Content features: Messages to increase mental health literacy and communicative content strategies

Visual features: Four-tiered model of visual frames

Emerging scholarship has explored visual communication on social media regarding various health-related topics (Nobles et al., Citation2020) including mental health or mental illness communication (Feuston & Piper, Citation2018, Citation2019; Griffith et al., Citation2021; N. Lee et al., Citation2020). Employing human or computational analysis, previous studies examined various features of images, including the image composition (e.g., photographic or illustrative, infographic or novel visual), objects and people presented (e.g., human face or body, social groups, scenery, food, drink, and animals), pixel features, and topic or meaning of the visuals (Kearney et al., Citation2019; Y. Kim & Kim, Citation2020). Despite the diversity of visual features analyzed in various studies, a systematic, theoretical investigation of visual images of health communication is lacking, making it difficult to compare the results and generate replicable and design-specific recommendations (King & Lazard, Citation2020).

To develop a more systematic investigation of visual messages, this study adopted the visual framing theory, a four-tiered model of visual frames as the analytical framework to examine visuals (Rodriguez & Dimitrova, Citation2011). We argue that with its four-tier system, this model can cover most of the image features analyzed in previous studies and categorize different image features into different levels for analysis. As a result, a systematic investigation of visuals can be generated.

The four-tiered model of visual frames suggests four levels to analyze visual images: denotative, stylistic-semiotic, connotative and ideological (Rodriguez & Dimitrova, Citation2011). The first three levels were applied to our analysis in view of their relevance. First, the denotative level is concerned literally with who or what is present in the visual messages. Health communication literature has long recognized the value of tailoring or targeting a variety of demographic constructs in health-related messages to increase their effectiveness (Pope et al., Citation2018). In relation to mental health, two demographic factors are worth noting: ethnicity and gender. Ethnic minorities suffer from mental health for various reasons. For example, African American adults who live in poverty are more likely to report serious psychological distress due to socioeconomic challenges, demonstrating the need to segment a population to achieve effective health outcomes (“Mental and Behavioral Health―African Americans,” Citation2021). The stigma of mental illness is prevalent in Asian cultures, making this population reluctant to seek help (Wynaden et al., Citation2005). Even worse, across all racial categories in the U.S., the indigenous population has the highest rate of suicide (Ivey-Stephenson et al., Citation2017). In terms of gender, men and women are vulnerable to different types of mental illness: women have higher prevalence rates of anxiety and depression, while men are more likely to be diagnosed with substance abuse or antisocial disorders (Eaton et al., Citation2012). Therefore, mental health promotion and intervention messages should be tailored to those populations’ needs and preferences (Pirkis et al., Citation2019), and the first step is to present populations at risk in the message (Lama et al., Citation2018). Accordingly, to examine the visual component of mental health Instagram posts at the denotative level, this study asks:

RQ 3:

How are demographic segments (gender and ethnicity) presented on mental health organizations’ Instagram posts?

Second, the stylistic-semiotic level examines pictorial variables, which may include but are not limited to camera position, actions or poses, tonal shades, and color (Rodriguez & Dimitrova, Citation2011). Visual persuasion theory (Messaris, Citation1997) explores how visual messages influence attitudes and behaviors of individuals. Generally, visual imagery can be categorized as photographic, presenting reality through people, places and things; or illustrated, displaying shared meaning through symbols such as logos, signs, and cartoons (King, Citation2015, Citation2016). Illustrated images may also involve an infographic, a visual representation of data highly effective in health education (Martin et al., Citation2019). These two types of composition may be used based on different purposes. Photos include real people, objectives, events or sceneries or a combination of them, and can be used to present a vivid picture of reality. Hence, demonstrate the severity or urgency of the health problem (Gibson & Zillmann, Citation2000; Yoo, Citation2016). Illustrations use symbols to present an abstract representation of the reality and are used to convey comprehensive or abstract health related information (King, Citation2015), thus, provide a macro-level understanding of the health issues. Considering the differences between these two types of visual imagery and the different effects of each category, the following question is proposed:

RQ 4:

How are visual compositions presented on mental health organizations’ Instagram posts?

Finally, the connotative level asks more generally and associatively about the concepts attached to visuals. In other words, this level of analysis concerns the “social meaning” (Rodriguez & Dimitrova, Citation2011, p. 56) as presented through visual symbols. Following this logic, this study operationalizes its social meaning as major topics that are presented in pictures. Using the framing approach (Entman, Citation1993), which is concerned about how something is presented to the audience, previous studies have analyzed textual mental health messages on social media and suggested that awareness, individual feelings, and anti-stigma were the three most prevalent framing topics (Ju et al., Citation2021; Pavlova & Berkers, Citation2022). However, how the framing topics are used in visual communication of mental health is yet to be explored, hence, on the connotative analysis level, we ask:

RQ 5:

What are the visual framing topics presented on mental health organizations’ Instagram posts?

