Volume 32, Issue 2 p. 336-349
RESEARCH REPORT
Open Access

Content Hungry: How the Nutrition of Food Media Influences Social Media Engagement

Ethan Pancer

Corresponding Author

Ethan Pancer

Saint Mary’s University

Correspondence concerning this article should be addressed to Ethan Pancer, Sobey School of Business, Saint Mary’s University, 923 Robie Street, Sobey 147, Halifax, NS B3H 3C3, USA. Electronic mail may be sent to [email protected]

Search for more papers by this author
Matthew Philp

Matthew Philp

Ryerson University

Search for more papers by this author
Maxwell Poole

Maxwell Poole

Saint Mary’s University

Search for more papers by this author
Theodore J. Noseworthy

Theodore J. Noseworthy

York University

Search for more papers by this author
First published: 20 April 2021
Citations: 8

The authors thank Megan Pollock for her assistance with data collection and gratefully acknowledge SSHRC (#435-2018-1319 and #430-2019-00218) for their financial support.

Accepted by Priya Raghubir and Thomas Kramer, Editors; Associate Editor, Brent McFerran

Abstract

What motivates people to consume and engage with food media on social networks? We adopt an evolutionary lens to suggest that the valence of people’s affective state varies by the implied caloric density of food media, which has a direct impact on social media engagement. First, we analyze a catalog of Buzzfeed’s Tasty videos based on nutritional content derived from the dish’s ingredients and find that visualizing caloric density (i.e., calories per serving) positively influences likes, comments, and shares on Facebook. We then replicate this phenomenon in an experiment, providing preliminary evidence for the role of affect as an explanatory mechanism. We conclude by isolating the role of affect with a classic misattribution task, which attenuates the elevated engagement resulting from exposure to calorie-dense food media. These findings contribute to the dialogue on the antecedents of social media engagement and offer implications for content developers, advertisers, consumer health advocates, and policymakers.

The social media landscape has fundamentally altered how consumers are exposed to food. Users are inundated with visual displays of food, with over 400 million posts on “#food” and 250-million on “#foodporn” on Instagram alone (Instagram, 2021). Perhaps most notably, Buzzfeed’s Tasty has become the world’s largest digital food network, amassing over 100-million followers on Facebook and over 2 billion views monthly (Tubular, 2020). Given the ubiquity of food media online, understanding the specific characteristics that shape engagement is of critical importance to content producers looking to tailor media toward viewer preferences, advertisers seeking to increase impact, and health advocates interested in helping consumers make better food choices.

While extant literature has focused extensively on food media (Chandon & Wansink, 2012; Hagen, 2021; Wadhera & Capaldi-Phillips, 2014) or digital engagement (Lamberton & Stephen, 2016; Wilson, Gosling, & Graham, 2012), there is a paucity of work integrating these domains to examine the specific characteristics of food media that predict digital engagement. In the current research, we focus our investigation on the nutritional makeup of dishes depicted in food media. In doing so, we test a novel proposition—whether the caloric density of food media influences social media engagement (i.e., likes, comments, shares).

From a nutritional perspective, humans have long sought foods with characteristics that the brain instinctively recognizes as valuable. Evolutionarily, the ability to visually recognize calorie-dense foods was advantageous during foraging (Allman & Martin, 1999; Drewnowski & Almiron-Roig, 2009; Gehring, 2014). Since human brains evolved to learn that seeing calorie-dense foods typically precedes pleasurable and satiating consumption (Harrar, Toepel, Murray, & Spence, 2011; Toepel, Knebel, Hudry, le Coutre, & Murray, 2009), humans have developed a desire to visually attend to food (Spence, Okajima, Cheok, Petit, & Michel, 2016). To this day, finding and eating calorie-dense foods typically makes people feel good (Drewnowski, 1997; Moss, 2013), releasing dopamine and stimulating pleasure centers of the brain (Volkow, Wang, & Baler, 2011). Indeed, the activation of food-related schemas can alter emotional states, physiological responses, and desires (Collins & Stafford, 2015; Evers, Dingemans, Junghans, & Boevé, 2018; Hagen, Krishna, & McFerran, 2017, 2019; Hingston & Noseworthy, 2018, 2020; McFerran, Dahl, Fitzsimons, & Morales, 2010a, 2010b). For example, picturing oneself eating a steak, which is calorie-dense, can increase salivation and the desire to consume (Dadds, Bovbjerg, Redd, & Cutmore, 1997). Based on these findings, we suggest that nutritional content can be broadly gauged based on a dish’s appearance and that exposure to calorie-dense dishes can elevate affect.

