. The engagement achieved by luxury fashion brands translates into a greater number of likes.
H5b. The engagement achieved by luxury fashion brands translates into a greater number of comments.

Method
The design of this study is exploratory, non-experimental and cross-cutting. Data from the official accounts of Dior and Chanel on Instagram were used. Saunders et al. (2007) propose this case study technique when carrying out exploratory work. Similarly, studies on the fashion sector prioritize the use of content analysis. Therefore, in order to achieve the proposed objectives, this study employed a quantitative methodology and, specifically, content analysis, which consisted of tracking and drawing conclusions from the data collected qualitatively (Easterby-Smith et al., 2015).
For the evaluation of the behavioural engagement (Bråten et al., 2022) of the followers of these brands, the reactions to each post published by the companies on Instagram were recorded. The field work was carried out during the period of one calendar year, but the data collection was carried out six months after this period in order to ensure that all posts achieved a maximum degree of reaction. We then analysed a daily post among all the posts published by each brand and carried out a random systematic sampling among the posts of the day. With this sampling, we aimed to analyse a complete year, making sure at the same time that the time or order the post was published during the day did not influence the observed values. In the end, and given that some days the brands did not publish any posts, a total of 598 posts were analysed. Each was categorized based on the 31 categories, and the number of reactions that the post generated was counted both in terms of likes and comments. Regarding the comments, these were counted numerically without carrying out a sentiment analysis of them.
In order to ensure rigorous and unbiased data coding, three external coders were trained who were not part of the research team. The reliability of the coded data was verified, followed by an inter-coder analysis to evaluate the agreement of the obtained results. Scott's Pi and Cohen's Kappa coefficients were used to measure this agreement, establishing the relationship between the coded data and the values that would be obtained by chance. The inter-coder analysis was conducted on a sample of 50 elements, and it was found that in all cases the values exceeded 0.80, with most cases having a value of 1. This latter value is expected in descriptive categories where subjectivity does not play a role.
For data processing and analysis, R version 3.5.1 was employed. Statistical decisions were made with a significance level of 0.05. Three levels were established for these decisions: 1) a highly strict level with a p-value <0.001 (represented as ***p); 2) an intermediate level with a p-value <0.05 (indicated as **p); and 3) a more lenient but still relevant level with a p-value <0.1 (designated *p). Frequency tables were generated, including both relative and absolute frequencies for qualitative variables. Additionally, summary statistic tables were obtained, providing measures such as sample size (N), median, and quartiles for quantitative variables. Bivariate tests were conducted for each variable, employing the chi-squared test for qualitative variables and the non-parametric Mann-Whitney test for quantitative variables. Bivariate tests were also performed to examine the relationship between the response variables (comments and likes) and each of the explanatory variables, using the non-parametric Mann-Whitney test or the Kruskal-Wallis test. For modelling the data, a multiple linear regression model was fitted for each response variable. Furthermore, the variables were transformed using logarithms. Therefore, the observed comments variable is defined as OCt = ln Ct, and the observed likes variable is defined as OLt = ln Lt.
All the initial explanatory variables that had a p-value less than 0.1 in the bivariate analysis were included in the analysis. The final model was obtained by eliminating all non-significant variables.

Results
With the aim of achieving the greatest solidity in our conclusions, the data was initially processed through a bivariate analysis on the impact generated by each of the variables in terms of comments and likes of the three models proposed; the data was then processed again through a multivariate analysis, thus achieving the maximum information from the coded data in order to be able to answer the research questions with the greatest soundness.

