Features of the videos

The results of the analyzed videos, although with a very high impact and volume of interactions, show high inequalities. Thus, the average number of followers is 979,164.40, although the standard deviation is 1,141,681.52, with the minimum being 3,150 and the maximum being 3,950,000, while the asymmetry was 1,489 and the kurtosis 1,150. This trend is reflected in all the impact variables, as can be seen in Table 2, showing that, although a large number of the contents provide relatively low values –as is also shown by the fact that the median is much lower than the average–, some contents with very high values significantly affect the central values.

Table 2. Central values and dispersion of the measures of consumption and interaction.

Source: Own elaboration.

Regarding the descriptive variables, a remarkable uniformity is observed in a large number of aspects. Thus, all the videos are aimed at the general population and none of them are aimed at professionals. In 12 of the cases analyzed (24%), there is an absence on camera while the voice is still being heard, and only in one video there is presence on camera, but without addressing it, while in 43 sample units (86%) the gaze and the voice goes to the camera. Regarding the relationship with the audience, all the videos except one use the second person, 46 videos (92%) invite them to watch other videos on the channel, 44 invite them to visit the account or other platforms, in 43 videos expressions are used to promote subscriptions, in 27 (54%) comments or opinions are directly requested, and in two videos it is even used in the specific name of a follower and specific comments from followers are answered.

Where a greater diversity is observed is in the use of emotional expressions that are present in 22 videos (44%): in 17 videos (34%) joy is expressed; in 1, sadness; in 1, anger; in 2, shame; in 3, frustration; in 9, surprise; in 3, fear: in 1, disgust; and envy was not observed in any of them. Similarly, in 22 videos the user is emotionally appealed: in 17 videos, through joy; in 1, with anger; in 2, with shame; in 3, with frustration; in 12, with amazement; in 3, with fear; and in no case with sadness, disgust, or envy.

In general, the videos are explanatory, so that in 41 of them the process is explained from the beginning to the end, and in the same number of videos information is offered on products or processes necessary to carry out the diet; Moreover, 22 videos (44%) refer to information without asking the user to take any action. 27 videos explain how the diet works, and 30 videos also describe how to apply it in a professional imitation style. 43 videos offer instructions on care after using the product or diet, 28 mention things that can be done after applying the product or diet, and 18 also refer to things that should not be done. Regarding the plurality and quality of the information, 44 videos (88%) mention aspects about which the available information is uncertain, 38 (76%) describe positive aspects of the product or diet in question, 11 (22%), negative aspects are described, and in 8 both positive and negative aspects are included. Likewise, 41 videos describe what can be expected from the diet, in 13 cases side effects are mentioned, in 5 contraindications are reported, and in 7 there is a warning of potential risks.

The contents are, in general, personal: only one video resorts to the interview of a person external to the channel, and in only one case, content is not used in the first person. 45 videos (90%), also, show the information based on the personal experience of the YouTuber. On the other hand, the promotional aspect of the videos is not dominant, although it is relevant: in 21 videos (42%) specific brands are mentioned, in 4 a specific center or professional is mentioned, and in 19 (38%) promotional messages are included to sell a product.

Regarding the variables of quality and credibility studied within the quantitative analysis, it was observed that only 7 of the 50 videos (14.0%) included information from experts in the field, and only 12% made reference to qualifications or relevant training. 41 videos (82.0%) mentioned previous experience with success stories, and 35 (70.0%) mentions of publications on the subject were included. In 7 of the 50 videos (14.0%), balanced information was included, while in 44 (88.0%) biased information was included; It should be noted that, although it is not usual, the same video could include information from both types. On the other hand, 16.0% of the videos included reliable sources of additional information, and 88.0% included sources of additional information that were unreliable or impossible to classify; again, a video could include both reliable and unreliable additional sources of information simultaneously.

Finally, the average duration of the videos is 10.45 minutes, with a standard deviation of 04.16 minutes. The shortest video is 02.45 minutes, and the longest is 24.32 minutes.

Relations between variables

After exposing the characteristics of the videos in the sample, we now proceed to study the effects of the different variables on consumption and interaction, so that we can answer the research questions.

Starting with the relationship that may exist between the descriptive variables of the videos and the quantitative variables on interaction and impact and duration, it was observed that no variable was significantly correlated with the number of followers -something understandable since it is an element that does not directly affect each video, but rather the entire account of the YouTuber – but neither does it affect the number of views or comments. For its part, the volume of likes of a video is significantly and positively correlated with the emotional appeal to frustration [R(50)=.603, p<.001] and with the description of contraindications [R(50)=.533, p<.001] And similarly, the number of dislikes is significantly and positively correlated with the emotional appeal to frustration [ R(48)=.515, p<.001], with the description of contraindications [R(48)=.524, p<.001] and possible health risks [R(48)=.544, p<.001].