When it comes to influencing consumers’ online behaviors, the link between affect and digital engagement is well documented (Eigenraam, Eelen, Van Lin, & Verlegh, 2018; Moore & Lafreniere, 2020). Positive content is more likely to go viral (Berger, 2014; Berger & Milkman, 2012). Additionally, positive and exciting consumer experiences are more likely shared (Berger, 2011; De Angelis, Bonezzi, Peluso, Rucker, & Costabile, 2012). More specifically, social media content that makes consumers feel good increases the likelihood of being liked, commented upon, and shared (Pancer, Chandler, Poole, & Noseworthy, 2019).

Taken together, prior research suggests that people should feel positively when viewing calorie-dense foods because they can forecast the pleasantness of the meal based on its appearance. Furthermore, when something makes people feel good, they are more likely to engage with it. We focus our investigation on digital behaviors associated with social interactions (e.g., likes, comments, and shares). Although likes, comments, and shares can differ in terms of underlying motivations and visibility, prior work has demonstrated that positive affect can motivate all of them (Pancer et al., 2019). Thus, we predict that visual exposure to food media that looks calorie-dense (vs. calorie-light) will drive social media engagement and that this process is based on the users’ positive affect.

Overview of the Studies

We report the results of a field study and two experiments. In Study 1, we examine social media posts from a digital food network to determine whether the caloric density of meals depicted influences social media engagement. We find that visualizing caloric density is positively associated with likes, reactions, comments, and shares. In Study 2, we replicate the phenomenon in an experiment and provide preliminary evidence for the mediating role of affect. In Study 3, we further isolate the role of affect by demonstrating that a classic affect misattribution task can attenuate engagement intentions for calorie-dense food media.

Study 1

Method

Dataset

We constructed a dataset by harvesting posts from Buzzfeed’s Facebook page Tasty (www.facebook.com/buzzfeedtasty), a popular food-centric page featuring time-lapse meal assembly videos. While the meals vary, the consistent video style makes it an ideal context for determining the influence of nutritional content on social media engagement.

Our dataset consisted of every video post from Tasty’s Facebook page over a 5-year period (July 31, 2015 – October 10, 2020) that featured the preparation of a single dish, yielding a corpus of 721 posts (MDA 1a for sample construction). Posts were downloaded using a custom Python script that interacted with the Facebook Graph API, allowing us to extract counts of likes, reactions (93.3% of which were likes), comments, and shares as well as the number of times each video was viewed along with other variables associated with the post (Table 1).

Table 1. Study 1 – Summary statistics
Mean SD
Outcome variables
Number of Likes 196,761.01 214,711.88
Number of Reactions 210,895.81 230,195.17
Number of Comments 26,236.70 42,451.58
Number of Shares 263,984.61 516,154.00
Likes (log-transformed) 11.36 1.68
Reactions (log-transformed) 11.44 1.66
Comments (log transformed) 9.12 1.80
Shares (log transformed) 11.03 2.08
Predictor variables
Caloric density (calories per serving) 507.14 288.51
Visual perception controls
Protein (g) 127.96 143.16
Sugars (g) 148.84 234.01
Fiber (g) 22.45 20.22
Nonsaturated Fats (g) 114.89 153.61
Sodium (mg) 5,498.81 6,897.81
Number of Servings 7.90 6.00
Number of Video Views 23,462,127.71 28,011,482.21
Food preparation controls
Number of Ingredients 9.29 4.17
Number of Preparation Steps 10.46 3.91
Appetizer (%) 11.4% 31.8%
Breakfast (%) 6.8% 25.1%
Desserts (%) 25.4% 43.6%
Dinner (%) 29.2% 45.5%
Drinks (%) 2.2% 14.6%
Lunch (%) 7.4% 26.2%
Sides (%) 3.8% 19.2%
Snacks (%) 7.0% 25.6%
Video-specific controls
Video Duration (seconds) 93.41 62.10
Number of times video was posted 2.07 1.99
Sample: 721

Variables

The Facebook posts contained no explicit nutritional information. Therefore, to extract nutritional information for each dish, we aggregated data from two sources. First, we used the unique identifier embedded in each post to scrape the corresponding recipe from the Tasty website (tasty.co), which provided the calories per serving, number of servings, as well as a text-based ingredient list. Second, we uploaded the ingredients to Nutritionix.com, which uses a natural language processing algorithm to extract nutritional information from text-based ingredients. For example, “1 cup milk” would return a result of 4.7 grams of total fat, 12 grams of carbohydrates, 8.5 grams of protein, and other nutritional outcomes. Aggregating data from these sources allowed us to create a dataset that matched nutritional information with each corresponding video post.

