- Dettagli della rivista
- Formato
- Rivista
- eISSN
- 2543-683X
- Prima pubblicazione
- 30 Mar 2017
- Frequenza di pubblicazione
- 4 volte all'anno
- Lingue
- Inglese
Online videos are an important source of information and entertainment for a substantial minority of the world. According to Alexa.com (
The social media video format allows influencers to chat directly to camera in an informal personal setting (e.g., bedroom, living room) and to respond to viewers through video commenting functions. The performers may strive to create a sense of intimacy with viewers to engage their attention (Marôpo, Jorge, & Tomaz, 2020; Raun, 2018). Repeated exposure to performers in naturalistic situations like this can lead audience members to feel a subconscious or conscious social bond with the performer: a
On YouTube, unlike TV and radio, there is a realistic chance of the viewer interacting with the performer through commenting on their video, possibly receiving a Like or reply from the performer or their team. This is like a PSI in that the relationship is between an audience member and performer, with little chance of turning into a personal friendship (e.g., Yuan & Lou, 2020). Nevertheless, it is not a classical PSI because it could be a real interaction, depending on whether the performer chooses to engage. Here are three examples of YouTube comment interactions, paraphrased for privacy: “Long term follower, thanks for always making beautiful, wholesome, helpful, videos packed full of ideas and advice! xx” [reply from influencer: “Wow thank you soooo much
Another concept that is useful to introduce is the strength of a PSI or semi-PSI. Since PSIs and semi-PSIs can be separate from PSRs, it seems reasonable to characterize those that are closer to PSRs as stronger. Strength is a key concept because this article introduces a method for assessing it in semi-PSIs.
The
It is also useful to distinguish between potential and actual PSI/PSR. This is helpful when discussing the properties of videos or influencers that make them likely to generate PSIs or PSRs.
The
Most previous PSI/PSR research has investigated either audience characteristics associated with pure PSRs/PSIs or media product characteristics associated with PSI/PSR potential (e.g., looking at camera, informal style, regular episodes).
This article introduces a new method to automatically assess a social media video's sPSI potential through semi-PSIs in its audience comments, validates the method, and uses it to identify factors associating with sPSI potential in UK female influencers’ YouTube videos. Female YouTube lifestyle influencers seem likely to elicit PSIs because they have large audiences, frequently adopt a direct, personal style and may influence their (presumably young female) audiences (Ladhari, Massa, & Skandrani, 2020; Lee & Lee, 2022; Thelwall & Cash, 2021; Torjesen, 2021). The restriction to the UK is a practical decision to increase homogeneity and reduce the chances of interpretation misunderstandings. The presumably female audiences of female influencers can be expected to have stronger parasocial relationships than the presumably male audiences of male influencers (Tukachinsky, Walter, & Saucier, 2020). The new method to assess a video's sPSI potential is the proportion of comments containing the term ‘you’. Referring directly to the influencer would be a semi-PSI, as defined above, in contrast to general comments (e.g., “love that colour”, “my nails cost £20”) that may not target the influencer. If validated, the you pronoun proportion semi-PSI indicator would give the first automated method to calculate any type of PSI indicator and, therefore, the first automated method to estimate video sPSI potential. The overarching research questions are as follows.
RQ1: Does the proportion of comments containing a you pronoun correlate with the sPSI potential of a YouTube influencer video? RQ2: Which factors associate with higher sPSI potential for YouTube influencer videos?
As introduced above, this study adopts and adapts the existing theory of parasocial relationships to video-based social media influencers. The study makes the assumption that existing PSR/PSI theory, although developed for radio and extensively applied to television, also applies to video-based social media influencers. It extends this theory by introducing the concept of semi-PSI and hypothesising that semi-PSI works similarly to traditional PSI. To be clear, semi-PSI is hypothesised to be additional to PSI rather than the social media version of it. PSIs and semi-PSIs are assumed to have different strengths according to the extent to which they reflect underlying PSRs. This study also extends PSR/PSI theory by introducing the concept of PSI/PSR potential for influencers or products, which has been implicit in prior research. Finally, at a methodological level, it introduces you pronoun proportions with the hypothesis (explicitly tested as H1) that they may be indicators of sPSI potential.