On the other hand, the duration was significantly correlated with the expression of sadness [R(50)=.507, p<.001] and frustration [R(50)=.488, p<.001], as well as with the emotional appeal through shame [R(50)=.507, p<.001] and fear [R(50)=.620, p<.001], and negatively, with the presence of instructions to personal care after using the product or diet [R(50)=-.533, p<.001].

Among the quantitative variables, it was observed that the number of followers and the duration are not related to the rest. Although the views, the number of likes, dislikes, and comments always correlate with each other in a positive and significant way. Thus, the number of views is correlated with the number of likes [R(48)=.876, p<.001], dislikes [R(48)=.800, p<.001], and comments [R(48)=.696, p<.001], also the number of likes is correlated with the number of dislikes [R(48)=.910, p<.001] and with the number of comments [R(50)=.609, p<.001], and finally, the number of dislikes and comments is also positively correlated [R(48)=.633, p<.001].

Among the descriptive variables, statistically significant correlations are not abundant, given that they are frequently variables with little or no diversity –variables in which the absence or presence is total or almost total are frequent–, although some readings are of interest. As it seems logical, there is a negative correlation between the absence on camera keeping the voice and the gaze and the voice directed at the camera [R(50)=-.448, p<.001]. Talking and looking directly at the camera correlates positively with the invitation to watch other videos on the channel [R(50)=.518, p<.001]. The use of the second person is positively correlated with the invitation to watch other videos on the channel

[R(50)=.484, p<.001] Likewise, the invitation to watch other videos on the channel is positively correlated with the invitation to visit the account on other platforms [R(50)=.572, p<.001].

The expression of emotions¹ has a strong positive correlation with the emotional appeal to the viewer [R(50)=.675, p<.001] and with the emotional appeal through joy [R(50)=.725, p< .001]. The expression of joy is positively correlated with the emotional appeal to the audience [R(50)=.640, p<.001] and with the use of joy [R(50)=.733, p<. 001] to address the audience. The expression of sadness shows an absolute positive correlation with the expression of anger [R(50)=1.000, p<.001], and a high correlation with the expression of frustration [R(50)=. 565, p<.001] Also, the expression of anger is positively correlated with the expression of frustration [R(50)=.565, p<.001], while the expression of shame is positively correlated with the expression of frustration [R(50)=.808, p<.001] and with the emotional appeal through anger [R(50)=.700, p<.001], from shame [R(50)=1.000, p <.001], frustration [R(50)=.808, p<.001], and fear [R(50)=.808, p<.001], as well as with the inclusion of an interview with a person external to the channel [R(50)=.700, p<.001], information on contraindications [R(50)=.612, p<.001], and the warning of possible risks [R(50)=.506, p<.001], while negatively correlating with the mention of aspects for which the available information is uncertain [R(50)=-.553, p<.001] and with the care instructions after use of the product [R(50)=-.506, p<.001]. The expression of frustration shows a positive correlation with the emotional appeal through anger [R(50)=.565, p<.001], shame [R(50)=.808, p<.001], frustration [R(50)=.645, p<.001], and fear [R(50)=.645, p<.001], besides the description of contraindications [R(50)=.477, p< .001], while it is negatively correlated with the inclusion of instructions for care after use of the product [R(50)=-.626, p<.001]. The expression of surprise is positively correlated with the expression of fear [R(50)=.539, p<.001] and with the emotional appeal to the audience through surprise [R(49)=.710, p<.001]. The expression of fear is positively correlated with the expression of disgust [R(50)=.565, p<.001] and with the emotional appeal through anger [R(50)=.565, p<.001].