Audience engagement with mental health information

Social media allows users to engage with the information through such functions as likes/reactions, reposts/shares and comments. Audience engagement can facilitate the consumption and dissemination of the information, therefore, widen the distribution of the message for organizations, because social media algorithms actively promote messages with high involvement (e.g., high numbers of likes; Luarn et al., Citation2015). In addition, audience engagement can drive positive behavioral outcomes such as advocacy or word-of-mouth in real life (Alhabash & McAlister, Citation2015; Men & Tsai, Citation2014), and further expand the effect of health promotion on social media. It is important for organizations to understand how to properly engage the audience with their social media messages. Focusing on content and visual elements of messages, this study explores:

RQ 6:

To what extent is audience engagement associated with mental health literacy, communicative content strategies, and three levels of visual features?

Codebook development and coding scheme

A codebook consisting of three parts, content features, visual features and audience engagement, was developed for analysis (see Appendix). The content component analyzed the whole Instagram post, including caption and image and covered two variables. The first variable examined the communicative content strategy, which included 1) information, 2) community, and 3) action (Lovejoy & Saxton, Citation2012). Posts were first coded into one of the three categories and then into one sub-category (only two of the three categories – community and action – had subcategories. For details, see Table 1). The second variable addressed messages designed to increase mental health literacy. Posts were coded into one or a combination of the following categories: 1) N/A; 2) maintaining good mental health, such as promoting a healthy lifestyle; 3) mental disorders and treatments, such as introducing facts and symptoms of anxiety; 4) anti-stigma; and 5) enhancing help-seeking efficacy, which informs the public how to seek detailed information or find professional help (Kutcher et al., Citation2016).

The second component analyzed the visual part of the message only, and focused on three levels of visual analysis of the posts. First, the denotative level examined who is covered in the image, in terms of gender and ethnicity. Each image was coded as having a presence of the following gender and ethnicity or not: female and male, Caucasian and non-Caucasian, using a dichotomous coding (0=non-presence, 1=presence). For a visual containing multiple people, all the applied coding categories were chosen. Second, the stylistic-semiotic level asked what form of visuals was present. Each image was coded into one of the five visual compositions: 1) photo, 2) comic/cartoon, 3) infographic, 4) words on pictures (with no design; e.g., words on plain background) and 5) words on pictures (with design, such as posters). This category also covered the number of visuals presented in each post. The last category concerned the connotative meaning of visuals, and each post was coded into one of the visual framing topics, developed inductively from the data and deductively from previous studies (Pavlova & Berkers, Citation2022). Finally, public engagement only recorded the number of likes and comments on each post because repost data were not available from Instagram.

Intercoder reliability

Following the suggestions of Lacy et al. (Citation2015), the first two authors of the study practiced with “similar non-study content” (p. 804), which were the Instagram posts from these two organizations published prior to August 2020, until intercoder reliability was established. After two rounds of tests, with each round containing 140 (equals to 19.3% of the study sample) of messages that were not included in the final study sample, all items achieved an acceptable level of intercoder reliability using Krippendorff’s Alpha, from 0.74 to 0.87, with an average of 0.80 (Krippendorff, Citation2018; see Table 1 for reliability statistics for each coding category). Due to the limited variance, the intercoder reliability of denotative level of coding (sex and ethnicity) and number of pictures were tested using percentages of agreement; both items reached 100% agreement. After the intercoder reliability was established, the two coders split the study sample data and coded independently.

Frequencies of content and visual strategies

Data were aggregated and submitted to the statistical analyses using R version 4.1.1 for descriptive analysis (https://www.r-project.org). The frequencies and percentages of each variable are reported in Table 1.

RQ 1 asked what mental health literacy was promoted on Instagram. The results showed that more than a half of the messages (n = 384, 53%) included at least one type of MHL information. Among the posts that cover only one type, 7.6% (n = 55) posts focused on obtaining and maintaining good mental health, 1.5% (n = 11) introduced disorders and treatments, 3.7% (n = 27) sought to decrease stigma against mental illness, 18.6% (n = 135) were to enhance help-seeking efficacy by offering another link that specifies where, from whom, and how to seek relevant information for details. The rest of the posts were a combination of different categories.

Regarding content strategies (RQ 2), less than half (n = 334, 46%) of the post focused on sharing information of mental health facts and knowledge, 28% (n = 204) attempted to build a community with the stakeholders, and 26% (n = 187) aimed to call for an action. For frequencies of specific categories of each strategy, see Table 1.

