Mirela Sandu – Influence of Creative Ad Concepts on Social Media Advertising Performance

This blogpost was written by former ETHOS Lab Junior Researcher, Mirela Sandu.


Richard Elliot and Krisadarat Wattanasuwan have attributed the need for emotional content to advertising’s relationship to constructing self-identity. Self-identity is today not the consequence of a social system, but instead as something that individuals continuously create. Therefore, advertising and the brands one chooses to use, contribute to this creation of self-identity. (Wattanasuwan, 1998) This is also supported by a more recent study by Summers et. al, who shows that advertising impacts self-perception. For example, exposing someone to environmentally friendly toys might change their own perception of themselves to mirror that image. (Christopher A. Summers, 2016)

The point of departure of my master thesis was my intertwined interest in data analysis, advertising and constructing self-identity, but also from a practical need I have experienced throughout my working life: when it comes to social media advertising, very few know what they are actually doing. Particularly regarding paid content. (We will call the others just organic posts) I also liked the idea of using a mixed-methods approach, which gave me the flexibility of combining qualitative and quantitative methods where I found it relevant.

While previous research focused on increasing engagement of organic content on social media, little knowledge exists about optimizing content for paid advertising. Furthermore, instead of focusing on the number of likes, shares and comments, my main focus was on the number of clicks and ad receives, which is a step closer to actual web purchase. I examined this phenomenon on a single-case study in collaboration with Toy Maker X, which is the owner of two leading toy retailers. Multiple linear regression was one of the main methods used in this study, as a way to predict what type of content has the highest number of clicks.

The data (never) speaks for itself

While I had access to all the company‚Äôs social media advertising data up to that point, from both brands on 3 geographical markets: Denmark, Sweden and Norway, my data set was actually comprised of 618 observations. Due to the nature of the study, I have used open coding to label the content of each individual ad manually, based on its imagery and copyrighting. While the idea of automating this process was definitely appealing, the image software recognition I have tried was labelling my images as ‚Äúnon-human person‚ÄĚ, which was simply not specific enough, to say the least.

After repeating the process of finding the ad, enumerate codes based on the image/video and then text, I have noticed emerging patterns: whether an ad included a toy, a child, or whether it was highlighting more emotional/functional values. Thus, functional refers to advertisements focused on the brand’s product utility, such as toys that encourage creative development or learning, or have an emphasis on price, which is seen as a functional/rational argument. (Rodolgo Vazquez, 2010) Emotional arguments on the other hand, focus on the brand’s symbolic utility and storytelling, such as branding commercials, or a cute message meant to appeal to the viewer’s feelings. (ibid) The last attribute stands for ads that emphasize a fun experience or use humor in its text and/or image. An example is a video ad for roller shoes, where a host and two children are testing the product, and where the dialog is very witty and filled with jokes.

The result of the data processing (cleaning and labelling) was 3 data sets: one for each brandwith ads from Denmark, Norway and Sweden, and a joint data set of both brands. The data sets from the individual brands include 309 rows and 48 columns, whereas the combined data set has 618 rows and 48 columns.

Therefore, several regression models were run in an iterative way. This facilitated the comparison between the different brands’ performance across countries and the continuous improvement of the model.

Results

The table below shows a comparison between the last versions of the predictive models for each brand, and the joint data set. The first columns shows the predictor variable, and the rest of the columns the coefficient (impact) of each predictor variable over the dependent variable, which is number of clicks.
While there are differences in the variables’ coefficients estimate, the results are consistent across models in terms of showing a positive or a negative correlation.

Dependent variable: Clicks (All)
Predictor variable Brand Toy 1 Brand Toy 2 Brand Toy 1 + Brand Toy 2
Coefficient estimate Coefficient estimate Coefficient estimate
(Intercept) -300** -15.40 277.00
ObjectiveTraffic      
Reach -0.0045*** -0.01*** -0.007***
Frequency -75.40*** -48.40*** -88.60***
Amount Spent (DKK)` 0.23*** 0.136*** 0.21***
`Relevance Score`   99.50*** 110.00*** 92.20***
ObjectiveReach -489.00***    
MonthOngoing 913.00***    
Discount 436.00***   747.00***
Seasonal -248.00*** -259.00** -357.00***
ObjectiveVideoViews      
ObjectiveCONVERSIONS      
Sensational     390.00***
‘Brand focus’   429.00***  
Toy   -743.00***  
Disney      
LEGO      
Children   -199.00*  
Formatmorethanthreepictures      
Formatonepicture      
Formatvideo      
Functional   334.00***  
Emotional 263.00***   316.00***
Funny   -185.00*  
Impressions   0.004***  
Offer   203.00*  
Genderm     -162.00.
Genderw -193*   -231.00*
Specifictoy_Trampoline   -843.00***  
Specifictoy_Animalcostume   1450.00*** 448.00*
`Specifictoy_creativetoys` -288.00**   -307.00**
`Specific_toy_boardgames`     -271.00.
`Specific_toy_popular`     -727.00***
Audience_page.liked   501.00*** 318.00**
Audience_wca.productview30 2100.00*** 1110.00*** 1886.00***
Month dummies? Yes Yes Yes
Multiple R-squared: 0.8126 0.6655 0.813
Adjusted R-squared           0.8076 0.6426 0.8038
Observations: 309 309 618
*p < 0.05   **p <0.01  ***p<0.001 .p<0.1