Emotional appeal through anger is positively correlated with emotional appeal through shame [R(50)=.700, p<.001], frustration [R(50)=.565, p<.001], and fear [R(50)=.565, p<.001].The emotional appeal through shame is positively correlated with the appeal through frustration [R(50)=.808, p<. 001] and fear [R(50)=.808, p<.001], as well as the inclusion of interviews with people outside the channel [R(50)=.700, p<.001] and the description of contraindications [R(50)=.612, p<.001] and possible risks [R(50)=.506, p<.001], and it correlates negatively with the mention of aspects on which the available information is uncertain [R(50)=-.553, p<.001] and with the presence of personal care instructions after using the product or diet [R(50)=-.506, p<.001]. The emotional appeal through frustration shows a positive correlation with the appeal through fear [R(50)=.645, p<.001], with the inclusion of interviews with people outside the channel [R(50)=.565, p<.001], and with the description of negative aspects of the product [R(50)=.476, p<.001], contraindications [R (50)=.758, p<.001], and possible risks [R(50)=.626, p<.001], and negative with the mention of aspects for which the available information is uncertain [R(50)=-.684, p<.001]. Finally, the emotional appeal through fear is positively correlated with the inclusion of interviews with people outside the channel [R(50)=.565, p<.001] and the description of negative aspects of the product [R(50)=.476, p<.001], contraindications [R(50)=.477, p<.001], and possible health risks [R(50)=.626, p<.001], and negatively with the mention of aspects for which the information available is uncertain [R(50)=-.684, p<.001], with the description of positive aspects of the product mentioning aspects for which the information available is uncertain [R(50)=-.450, p<.001], and with the mention of care instructions after use of the product [R(50)=-.626, p<.001].

1 Existing correlations between this variable and the different emotions expressed are not included, since this variable includes them all. The same happens with the emotional appeal to the audience and the different emotions used for this appeal.

The mention of a brand is positively correlated with the inclusion of promotional messages to sell a product [R(50)=.503, p<.001]. The mention of aspects on which the available information is uncertain is negatively correlated with the description of contraindications [R(50)=-.492, p<.001] and possible risks [R(50)=-.560, p<.001]. The description of positive aspects of the product has a positive correlation with the presence of instructions on personal care after using the product [R(50)=.583, p<.001]. On the other hand, as seems evident, the description of negative aspects of the product is positively correlated with the information on side effects [R(50)=.566, p<.001], contraindications [R(50)=.628, p<.001], and possible health risks [R(50)=.760, p< .001]. The explanation of how the diet works is positively correlated with the description of how to apply the diet with a more professional communication style [R(50)=.721, p<.001].

Finally, the reference to possible side effects is positively correlated with information on contraindications [R(50)=.562, p<.001] and possible risks [R(50)=.681, p<.001], which in turn are also correlated with each other [R(50)=.826, p<.001]. Mentioning what can be done after using the product is positively correlated with mentioning what cannot be done after using the product [R(50)=.497, p<.001].

With this, RQ1 can be answered by stating that there are no clear patterns regarding which variables give rise to greater consumption or interaction in videos about miracle diets by YouTubers; Except for the emotional appeal to frustration and the description of contraindications, which are significantly positively correlated with the number of likes and dislikes, no relevant influences were observed. However, consumption does influence interaction: that is, greater consumption (more views) leads to higher levels of interaction (more likes, dislikes, and comments).

Finally, it should be mentioned that no significant differences were observed between the videos in both languages. Some trends can be seen, such as the fact that the videos in Spanish have a greater volume of interactions than those in English or that the videos in Spanish also tend to show risks and contraindications, but the differences are small. Thus, in response to RQ2, no significant differences have been observed derived from the use of English or Spanish in the videos, so it can be stated that the videos of both languages follow similar patterns.

Finally, and in response to RQ3, hardly any correlations between the different variables have been detected, so we can interpret that they follow independent patterns. Except for a few specific cases

–correlations between variables on the presence of emotions or the emotional appeal to the viewer, or correlations between variables on informative elements–, the relationships that have been observed do not seem to follow a clear theoretical pattern. This seems to confirm the work that pointed to the need to understand videos as a whole rather than based on very specific characteristics (O'Keefe & Hoeken, 2021).

Discussion of observations

In general, this research has shown that the videos on health-related content produced and disseminated by influencers through YouTube are very personal, informative, and, although not generally, with relatively plural information, sometimes addressing risks or negative aspects, although positive visions and recommendations predominate. It should be noted that the products and diets addressed often have a high risk to health and do not represent practices recommended by experts, so, although their contraindications or associated risks are sometimes indicated and there is a large amount of information, the promotional component and the search for channel traffic seem to be more relevant factors than health in these videos. In this sense, according to Harris et al. (2020), one of the motivations of YouTubers to produce health content, and, specifically, diets such as those addressed in this research, focus on the possibilities of obtaining personal benefits. Gedeman (2019) specifies that the market for beauty influencers is massive; in fact, today it continues to represent an area with great investment possibilities.