Regarding the demographic segments (RQ3) in the pictures, most of the visuals did not depict a specific gender, and among those which did, the visual representations of men (n = 140, 19.3%) and women (n = 167, 23%) were more evenly distributed among all posts. In terms of ethnicity, most of the images did not feature a specific ethnicity and among those which did, racial minorities (n = 141, 19.4%) were covered more than Caucasians (n = 94, 13%).

RQ 4 asked how visuals were presented. Most posts included one picture (n = 629, 86.8%) and the rest had more than one. In terms of image composition, around 90% (n = 652) were illustrated pictures and 10% (n = 73) were photographic. Among illustrated images, more than a half (n = 394, 60.4%) of the posts were words on pictures with a background design (such as posters), 20.9% (n = 136) were words on plain backgrounds, 15.2% (n = 99) were comics or cartoons, and 3.5% were infographics (charts or graphs; n = 23).

Concerning how mental health organizations covered visual framing topics (RQ 5), the most frequently depicted topics were awareness and advocacy (n = 187, 25.8%), event/product/personnel information (n = 168, 23.2%), and mental health care (n = 111, 15.3%).

Predictors of audience engagement

To answer RQ 6, the statistical significance was set to an alpha level of .05 for all the analyses. Regarding each of the categorical independent variables, two separate ANOVAs were employed to examine the differences in the number of likes and comments, respectively. Also, for each of these independent variables, categories with fewer than 10 Instagram posts were excluded from the ANOVA analyses to avoid extreme inequality in group sizes. Significant results were followed by post-hoc pairwise comparisons using Tukey adjustment, where applicable. The pictures related to stylist-semiotic visual strategies were treated as a continuous variable, and so, two separate linear regressions were employed to examine the prediction of likes and comments, respectively. Table 2 presents the main effects of all the independent variables, the two ANOVAs and simple linear regressions.

Because all analyses are parametric tests and assume a normal distribution of the dependent variables (likes and comments), a Box-Cox power family of transformation was applied to likes and comments that are both positively skewed. Specifically, the maximum-likelihood estimation was used to optimize the tuning power parameter, λ, such that the distribution of the transformed data has the largest similarity to normality. When zero or negative values occur in the data, a second parameter, γ, was also estimated to reduce transformation bias and added to make all values positive before the data can be handled by the Box-Cox transformation (Box & Cox, Citation1964; Fox & Weisberg, Citation2019; Hawkins & Weisberg, Citation2017).

Content features

This study examined communicative strategies and organizational efforts to increase mental health literacy as the two content features. Among three communicative strategies, information and community messages can garner more engagement than action messages, and social support was one of the most effective among all subcategories. In line with prior research on text-based platforms (Lovejoy & Saxton, Citation2012; Saxton & Waters, Citation2014), this study confirmed the effectiveness of these two engaging strategies on a visual-dominant platform. In addition, the study also contributes to MHL literature by revealing its coverage on social media and its influence on audience engagement. According to Jorm (Citation2012), compared to major physical diseases, the public is still ignorant about how to prevent mental disorders, and often delay or avoid seeking professional assistance. One possible reason is the limited mental health education and resources; and in addition, the stigma associated with mental illness also results in a negative attitude toward seeking treatment (Crowe et al., Citation2018; Ross et al., Citation2020). Therefore, social media, particularly official accounts of leading mental health organizations, should become a significant source for promoting MHL to the public. The findings showed that among half of the posts that were designed to increase at least one type of MHL, the most frequently covered was information-seeking efficacy – either alone or combined with other types of MHL categories, indicating organizations’ efforts on educating the public about how and where to find help. In addition, audiences were more likely to respond to the posts that bring illnesses to their attention, explain symptoms and possible treatments, and add external links that guide them to better understand a certain condition. Another MHL category effective in involving audiences was mental health care, which demonstrated useful tips on how to achieve and maintain positive mental health. This finding, on one hand, signals the audience’s needs and interests for concrete information to deal with mental health issues; on the other hand, it reinforces the importance of information utility in the context of healthcare (H. S. Kim, Citation2015). Mental health posts that involve extra resources in a link about certain illnesses, or suggestions on maintaining mental wellness, are all useful and practical information that can satisfy the “utility” function.