YouTube influencers create an emotionally safe space (Mardon, Molesworth, & Grigore, 2018), usually talking informally to the camera as if to a friend, which promotes PSR formation (van Reijmersdal et al., 2020). PSR strength strongly correlates (Pearson: 0.6) with motivation to watch knitting vlogs (Gregg, 2020) and associates with time spent watching YouTube by children aged 9–12 (Tolbert & Drogos, 2019). Thus, PSRs seem to be effective for retaining YouTube viewers. They can also generate advertising and marketing revenue for the influencer through enhanced endorsement credibility (Dekavalla, 2020; Ladhari, Massa, & Skandrani, 2020; Sokolova & Kefi, 2020; Reinikainen et al., 2020; Trivedi & Sama, 2020) and make their public service messages more persuasive (Rosaen, Dibble, & Hartmann, 2019; Sakib, Zolfagharian, & Yazdanparast, 2020).
Semi-PSIs are also important on social media. Audiences may “Like” a video or comment underneath it, perhaps believing that the influencer will notice or respond (Reinikainen et al., 2020). The performer, in turn, relies on evidence of interaction to be visibly successful (Chen, 2016) and to promote their work on YouTube (Bishop, 2018). This incentivises them to interact and encourage audience responses (Tur-Viñes & Castelló-Martínez, 2019).
There is limited evidence that commenting (i.e., semi-PSIs) can be an indicator of PSRs or strong PSIs. One study found a statistically significant weak positive relationship between commenting on videos and PSR strength for a snowball sample of German-speaking people following any YouTube celebrity (Rihl & Wegener, 2019). For PSIs, viewers told to comment on and like/dislike a video in an experiment (i.e., semi-PSIs) felt stronger PSRs (Pearson: 0.6) with a US female travel vlogger promoting a product (Munnukka et al., 2019). The current study for the first time singles out comments that seem to indicate stronger semi-PSI: those that address the performer directly with you pronouns. The underlying belief is that these apparently stronger semi-PSIs are better indicators of PSI than comments alone. This generates the first hypothesis, which is leveraged for a method to address subsequent hypotheses and analyses.
H1 You pronouns: videos with higher proportions of comments containing variants of the pronoun ‘you’ have higher sPSI potential.
The terms PSR and PSI have been used interchangeably in some parasocial literature (Dibble, Hartmann, & Rosaen, 2016) and more is known about PSRs than PSIs, so this section mainly discusses PSRs. The two are definitionally distinct (Dibble, Hartmann, & Rosaen, 2016; Hartmann & Goldhoorn, 2011) but likely to be connected in practice because repeated PSIs could help build a PSR, a PSR seems likely to make PSIs more common, and the two concepts correlate moderately (Tukachinsky, Walter, & Saucier, 2020). Whilst the correlation and possible connections do not prove that factors affecting PSRs also affect PSIs, factors affecting PSRs at least form a plausible starting point for investigating factors affecting PSIs. Moreover, the current article focuses on strong PSI, which has a definitional connection with PSR. The rest of this section primarily introduces hypotheses about the sPSI potential of videos through prior PSI or PSR research (sometimes also from research conflating the two or implicitly focusing on sPSI). As explained in the methods, these hypotheses will be tested through semi-PSI evidence from you pronoun proportions. Known PSR factors that cannot feasibly be related to the data collected are not included (e.g., Performer physical and social attractiveness [needs viewers’ subjective judgements]; binge watching and episode regularity [no viewing data]). Four mimic informal communication between friends and one is specific to PSRs.
H2 Gaze: An influencer talking primarily to camera in close up (head and upper body) has higher sPSI potential.
H3 Realism: A video perceived to be a realistic portrayal of the influencer as a friend has higher sPSI potential.
H4 Self-disclosure: A video discussing personal issues that the influencer has previously not shared in public has higher sPSI potential.
H5 Time: An influencer's more recent videos have higher sPSI potential.
H6 Popularity: Influencers with more subscribers (i.e., more popular in one sense) produce videos with a higher proportion of sPSIs compared to PSIs.
Qual1: Which types of videos have higher sPSI potential than suggested by the above factors?
The research design was to obtain a large sample of popular UK female YouTube lifestyle influencers and obtain statistical evidence of their popularity and the popularity of their videos from YouTube, in addition to you pronoun proportion data for each for their videos. A paired design was chosen, comparing two videos from the same channel rather than comparing between channels because each channel has a different style and focus (Figure 1). The paired design also partly factors out the average audience PSR strength for individual influencers by contrasting two videos from the same influencer. Comparing videos between different types of influencer could have therefore introduced spurious relationships from the different topics covered. The research design was as follows, and is described and justified in detail below.