Given the questionable quality and reliability of the videos, their effects must be accurately analyzed. In fact,Coates et al. (2019) observed that celebrity endorsement of healthy foods encourages younger people to eat more consciously, therefore, it is reasonable to predict that if the advice is harmful, it may also be assumed by the audience. In this sense, the fact of hardly detecting any influence of the different variables studied on consumption or interaction with the videos could point towards passive consumption, almost indifferent to the characteristics of the video, which could be considered a risky attitude if it led to non-critical consumption. This uncritical consumption is reinforced given one of the most striking observations of the study, the positive correlation between dislikes and other variables such as views, comments, and, above all, likes. It is appreciated that the dislikes option, rather than reducing views or the number of likes, seems to be another form of interaction. This may be because the factor that increases the number of interactions – likes, dislikes, and comments – is the volume of views and the popularity of the video, rather than its content or characteristics.

The work has not addressed whether the comments were positive or negative, although in general terms it is a form of interaction that is also positively related to the rest of the quantitative variables. Still, it would be convenient for future works to observe whether the meaning of the comments may be related to other variables. Nor has it been analyzed whether these interactions have effects on the attitudes and behaviors of the audience, although it can be assumed that the greater the impact and the greater the audience, the greater the probability of effects on viewers, especially if it is an uncritical consumption.

The foregoing is relevant because influencers are perceived as knowledgeable about the products and services of their domain, and it is for this reason that the consumption of these videos by viewers must be aware of and guided by regulatory processes that allow recommendations to be differentiated from advertising spaces. Similarly, from training in communication and dissemination of content, there must be a process of digital literacy that allows exploring and searching for content in a more conscious, critical, and efficient way. In fact, to achieve a transformative impact on the positive, it is essential to acquire the necessary skills for the development of that awareness of self-care (Chang, 2019; Hibbard, Peters, Dixon, & Tusler, 2007; Koh & Rudd, 2015) with the ultimate goal of improving the decision-making process on health (Mason and Scammon, 2011; Powers et al., 2010; Southwell et al., 2019) (Mason & Scammon, 2011; Southwell et al., 2019).

It should be pointed out that all the observations made on the correlations between the different descriptive variables of the videos should be taken with caution because, although most offer logical and theoretically sustainable data, they are subject to the effect of the limited sample size, with 50 videos analyzed. The use of inferential statistical tests, precisely, makes it possible to discern the significant effects from those in which the effect of chance cannot be ruled out; Added to this is the decision to use a smaller margin of error than usual (α=.01), increasing the confidence margin. However, the limitation that the study poses in this regard and the need to verify, with larger samples, the observations made in this exploratory research, are recognized.

Additionally, the results of this research must be interpreted by taking into account two aspects. In the first place, variables related to the characteristics of the videos on YouTube have been analyzed, rather than aspects related to the authors or the content itself, having left out of the study factors external to YouTube, such as where each video has been promoted. The analysis tried to explore a series of characteristics less studied in previous studies to find out their potential effects, something that has been fulfilled, although the possible influence of other variables not studied, and which may have had effects on the observed consumption and interaction, cannot be ruled out. Second, to approach the

study reliably and guarantee observations with a high degree of certainty, it was decided to always report values with a high degree of significance (p<.001); this guarantees that the results are reliable, especially when dealing with a limited sample like this, but it prevents the visualization of correlations or differences that may be significant with lower significance values (P<.05 or p<.01).

Likewise, the limitation of the sample size must be taken into account, which prevents the elaboration of more forceful statistical observations, but whose justification is the need to have a manageable number of videos to be able to carry out the views, since there are no computational methods – as is the case with content in text format – that can reliably speed up this task. Given that it is an exploratory work, it is understood that the size of the sample is sufficient since it allows opening new lines of work in this type of content and reflecting on the impact of these videos on the health of the audience.

On the other hand, the work that has been proposed here adopts an eminently quantitative perspective, which allows numerous variables to be consistently addressed in the videos analyzed, but it is not the only way to expand knowledge of the subject. On the contrary, it is recommended that future lines of work continue to delve into more qualitative analysis –for example, analysis of the discourse of videos–. Another essential line of work will be reception studies, since the data analyzed here have made it possible to evaluate exploratory questions about interaction and consumption, but they do not manage to understand the effects and reception process of the videos in question, a key element for its potential effects on the health of consumers.

Finally, the values of the number of followers, views, likes, dislikes, and comments are expected to have increased from the moment the fieldwork was carried out. Although it was found that the date was not a determining variable in this aspect and videos collected over more than a year were included to avoid this effect, future studies with a greater temporal distance or with a longitudinal perspective will complement this aspect of the present work.