An interesting finding is that even though budget increases the total number of clicks, the type of content also has a significant impact. For example, as seen in the last model, spending 5DKK extra would lead to one more ad click, whereas content labeled as emotional leads to 316 more clicks. Having a discount in an ad would lead to an increase of 747 in the number of clicks, making it a variable with a very strong positive impact over the dependent variable. The model also shows that gender targeting is only limiting the audience instead of improving the ads‚Äô performance. However, this could be because currently the ads are not customized to specific audiences. Instead, it is more effective to target people who have visited the website in the past 30 days, which leads to an increase of 1886 in ad clicks ‚Äď the most effective targeting option based on my analysis.

An important part of the research’s findings is identifying the differences between the two brands. Firstly, there are differences between which predictor variables have an impact over the total number of clicks. For Toy brand 1 for example, there was not enough statistical evidence to show a relationship between sensational content, brand focus, toy, and mentioning Disney or LEGO, and the total number of clicks. Furthermore, Toy brand 1’s ads are performing better when they include emotional content, often related to the brand’s symbolic utility. (Rodolgo Vazquez, 2010) In contrast, Toy brand 2’s target group has a significant preference for functional content. Interestingly enough, funny content has a negative effect over the total number of clicks for Toy brand 2, suggesting that the brands’ fans are not interested in humorous messages, but instead in the functional utility of its products.

Interestingly, children present in the visuals of an ad actually have a negative impact over the total number of clicks. This could be because of the audience’s preference for functional content, whereas children imagery is most often closer to the emotional dimension (which resonates better to Toy brand 1’s audience according to the analysis).

Reflections

Central to the concept of branding is that consumers buy products not only based on their functional attributes, but also on their symbolic utilities. According to Elliot et. al, symbolic meanings have two effects: social-symbolism, constructing the social world, and self-symbolism, constructing self-identity. Thus, modern consumption and advertising revolve around the idea that consumers use products, services and media to express who they are. In addition, the development of self-identity is intertwined with social-identity, as one’s image of the self needs to be validated through social interaction. (Jenkins in Elliot:133) Shared meanings among consumers are created through the process of socializing.

From this perspective, the importance and the effect of advertising in society, on a platform labelled as social, becomes even more important to discuss. From my research, I have noticed that both brands are performing targeted advertising based on gender. While none of the ads have been customized to their audience, doing so could reinforce stereotypes in our societies. This issue becomes even more sensitive as the end users are children, who mostly rely on their parents to choose their toys for them. Furthermore, even though restricted, in some markets advertising to children is actually allowed, which means that companies could choose to advertise dolls to girls and STEM toys to boys, thus reinforcing existing gender bias.

Advertising has been wildly recognized for its effectiveness in creating and changing cultural trends. Within the toy industry, this can mean re-enforcing existing stereotypes, or it can strive to promote values which can bring our collective social identity forward. This responsibility should be equally shared by toy manufacturers, retailers and advertisers, which have the opportunity to shape the way children play and learn.

List of references

Christopher A. Summers, R. W. (2016). An Audience of One: Behaviorally Targeted Ads as Implied Social Labels. Journal of Consumer Research, 156-178.

Edith G. Smit, G. V. (2014). Understanding online behavioural advertising: User knowledge, privacy concerns and online coping behaviour in Europe. Computers in Human Behavior, 15-22.

Wattanasuwan, R. E. (1998). Brands as symbolic resources for the construction of identity. International Journal of Advertising, 131-144.

Rodolgo Vazquez, A. B. (2010). Consumer-based Brand Equity: Development and Validation of a Measurement Instrument. Journal of Marketing Management, 27-48.