We controlled for two broad categories of covariates: food preparation characteristics and video-specific features. We accounted for the complexity of the dish by counting total ingredients and preparation steps to mitigate concerns that the relative simplicity of a meal could explain our findings. Tasty also classified each meal into one of eight types: appetizer, breakfast, dessert, dinner, drinks, lunch, sides, or snacks. Accounting for this helped control for any influence a particular meal occasion has on engagement. With respect to video features, we accounted for overall video length, the number of times a video was posted to the page, and release date timing controls were recorded to control for variance associated with video durations, repeat viewing, and timing effects.

Results and Discussion

Several videos generated a disproportionately high number of likes, reactions, comments, and shares, which positively skewed our data. To account for this, we calculated the natural logarithm of each outcome as dependent variables in a series of OLS regression analyses. All four engagement outcomes were significantly and positively correlated (Table 2), which supported the notion that these variables share a commonality.

Table 2. Study 1 – Correlation matrix
Variable 2 3 4 5 6 7 8 9 10 11 12
1 Calories 0.37** 0.07 0.35** 0.37** 0.35** −0.35** −0.03 −0.07 −0.06 −0.02 −0.06
2 Protein −0.02 0.01 0.48** 0.73** 0.02 0.02 −0.04 −0.04 −0.01 −0.02
3 Sugars −0.09* 0.25** −0.01 0.27** −0.08* −0.15** −0.14** −0.17** −0.16**
4 Fiber −0.09* 0.01 −0.32** −0.01 −0.08* −0.08* −0.07 −0.07
5 Nonsaturated fats 0.40** 0.10** −0.02 −0.13** −0.13** −0.11** −0.12**
6 Sodium −0.01 −0.02 −0.10** −0.10** −0.07 −0.08*
7 # Servings 0.00 0.00 0.01 0.00 0.00
8 Views 0.56** 0.57** 0.59** 0.63**
9 Likes 0.99** 0.95** 0.97**
10 Reactions 0.96** 0.98**
11 Comments 0.96**
12 Shares
  • Engagement outcomes are log-transformed.
  • ** p < .01;
  • * p < .05.

We first regressed our engagement measures on the broad basket of calories in a simple OLS regression (Tables 3–6; Model 1). This model revealed no evidence of a significant effect for any engagement variable (Fs < 3.14, ps > .05). Given that calories are merely a measurement of energy derived from either carbohydrates, protein, or fats, our next step was to explore whether the components may have a competing influence on engagement. While each gram of carbohydrate or protein yields 4 calories, each gram of fat yields 9 calories (FDA, 2021). Critically, these components are not equally distinguishable by sight alone. Fats (particularly saturated fats) tend to manifest more visually in food (DiFeliceantonio et al., 2018; Heinze, Preissl, Fritsche, & Frank, 2015; Toepel et al., 2009). Fats often give food their visual structure and texture (Teicholz, 2014). Moreover, perceived fat predicts inferences about weight gain, particularly when no explicit nutritional information is provided (Oakes, 2005; Oakes & Slotterback, 2005). Thus, not only are meals high in fat objectively more calorie-dense, they also tend to be perceived as being more calorie-dense.