Results of the bivariate analysis
With the reference data, Chanel received a higher number of interactions for the three models than Dior for both response variables (comments and likes) ( Table 1).
For the Çukul (2015) model, it is significant that for both brands the Advertising variable had a negative impact on likes, with a high statistical significance in the case of Chanel and intermediate in the case of Dior. Posts whose format and/or intention was of an advertising nature generated fewer likes; however, there was no notable statistical significance in the generation of comments. The presence of the Product variable had a positive impact on comments for both brands, again with a greater significance in the case of Chanel. The presence of the product in the post generates an engagement that could be said to be more far-reaching if we consider that the comments require greater involvement and more time spent on them (John et al., 2017). No data for the Social responsibility variable was obtained for Chanel; in the case of Dior, however, posts with this variable were observed, having a positive impact on both response variables, though with a greater statistical significance for the comments than for the likes. Regarding this absence, it should be evaluated whether this represents a deliberate strategy on the part of Chanel. The study of the coded data following the application of the Goor (2012) model yielded results that, as a whole, were somewhat more relevant in number and statistical significance in this initial bivariate analysis, as can be seen in Table 2. It is notable how in this model, which analyses the impact of different post strategies, those posts that reflect the Emotions variable have a negative impact for both brands, both in terms of comments and likes, although in the case of Dior, no statistical significance was observed, whilst for Chanel, we obtained a high significance for likes and an intermediate level of significance for comments. The presence of the Self-efficacy variable also did not have a positive impact on the response variables; indeed, it generated a negative impact of relevant statistical significance, especially for Chanel. Similarly, the presence of the Sales Response variable for both brands had a negative impact on both comments and likes, in this case with a higher level of statistical significance for Dior than for Chanel. In contrast to this set of negative results, it is worth contrasting the positive impacts of the Relationship variable, especially for Chanel and Chanel likes, with posts that contain the Persuasion variable (not extrapolated to Dior). These two items -Relationship and Persuasion-highlight the potential of subjective elements that try to link the brand with the user or potential client.  Using the coded data, model 3 addresses aspects related to the format and elements present in the posts. The analysis highlights two variables -Text in image and Video-that have the same negative impact of high statistical significance on both likes and comments (Table 3). In addition to this result, some of the variables introduced into model 3 generate results with both positive and negative statistical significance. Among the positive ones, the only thing common to Dior and Chanel is the positive effect of the presence of the Clothes variable in generating more likes, although this impact is not seen for the comments. For Dior, the only other variable with a positive statistical significance in terms of generating comments is the use of Series of posts, a variable that, as will be seen later, is reinforced by the result of its impact in the multivariate analysis. For Chanel, the analysis yields better results, including the positive effect on both comments and likes of the presence of Accessories. The other variable that has significant effects on comments, in this case negative, is the presence of No product/evocative. Among the variables that have a positive impact on Chanel likes are People and People's faces (with a high level of significance that does not occur for Dior), and the presence of Location and Username mentioned, while the variables Studio and Make-up generate fewer likes.

Results of multivariate analysis
The multivariate approach modulates some of the already mentioned results of the bivariate analysis. In the case of Dior, the positive impacts of model 1 variables such as Product and Social responsibility are confirmed. The latter variable is even reinforced both for its positive impacts on comments and likes and for its level of significance. The negative effects of the presence of the Sales response variable and Video are also confirmed, as well as the positive effects on comments of the variable Series of posts and on likes of the presence of the Clothes variable. In the case of Chanel, the presence of Product is positive for comments, with Accessories being the only variable that has a positive impact on both comments and likes. The Series of posts variable has a negative impact on comments for Chanel (in contrast to Dior), while comments for Chanel are negatively affected by some variables already noted in the bivariate analysis, such as No product/evocative, Studio and Text in image, this latter variable also being negative for Chanel comments. Similarly, the negative effect of Emotions on Chanel likes is confirmed, with a high level of significance, while the role of Relationship in Chanel comments is weakened (Table 4).
The analysis of the data through the multivariate model for the likes response variable corroborates all the results from the bivariate analysis. In the case of Dior, the four variables with statistical significance in the multivariate modelling have the same positive or negative impact as in the bivariate analysis. For the variable Video, the results are confirmed with the same high level of statistical significance and clearly show a negative impact of the presence of this variable on the generation of likes. The positive impact of both the presence of the variables Clothes and Social responsibility on likes is confirmed. The same occurs for Chanel, with the results previously obtained in the bivariate analysis being confirmed. All the variables that emerge as significant in the multivariate modelling have the same impact (i.e. positive or negative) as in the previous bivariate analysis, in all cases with significant levels of statistical significance, with the Video variable drawing special attention, as its presence has a strong negative effect on the generation of likes (Table 5). "The fact that a video generates less engagement than a photo on Instagram may be due to various factors: a photo requires less attention and time to look at, as Instagram users tend to scroll quickly through their feed, dedicating limited time to each post. As such, photos are quicker to consume and capture attention more immediately, while videos require more time and engagement to be viewed in their entirety. Moreover, photos tend to capture specific moments, while videos can be longer and more narrative. In an environment in which attention is fleeting, it is possible that users prefer to interact with content that can be consumed quickly and that allows them to scroll easily to the next post. Another factor driving higher engagement with photo posts is the sound barrier, in the sense that videos often have audio components, which may limit their consumption." Table 6 shows a summary of the modelling results for both brands, both for comments and likes.