Visual features

The denotative-level visual analysis showed that most messages did not feature a specific gender or ethnicity, demonstrating a generic approach to mental health communication and promotion from organizations on social media. Although this generic approach runs counter to the recommendations from message tailoring literature (Pope et al., Citation2018) that call for targeting specific demographics in health messages, it worked in engaging social media audiences. There are two possible explanations. First, it matches the public discourse of mental health on Instagram where the majority posts only identify the general public without specifying any social groups (N. Lee et al., Citation2020). Second, this approach matches the goal of organizations, which is to raise awareness of mental health among the public, that is, people from diverse backgrounds. Also, demographic information of Instagram users showed that it has almost equal number of female (49.3%) and male (50.7%) users (McLachlan, Citation2022) and this platform has been used by different ethnicity groups with 35% of causation, 49% African Americans, and 52% Hispanic population (Auxier & Anderson, Citation2021). With this distribution of users across genders and ethnicities, the generic approach should be more effective.

At the stylistic-semiotic level, numbers of images and visual compositions were examined. In accordance with a previous study (Y. Kim & Kim, Citation2020), the results found that words on pictures with design were the most frequent visual form on Instagram, signaling the organization’s preference of communicating an abstract representation of mental health/illness. In regard to the number of pictures, although most posts only contained one picture, the number of visuals was positively related to engagement. This is somewhat surprising because other studies reported a negative correlation between the number of words in social media posts and audience engagement (Ju et al., Citation2021). More visuals, similar to more word counts, means more information, which should decrease audience involvement. However, the positive correlation in the current study indicates a difference in information processing between visual and textual messages, which triggers audience reactions. Visual and textual information is processed differently in that people use a verbal code to process language but an imagery code for visual messages (Paivio, Citation1986; Sadoski & Paivio, Citation2004). The process also produces different outcomes in which people demonstrate a greater recall of pictures and images as they enhance the sensory experience (Blanco et al., Citation2010; Hong et al., Citation2004). It is possible that due to the enhanced experience of seeing a picture, Instagram users reacted positively to more pictures through engaging with the content. The results of visual compositions, illustrative images―particularly comics/cartoons and words on plain backgrounds―were found more effective in eliciting engagement than photos. On one hand, these results align with previous findings that comics and cartoons are effective visual tools (Nan et al., Citation2022). On the other hand, it contradicts the notion that realistic images featuring real people or actions are more compelling for communicating health messages, as audiences can identify with the content (Ophir et al., Citation2019). Such results added complexity to the literature, and future studies should further explore the mechanism underlying this complexity.

Finally, the connotative level showed that anti-stigma was the most effective visual topic to elicit engagement, which confirmed the notion that stigma-reducing discourse is a powerful tool for mental health promotion on social media (Pavlova & Berkers, Citation2022). Social support topics also engage the audience because this type of visuals can meet the needs for community and support of the audience, as previous studies indicated that social media users seek mental health information to find community or support (Pavlova & Berkers, Citation2022). Lastly, classification framed visuals providing specific information about symptoms or treatment was the third most engaging visual topic. This result confirms that information utility is a predictor of user’s engagement of health information (Cappella et al., Citation2015). Taken together, the results support previous findings that various types of content meet public demands in different ways, and the differences are reflected in the likes, reposts and comments (J. Lee & Xu, Citation2018; Keib et al., Citation2018). Our study demonstrated that in the mental health communication context, anti-stigma, support and community, and information utility were the top three visual topics demanded from the audience.

Theoretical contributions and practical implications

This study yields several theoretical, methodological, and practical implications. First, it enriches the theoretical framework of communication strategies from effective public health messaging (Nan et al., Citation2022), particularly message execution, by offering evidence of effective visual strategies. Specifically, our findings showed certain stylistic-semiotic and connotative elements were effective in engaging the audiences. The next step should explore the relations between audience engagement and attitudinal or behavioral change to further test the effectiveness of these visual execution strategies.

Second, this study contributes to the methodological development of visual message analysis. As mentioned, due to the lack of a unified analysis framework, it is difficult to compare the research results to generate replicable and design specification recommendations for health promotion (King & Lazard, Citation2020). Our study used the four-tiered model of visual framing theory (Rodriguez & Dimitrova, Citation2011) and demonstrated that this analytical framework can be applied to social media visual message analysis and generate meaningful results. By employing a similar framework, future research can be compared and hopefully, through these results, replicable recommendations can be produced.

Regarding practical implications, this study provides valuable information to organizations about how to use visuals as a powerful tool to assist mental health promotion. First, we recommend using more illustrative visuals as they are more engaging and do not require professional photographers nor consent from the people being featured, saving time and resources for nonprofit organizations that are already on tight budgets. Second, anti-stigma themed visuals should also be used as they can actively engage the audience and help spread the word. In terms of content design, organizations should be cognizant of the public’s engagement with MHL and promote relevant content, including disorder diagnoses, self-care, reducing misunderstanding and stigma, along with external resources. As our health systems “have not yet adequately responded to the burden of mental disorders” (WHO, 2021, p. 4), such information can help with early detection and wellness management, thus reducing the gap between the need for treatment and its provision.