Table 3. Study 1 – The effect of caloric density on likes (log-transformed)
OLS regression, likes
(1) (2) (3) (4)
Calories −0.066 0.083* 0.094* 0.081*
(.076) (.044) (.026) (.040)
Visual perception
Protein (g) 0.005 0.060 0.037
(.919) (.218) (.413)
Sugar (g) −0.111** −0.115** −0.082*
(.002) (.005) (.029)
Fiber (g) −0.106** −0.094** −0.093**
(.002) (.008) (.005)
Nonsaturated Fats (g) −0.098* −0.097* −0.082*
(.013) (.014) (.024)
Sodium (mg) −0.099* −0.091* −0.040
(.019) (.037) (.322)
Number of Servings 0.053 0.052 0.035
(.155) (.179) (.331)
Number of Video Views 0.551** 0.550** 0.537**
(.000) (.000) (.000)
Food preparation
Number of Ingredients −0.064 −0.017
(.065) (.600)
Number of preparation steps −0.094** 0.001
(.005) (.965)
Appetizer (%) −0.019 −0.075
(.713) (.110)
Breakfast (%) −0.041 −0.032
(.351) (.431)
Desserts (%) −0.002 −0.022
(.975) (.669)
Dinner (%) −0.060 −0.092
(.371) (.141)
Drinks (%) −0.017 0.006
(.618) (.848)
Lunch (%) −0.026 −0.050
(.570) (.245)
Sides (%) 0.004 −0.013
(.923) (.724)
Snacks (%) 0.006 −0.045
(.888) (.253)
Video-specific
Video duration (seconds) −0.239**
(.000)
Number of video postings −0.122**
(.000)
Timing (Days) 0.061*
(.031)
Timing (Day of Week) −0.034
(.227)
Timing (Day of Year) −0.171**
(.000)
R 2 0.004 0.354 0.370 0.466
R2 Change 0.349** 0.016 0.096**
  • Reporting standardized beta coefficients with p values in parentheses
  • p < .10;
  • * p < .05;
  • ** p < .01.
Table 4. Study 1 – The effect of caloric density on reactions (log-transformed)
OLS regression, reactions
(1) (2) (3) (4)
Calories −0.061 0.089* 0.099* 0.085*
(.102) (.032) (.020) (.031)
Visual perception
Protein (g) 0.005 0.060 0.038
(.918) (.214) (.403)
Sugar (g) −0.107** −0.113** −0.081*
(.003) (.005) (.031)
Fiber (g) −0.109** −0.097** −0.095**
(.001) (.007) (.004)
Nonsaturated Fats (g) −0.095* −0.094* −0.079*
(.015) (.017) (.030)
Sodium (mg) −0.099* −0.091* −0.041
(.019) (.038) (.321)
Number of Servings 0.055 0.053 0.036
(.141) (.171) (.317)
Number of Video Views 0.556** 0.554** 0.541**
(.000) (.000) (.000)
Food Preparation
Number of Ingredients −0.064 −0.018
(.064) (.577)
Number of Preparation Steps −0.088** 0.005
(.008) (.886)
Appetizer (%) −0.019 −0.075
(.702) (.110)
Breakfast (%) −0.042 −0.034
(.331) (.404)
Desserts (%) 0.002 −0.019
(.977) (.720)
Dinner (%) −0.061 −0.093
(.363) (.138)
Drinks (%) −0.017 0.006
(.624) (.849)
Lunch (%) −0.027 −0.051
(.556) (.239)
Sides (%) 0.004 −0.012
(.910) (.741)
Snacks (%) 0.004 −0.047
(.929) (.240)
Video-Specific
Video Duration (seconds) −0.236**
(.000)
Number of Video Postings −0.118**
(.000)
Timing (Days) 0.060*
(.033)
Timing (Day of Week) −0.033
(.241)
Timing (Day of Year) −0.168**
(.000)
R 2 0.004 0.357 0.372 0.465
R2 Change 0.353** 0.015 0.093**
  • Reporting standardized beta coefficients with p values in parentheses.
  • p < .10;
  • * p < .05;
  • ** p < .01.
Table 5. Study 1 – The effect of caloric density on comments (log-transformed)
OLS regression, comments
(1) (2) (3) Final
Calories −0.017 0.119** 0.134** 0.122**
(.654) (.003) (.001) (.002)
Visual perception
Protein (g) 0.009 0.048 0.030
(.846) (.309) (.508)
Sugar (g) −0.132** −0.119** −0.088*
(.000) (.003) (.020)
Fiber (g) −0.105** −0.099** −0.097**
(.002) (.005) (.003)
Nonsaturated Fats (g) −0.072 −0.066 −0.050
(.063) (.089) (.169)
Sodium (mg) −0.076 −0.069 −0.027
(.068) (.110) (.505)
Number of Servings 0.062 0.067 0.050
(.090) (.080) (.167)
Number of Video Views 0.572** 0.576** 0.561**
(.000) (.000) (.000)
Food preparation
Number of Ingredients −0.061 −0.022
(.073) (.499)
Number of Preparation Steps −0.105** −0.027
(.001) (.403)
Appetizer (%) 0.029 −0.023
(.562) (.622)
Breakfast (%) −0.037 −0.037
(.382) (.357)
Desserts (%) 0.017 −0.009
(.763) (.859)
Dinner (%) 0.000 −0.036
(.999) (.566)
Drinks (%) 0.025 0.046
(.469) (.152)
Lunch (%) −0.012 −0.036
(.798) (.402)
Sides (%) 0.014 −0.004
(.725) (.918)
Snacks (%) 0.025 −0.022
(.549) (.572)
Video-specific
Video Duration (seconds) −0.224**
(.000)
Number of Video Postings −0.076*
(.013)
Timing (Days) 0.082**
(.004)
Timing (Day of Week) −0.050
(.081)
Timing (Day of Year) −0.137**
(.000)
R 2 0.004 0.357 0.372 0.465
R2 Change 0.353** 0.015 0.093**
  • Reporting standardized beta coefficients with p values in parentheses.
  • p < .10;
  • * p < .05;
  • ** p < .01.
Table 6. Study 1 – The effect of caloric density on shares (log-transformed)
OLS regression, shares
(1) (2) (3) (4)
Calories −0.057 0.059 0.081* 0.068
(.126) (.128) (.041) (.065)
Visual perception
Protein (g) 0.019 0.069 0.049
(.651) (.127) (.243)
Sugar (g) −0.102** −0.091* −0.057
(.002) (.016) (.105)
Fiber (g) −0.083** −0.086** −0.085**
(.010) (.010) (.006)
Nonsaturated Fats (g) −0.080* −0.076* −0.063
(.032) (.039) (.065)
Sodium (mg) −0.071 −0.066 −0.022
(.074) (.105) (.571)
Number of Servings 0.042 0.050 0.033
(.235) (.163) (.326)
Number of Video Views 0.621** 0.624** 0.619**
(.000) (.000) (.000)
Food preparation
Number of Ingredients −0.044 0.002
(.176) (.935)
Number of Preparation Steps −0.150** −0.054
(.000) (.079)
Appetizer (%) −0.020 −0.071
(.671) (.107)
Breakfast (%) −0.049 −0.043
(.232) (.259)
Desserts (%) −0.003 −0.027
(.952) (.583)
Dinner (%) −0.045 −0.076
(.474) (.192)
Drinks (%) −0.030 −0.009
(.363) (.761)
Lunch (%) −0.014 −0.037
(.750) (.356)
Sides (%) 0.015 −0.001
(.687) (.967)
Snacks (%) −0.001 −0.051
(.973) (.174)
Video-specific
Video Duration (seconds) −0.222**
(.000)
Number of Video Postings −0.137**
(.000)
Timing (Days) 0.066*
(.013)
Timing (Day of Week) −0.039
(.144)
Timing (Day of Year) −0.126**
(.000)
R 2 0.003 0.424 0.450 0.531
R2 Change 0.421** 0.025** 0.081**
  • Reporting standardized beta coefficients with p values in parentheses.
  • p < .10;
  • * p < .05;
  • ** p < .01.