H1: test whether videos judged to have higher sPSI potential tend to have higher you pronoun proportions. H2, H3, H4: test whether videos with higher you pronoun proportions (or higher judged sPSI potential) are more likely to use gaze, friendliness and self-disclosure (if H1 is accepted). H5: test whether newer videos from an influencer tend to have higher you pronoun proportions (if H1 is accepted). H6: test whether more popular influencers’ videos tend to have higher you pronoun proportions (if H1 is accepted). Qual1: manually examine videos with high you pronoun proportions despite not having high scores on H2, H3, and H4 for possible additional video sPSI potential factors (if H1 is accepted).
Figure 1
The manual and automated stages of the data collection and classification components of the research design. Results from earlier stages are also used in some statistical analyses (e.g., for popularity).
As described and justified below, the two videos chosen for each influencer had the highest and lowest proportion of comments containing the word
There is no register of lifestyle influencers on YouTube, so a large set was identified by browsing YouTube and the wider web. Sources included the Lifestyle YouTube category, recommendations in video comments, online magazine articles about prominent influencers, subscription/following lists on YouTube and Instagram for people with lifestyle interests, and searches in YouTube (e.g., “make-up haul” “fashion haul”). YouTube was difficult to search for a comprehensive set of relevant influencers, so all authors searched and browsed separately using ad-hoc queries, browsing and intuitive explorations to make a reasonably comprehensive list, although the final list is almost certainly incomplete. Lifestyle influencers’ channels were retained if they matched the following criteria:
A single female influencer or (in three cases) two female influencers working as a team. For consistency, female influencers working with male partners were excluded, but those occasionally making videos with a male partner were allowed. Trans women would have been included, of course, had any been found. British and living in the UK. An influencer was assumed to be British if they had a British accent and declared a UK location or had a UK focus. A focus on fashion or lifestyle but not cooking since cookery vlogs have a narrower focus and a “how to” style. Fashion and lifestyle content seemed to have a large overlap, so no attempt was made to distinguish between the two. At least 2,000 subscribers at the time of data collection so that they could be regarded as broadly popular (for reference, the first author's six year-old channel with 30 videos had only 54 subscribers by June 2021).
The influencers were from different ethnicities, apparent social class backgrounds, sexualities and ages, although most seemed to be aged 19 to 35. Many channels had unique characteristics, including a focus on family, make-up, cruelty-free make-up, luxury fashion, high street fashion, student life, travel, and fitness. Some influencers had evolved through romantic partnerships, marriage, separation, and children. They may also occasionally address issues that do not normally fit their theme or create spin-off channels for different content.
The final set included 289 UK female influencer channels, 286 individual females and 3 female duos. Two channels were from the same person (Zoella and Zoe Sugg), which were retained since Zoella is a highly successful influencer and these channels contain differing content types. A second pair of channels (Rose and Rosie, Rose and Rosie Vlogs) were from the same team but with a different style and attracting a million subscribers for one and nearly half a million for the other, so these were also retained. On 6 December 2020 these channels collectively accounted for 111,194,290 subscribers (including duplicates, excluding one channel hiding its subscriber count; median 107,500 per influencer) and 11,811,612,486 individual video views (median 9,059,454 per influencer). Only registered YouTube users can comment or subscribe, so at least a median of 107,000 viewers per influencer were enabled to comment.
YouTube has an Applications Programming Interface (API) that allows lists of videos to be requested for any channel as well as the comments on each video. This service (version 3) was used in the software Mozdeh (
The comments were processed to identify semi-PSI evidence. Comments were classified as semi-PSI, likely addressing the influencer, when they contained a variant of the pronoun “you” (you, you’ve, you’re, your, yourself, yourselves). Since replies to other commenters were excluded, a comment written underneath the video of an influencer containing a you pronoun seems likely to address the influencer. To test this, 100 matching comments were selected with a random number generator, and all appeared to be addressing the influencer directly. In contrast, only 7% of 100 non-matching comments directly addressed the influencer (e.g., with “u”, “ur”, or their name) and a further 11% addressed the influencer through a request (e.g., “Do lips next”) or implicitly (e.g., “happy birthday”, “thanks”, “congratulations”, “keep the black dress”). Thus, you pronouns in comments seem to be more frequently attempts to interact with the influencer than other comments, on average.