Discussion
The general aim of this study was to examine the relationships between the content, the elements present in the images, the communication strategies, the formats of the posts and the categories  of the products presented in Instagram posts and their impact on the generation of engagement. A total of 598 posts were analysed, concentrating on two luxury fashion brands and thus avoiding potential industry-specific biases. This study contributes to a broader understanding of engagement on Instagram in the context of the luxury sector by analysing an unusually large number of variables. Our results contribute both to previous literature and to clarifying important aspects in this emerging field of research. The findings are relevant since, according to the reviewed literature, most brands use Instagram for marketing purposes (Bug & Heene, 2020). However, only 33% of marketers make Instagram a high priority. This low use is possibly due to a lack of sufficient knowledge on how to translate its features into higher brand value (Digital Marketing Institute, 2018).
Of all the social networks, Instagram is the most effective in generating user participation (Eslami et al., 2022). For fashion companies, marketing through social media is taking a leading role, forcing them to take care of their messages on the networks in order to achieve engagement with their users.
The results reveal that posts attract different levels of engagement and in different ways (positive/negative) depending on the variables present in the posts. Our findings strongly show that, for both brands, the presence of the product in the posts can generate a higher volume of comments. Similarly, the presence of Social responsibility has a strong positive effect on the generation of both comments and likes, but in this case only for Dior. This variable is not present in Chanel posts, so we need to evaluate whether this responds to a specific strategy by Chanel; however, the results highlight this variable in the generation of engagement in terms of both likes and comments. Along the lines suggested by previous studies regarding consumer rejection of advertisements (Wei, 2022), our study shows that when the posts include the Sales response strategy, thereby seeking immediate sale, the impact on both comments and likes is negative. This strategy should therefore be avoided if user engagement is the intended outcome.
Contrary to expectations, the low impact on the engagement achieved through the variables People and People's faces is striking, since intuition would lead us to think that their presence humanizes the posts and generates user identification. Although the positive impact generated is of low statistical significance, in this regard, it should be noted that this result coincides with Lindell (2019), for whom photos that show faces attract more likes, and Yan et al. (2022), who conclude that photos with a face are more likely to receive likes.
Finally, our results also strongly show, due to its statistical significance and to the fact that it occurs for both brands, that posts in video format generate fewer likes.

Theoretical
This study contributes to the development of literature on marketing communication, specifically on social networks, by proposing a framework for studying the engagement generated by the posts of fashion brands on Instagram. This research framework integrates into a single investigation the study of the content of posts, the communication strategies, the elements present in the images and different formal elements of the posts, and the relationship of all these aspects with the engagement generated.
Our research framework integrates the category systems of Çukul (2015)and Goor (2012), to which we added our own category system that was specifically developed for the study of Instagram in the luxury fashion sector. This study is thus based on the analysis of posts on social networks based on soft and hard criteria.
This combination can be useful to replicate this framework in future research for these or other brands, and can even, with minor adaptations, be applied to other sectors or industries.

Managerial
There are various implications that may be of interest to social media marketing professionals in the luxury fashion sector. The findings obtained in this study should be taken into account in the management of Instagram posts for luxury brands in order to improve the engagement they generate.
The most significant finding is the low interaction that posts in video format generate. It would be speculative to pronounce on the reasons behind this, but its analysis would be an important subject of study for further research. It is equally notable that the presence of the product in posts increases engagement, particularly in terms of comments. Since comments are a type of engagement that requires more effort on behalf of the user, the fact that the presence of the product in the image increases the number of comments has significant implications. These results are therefore an invitation to the managers of luxury brand Instagram accounts to publish with greater focus on the product. And, as a complementary consideration, the fact that the displayed product is an accessory also generates greater engagement in terms of comments and likes, at least in the case of Chanel. This is also something to take into account for the communication of a sector with a large aspirational component and in which this product category is more affordable than, for example, dresses.
"Therefore, marketing managers can leverage CSR-related content, as this not only generates greater engagement but also helps construct a positive brand image. In addition, although videos are a popular content form on Instagram, they generate limited engagement among followers, something that marketing managers should take into account.
The results of this study can help businesses to have a clearer idea of the potential applications of Instagram, which could improve their understanding of their brand's impact on this social network." Regarding the communication strategy, it is evident that when it is based on communicating social responsibility actions, engagement increases significantly. We have been able to verify this in the case of Dior, although we did not observe any post that used this strategy in the case of Chanel. This result is an invitation to use this type of communication strategy if greater engagement with users is the intended outcome.

Limitations and directions for future research
Despite the time span of this study and the sampling base, the results obtained are still partial within the luxury fashion sector. It is difficult to extrapolate these results to other brand categories or other sectors.
Regardless, we have shown that the method applied is valid for obtaining wide-ranging results in variables of differing natures. The door is therefore open to extend this study to other time periods or other brands or sectors to broaden or compare the conclusions.
Based on the findings obtained here, in future the elements that bring about these behaviours could be investigated with a more user-focused type of research.