To account for the perceptibility of caloric density, our second model partialled out the effects of the various nutritional components that are not easily discernable by sight. This included aspects of carbohydrates (sugars and fiber), protein, nonsaturated fats, and sodium. While caloric density is reported at the per serving level, it is visualized in the aggregate, requiring the number of servings to be accounted for. Given the nature of online media, the model also controls for visual exposure using the number of video views. This model (Model 2) revealed a significant effect of caloric density, controlling for nonvisual aspects, on likes, F(8, 713) = 48.77, p < .001, R2 = .354, reactions, F(8, 713) = 49.38, p < .001, R2 = .349, comments, F(8, 713) = 52.76, p < .001, R2 = .372, and shares, F(8, 713) = 65.63, p < .001, R2 = .424). Caloric density was a significant predictor for three of the four measures of engagement (likes: β = .08, p = .044; reactions: β = .09, p = .032; comments: β = .12, p = .003). The only exception was for shares (β = .06, p = .128). Most of the less-visible nutrients including sugar, fiber, nonsaturated fats, and sodium had significant negative influences on each engagement outcome. Unsurprisingly, the number of video views also had a consistent significant positive impact (likes: β = .55, p < .001; reactions: β = .56, p < .001; comments: β = .57, p < .001; shares: β = .62, p < .001).

In our third model, we included all food preparation characteristics as controls to examine the robustness of Model 2 (i.e., ingredients, steps, and meal occasion). The results were significant for all engagement variables (likes: β = .09, p = .026; reactions: β = .10, p = .020; comments: β = .13, p = .001; shares: β = .08, p = .041).

Finally, in our fourth and final model, we controlled for video-specific characteristics (i.e., duration, # of reposts, and release timing). The results were robust (likes: β = .08, p = .040; reactions: β = .09, p = .031; comments: β = .12, p = .002; shares: β = .07, p = .065). Based on the average Tasty video, the addition of a mere 10 calories per serving, controlling for nonvisual factors of influence, corresponds to an increase of 4,193 likes, 1,495 comments, and 3,506 shares. See MDA 1b for evidence that the patterns in Tables 3 through 6 are robust to numerous alternative model specifications, with the exception of shares in the raw data since shares increase views and views were controlled for in the model.

What could not be addressed in this field data was whether positive affect underscored this phenomenon. Study 2 was designed to test this assumption.

Study 2

Method

Our intent with Study 2 was to replicate the findings of the field study in a more controlled setting. In doing so, we sought to experimentally test whether certain types of food media influence positive affect, and, in turn, influence social media engagement.

Design and procedure

Two-hundred and twenty-nine participants (36.6% female; Mage = 39.8) recruited through Amazon’s Mechanical Turk were assigned at random to one of three conditions (No-Video Control vs. Calorie-Light Video vs. Calorie-Dense Video) in a between-subjects design.

Participants in either the Calorie-Light or Calorie-Dense video conditions were exposed to a 30-s overhead food preparation video chosen from tasty.co. Past experimental work on the consequences of caloric density often compared a burger (calorie-dense) with a salad (calorie-light) (Chernev, 2011; Chernev & Gal, 2010; Romero & Biswas, 2016; Wilcox, Vallen, Block, & Fitzsimons, 2009). Consistent with the paradigm, a burger and a salad video were selected from Buzzfeed’s Tasty that had similar filming styles, preparation speeds, and lengths, but varied in their actual caloric density (MDA 2a for stimuli).

Measures

Following the video, participants rated their likelihood (anchored: 1 = Extremely unlikely; 7 = Extremely likely) to “Click the thumbs up ‘like’ button”; “Comment on this post”; “Share this on your own Facebook feed”; “Share this with specific friends on Facebook”; and “Follow this Facebook page”. Items were averaged to create an engagement likelihood index (α = .94). All participants then completed an affective state scale (Noseworthy, Di Muro, & Murray, 2014) which consisted of four bipolar items on a 9-point scale, completing the phrase, “I currently feel…” unpleasant/pleasant; negative/positive, sick/fine, sad/happy (α = .94) before completing a demographic questionnaire (see MDA 2a for measures).

Results and Discussion

A one-way ANOVA revealed a significant omnibus effect on affect, F2,226 = 3.28, p = .040, η2 = .03. Planned contrasts confirmed that affect was more positive after watching the calorie-dense video (M = 7.72, SD = 1.49) relative to both the calorie-light (M = 7.10, SD = 1.70), t226 = 2.40, p = .017, d = .39, and no-video control (M = 7.21, SD = 1.70), t226 = 1.92, p = .056, d = .32. The calorie-light video had no significant effect on affect relative to the no-video control, p = .682. This confirmed that positive affect was being elevated in the calorie-dense condition.

Next, we explored engagement intentions between the calorie-dense and calorie-light video conditions. The results revealed a significant effect of caloric density on engagement intentions such that participants reported having higher engagement intentions after viewing the calorie-dense (M = 4.51, SD = 1.77) versus calorie-light video (M = 3.40, SD = 1.94), t157 = 3.77, p < .001, d = .60.

A follow-up simple mediation analysis (Hayes, 2017; Model 4, 10,000 draws) confirmed a significant indirect effect of caloric density (0 = calorie-light, 1 = calorie-dense) on social media engagement intentions through affect (CI95%: .020, .472).

These results were consistent with those of the field study. The caloric density of food media is positively related to engagement intentions. This study also provides preliminary evidence of affect as an explanatory mechanism. The primary substantive takeaway was that, again, consumers were more likely to engage with calorie-dense than calorie-light food media. However, the role of affect in this design is correlational. Our next step was to explore whether we could reduce the correlation between affect and engagement and thus test a fundamental parameter in the causal pathway.

Study 3

Our intent with Study 3 was to test whether the link between caloric density and engagement is predicated on attributing positive affect to the food media. To do so, we employed a classic misattribution paradigm (Proulx & Heine, 2008; Taylor & Noseworthy, 2020). Specifically, we tested the possibility that attributing positive affect to a different source may dampen intentions to engage with calorie-dense food media, while having little to no effect on calorie-light foods. In doing so, this would not only indirectly manipulate the correlated nature of the mechanism, but also provide insight into the degree to which participants are cognizant of what drives their online engagement.

Method

Participants and design

Two-hundred and forty-six participants (50.4% female; Mage = 43.9) recruited through Amazon’s Mechanical Turk were assigned at random to one of four conditions in a 2 (Food Media: Calorie-Light vs. Calorie-Dense) × 2 (Misattribution: Present vs. Absent) between-subjects factorial design.

Stimuli

To avoid mono-operationalization and conceptually replicate the effects of Study 2 with another subset of videos from Tasty, we selected a new calorie-dense stimulus that had precedence in the literature: chocolate cake (Chandon & Wansink, 2007; Chernev, 2011; Shiv & Fedorikhin, 1999; Wertenbroch, 1998), specifically a chocolate hazelnut mug cake (729 calories per serving). We also selected a different calorie-light stimulus, specifically lemon pepper cauliflower bites (148 calories per serving). To mitigate the likelihood of a third factor explaining our results, the videos were edited to be the same duration and played the same music.

Procedure

Participants were informed that the study examined the effects of sounds on consumption. Participants were asked to turn on their computer’s sound and informed that they will watch a food preparation video. All participants were then warned that some videos may inadvertently contain high frequency sounds that are not perceptible to the human ear. This was a guise that served as the foundation for the misattribution task (Payne, Cheng, Govorun, & Stewart, 2005; Taylor & Noseworthy, 2020)—in reality, participants were not exposed to any sound other than the video’s music. Consistent with prior procedures (Taylor & Noseworthy, 2020), participants in the misattribution condition then read that high frequency sounds “can impact a person’s emotional state, thus influencing their judgment.” By contrast, participants in the no misattribution condition read that high frequency sounds “have no discernible impact on people.” All participants were then exposed to one of the aforementioned calorie-light or calorie-dense food preparation videos.

The videos were pretested to confirm differential effects on affect (MDA 3a). We relegated our test of affect to a pretest rather than collecting in the actual study because misattribution tasks follow an indirect paradigm—that is, it assumes that while affect is consciously discernable, the link from affect to engagement is not. This reasoning is based on recent evidence (Pancer et al., 2019). Thus, explicitly asking affect in the context of a misattribution task could lead to distortions due to self-presentation (Fazio, Jackson, Dunton, & Williams, 1995; Teige-Mocigemba, Becker, Sherman, Reichardt, & Klauer, 2017) or biases due to introspective limits (Nisbett & Wilson, 1977).

Participants responded to the same engagement intentions scale (α = .94) as Study 2. To control for rival explanations on the visual appearance of food, we also included a battery of scales on food perception (Hagen, 2021), where participants also rated the perceived healthiness, naturalness, amount, freshness, tastiness, prettiness, and orderliness of the meal (MDA 3b for measures) before completing a demographic questionnaire.

Results and Discussion

A two-way ANOVA confirmed that the calorie-dense dish was perceived as being higher in calories than the calorie-light dish (MDense = 6.11 vs. MLight = 2.22), F1,242 = 852.36, p < .001, η2=.78. There was no main effect of misattribution (p = .61) or its interaction with caloric density (p = .47).

A two-way ANOVA on engagement intentions yielded a significant main effect of food media, such that participants were more likely to engage with the calorie-dense video (M = 3.51, SD = 1.76) relative to the calorie-light video (M = 2.76, SD = 1.49), F1,242 = 13.25, p < .001, η2 = .05. Participants were also more likely to engage in the control condition (M = 3.37, SD = 1.69) relative to the misattribution condition (M = 2.91, SD = 1.63), F1,242 = 4.62, p = .033, η2 = .02. As predicted, these main effects were qualified by a significant interaction, F1,242 = 4.20, p = .041, η2 = .02. As illustrated in Figure 1, participants in the control condition were more likely to engage with the calorie-dense video (M = 3.94, SD = 1.76) relative to calorie-light video (M = 2.77, SD = 1.39), F1,242 = 16.05, p < .001, η2 = .06. However, participants in the misattribution condition did not differ in engagement intention across the two food media conditions (MDense = 3.08 vs. MLight = 2.75), p = .26. Indeed, the misattribution task attenuated engagement intentions for calorie-dense food media (MControl = 3.94, SD = 1.76; MMisattribution = 3.08, SD = 1.66), F1,242 = 8.89, p = .003, η2 = .04, but had no effect for calorie-light food media (MControl = 2.77; MMisattribution = 2.75), p = .95. When controlling for the food perception measures, the effect still holds (see Table 7 for contributing weights). This suggests that the positive response to calorie-dense food media is the result of affect informing engagement. Thus, having participants consciously attribute their affect to a different source eliminated the elevation in engagement intentions for calorie-dense food.

Details are in the caption following the image
Study 3 – The impact of caloric density on social media engagement intentions as a function of misattribution.
Table 7. Study 3 – The effect of caloric density and misattribution of affect on social media engagement when controlling for food perception measures
OLS regression
(1) (2)
Misattribution −0.006 0.048
(.945) (.534)
Caloric Density 0.350** 0.440**
(.000) (.001)
Interaction: MisAtt*CalDense −0.218* −0.204*
(.041) (.029)
Food perception controls
Healthiness 0.177
(.169)
Naturalness 0.128
(.186)
Amount 0.161*
(.013)
Prettiness 0.099
(.233)
Orderliness 0.025
(.715)
Tastiness 0.268**
(.000)
R 2 0.085 0.321
R2 change 0.236**
  • Reporting standardized beta coefficients with p values in parentheses.
  • p < .10;
  • * p < .05;
  • ** p < .01.

General Discussion

Our research offers some initial insight into how the nutritional composition of food media influences social media engagement. As consumers’ preoccupation with digital food media continues to grow (Tubular, 2020), understanding the factors that increase engagement with this content is crucial. Most notably, social media platforms tend to use rank-ordering algorithms to prioritize and display content that receives more engagement (Gillespie, 2016; Hogan, 2015; Zulli, 2018). Simply posting content online does not mean it will be viewed. Rather, it is engagement with content that can amplify reach. Thus, identifying that the visualization of caloric density positively relates to engagement can help inform strategies of content producers as well as health food advocates.

These findings can stimulate several directions for future research. For instance, what is it about visualizing caloric density that seems to elicit positive affect? It could be argued that in our model, after we controlled for the nutritional components that were not easily discernable by sight, the only nutritional component remaining with any notable influence on caloric density was saturated fats. Saturated fats are prevalent in butter, cheese, meats, and oils (American Heart Association, 2021), and are known to give foods their juicy, chewy, and creamy sensory experiences (Teicholz, 2014). To test this, we collected supplementary data from the field to see if saturated fats would predict a different form of digital engagement—recipe bookmarking on Yummly.com, the largest online recipe aggregator (MDA 4 for reporting). We found that the amount of saturated fat in a recipe was associated with more bookmarking (β = .03, p < .001) even with similar controls to Study 1. This could be one explanation, and one which has precedence in practice. Food photographers often spray meat with WD-40 or brush it in dish soap, adding a sheen that makes it look “moist, plump, and juicy… these are the visual cues that make your mouth water when you look at it” (Chapin, 2016). Indeed, some alternative meat producers (e.g., Beyond Meat) have found that the product is better received if it both tastes and looks like the real thing (Piper, 2020). Is it possible to make other healthy foods like vegetables more appealing by applying visual characteristics associated with fattier foods (i.e., coating them with a sheen)? Future research should consider identifying these visual characteristics of nutrients to better inform strategies for garnering engagement with more health-conscious food media content. But food advertisers should exercise caution when modifying visual elements of meals. Research on prototypicality has noted that modifying deeply held beliefs about what food should look like (e.g., coffee that is not brown) can have negative consequences on preference (Noseworthy, Murray, & Di Muro, 2018). Therefore, future research also needs to identify limitations when altering the appearance of foods.

While our findings suggest that exposure to calorie-dense videos might elevate a person’s affective state, there are likely consequences to this seemingly innocuous activity. Not only does food promotion influence what people eat (Spence et al., 2016; Taylor, Noseworthy, & Pancer, 2019), but it may also shape their social dynamics in terms of what they share with others, ultimately influencing and normalizing what others eat. This is important given that obesity results from overnutrition (Livingston & Zylke, 2012) and people tend to underestimate the negative consequences of caloric intake when they engage in shared consumption (Taylor & Noseworthy, 2021). Future research could explore whether this also occurs in a virtual setting where content is shared. If so, not only would the sharing of food media influence and normalize what others might eat, but it could lead people to underestimate consequences to their health.

While our results focused on the consistent effect of caloric density on engagement, several other effects emerged in our data that future research can explore further. Of note was that video length and reposting were both negatively related to all engagement metrics. The effect of these variables suggests that brevity and novelty matter, especially in a distraction-laded digital environment such as browsing social media newsfeeds. Therefore, our findings suggest parsimony when developing and promoting meals online. Future research should investigate the limits of content duration and social media engagement further.

Overall, the notion that depictions of calorie-dense foods can increase audience engagement is relevant for content developers and advertisers interested in maximizing engagement. Yet, for advertisers, more research is needed to understand if this engagement stemming from exposure to calorie-dense food media content translates into a successful campaign by, for example, improving brand attitudes or increasing sales.

Conflict of Interest

The authors have no conflict of interest to declare.