For each video, the proportion of (non-reply) comments containing a variant of the you pronoun was calculated using a program written for this (available as a button in the YouTube tab of the free software Webometric Analyst at
Videos with fewer than 100 comments were excluded because the you proportion would be a less reliable semi-PSI indicator for these. The 100-comment threshold was chosen in advance as a simple round number, allowing proportions to be accurate to 1%. This was chosen as a trade-off between a larger threshold with fewer videos and a lower threshold with less accurate proportions. After excluding videos with fewer than 100 comments, influencers with fewer than 100 remaining videos were also excluded, leaving 118. The arbitrary threshold of 100 was set to increase the likelihood that there was substantial parasocial variation within the set for an influencer. On 6 December 2020 these 118 channels collectively accounted for 91,001,600 subscribers (including duplicates, excluding one channel hiding its subscriber count; median 437,000 per influencer) and 10,533,276,527 individual video views (median 45,035,114 video views per influencer).
A manual content analysis of videos was used to identify PSI factors (for H2, H3, H4) and overall sPSI potential (for H1). For this, a sample of videos was selected to represent likely high and low sPSI potential. For each influencer, the video with the highest you pronoun proportion (henceforth: high you video) and the video with the lowest you pronoun proportion (henceforth: low you video) were selected for manual investigations of factors from H2 to H4. These represent apparent sPSI extremes for each influencer and this sample should therefore be easiest to analyse for sPSI potential differences. A paired design (one high and one low you video per user) is more robust than selecting the overall highest and lowest you videos because the latter approach might primarily reflect the strategies of a single influencer. Moreover, a single influencer's strategy might relate to sPSI in an unusual way.
The paired design resulted in 236 videos, two from each of the 118 influencers with at least 100 videos having at least 100 comments each. During the manual classification, one of the videos became private and the channel was excluded, leaving a final set of 117 channels and 234 videos.
The paired videos were manually classified for the presence of factors and overall sPSI potential (based on the Wikipedia description of PSI which describes sPSI). The former was used to assess the hypothesised video-based sPSI factors (H2, H3, H4) and the latter to validate the methods assumption that the proportion of you pronoun comments is an indicator of sPSI potential for YouTube videos (H1).
Four researchers with experience of coding social media sites conducted the content analysis. A codebook was created to describe the classes to be used, based on the research hypotheses and intuitions obtained by browsing video channels before the data collection started. Because YouTube influencers often seem to follow a direct camera gaze style, H2 was split into five related parts to investigate it at a fine-grained level. Some facets were coded by watching the first 3 minutes and the last 2 minutes (5 minutes in total) and other facets by also hovering the cursor along the rest of the video:
Watching 5 minutes: Eye contact, Main focus, Friendly, Personal issues disclosed, High sPSI potential. Watching 5 minutes and cursor hovering: Faces camera, Head and shoulders, Alone.
For the pilot testing phase, 134 videos were selected from the 76 influencers that had at least 50 but fewer than 100 videos with at least 100 comments each to train the coders and fine-tune the codebook. Thus, there was no overlap in the influencers between the pilot set and main set. To help with training, the pilot sample included the classification of videos as high or low you (based on you proportions, but the coders were not told about the you pronoun heuristic used) and the main sample did not. The coders were instructed not to read the comments under a video but make classifications based on the video alone. One of the aspects coded was the sPSI potential of the video. Although PSI and PSR are normally investigated using validated instruments asking multiple questions, a direct question was more appropriate because the coders were not interacting with the videos as viewers and understood the PSI concept, relating it to PSR. Inter-coder agreement was assessed on the pilot sample, discussing cases of disagreement. The same people then coded the main sample, with two coders independently classifying each video, and each coder paired with the other coders for approximately an equal number of videos (i.e., a partial crossed design).
Krippendorff's interval alpha was used to assess the levels of agreement between coders on the six classes that reference percentages of the videos with the given feature, and ordinal alpha for the two classes using qualitative scales. Krippendorff recommends a score of at least 0.8 for firm conclusions and 0.667 for tentative conclusions (Krippendorff, 2019) when the results are used for content analysis purposes. The inter-coder agreement is adequate for tentative conclusions, except for the Friendly class (Table 1). For completeness, the Friendly class results are reported in the paper alongside the others, but these are suggestive rather than sufficient for tentative conclusions.
Inter-coder consistency scores for facets of UK female influencer videos Krippendorff's alpha at the difference between high you and low you videos within a channel level (n=117). Interval alpha is used for the middle six and ordinal alpha for the other two.
| Facet | 1 | 2,3,4 | 5 | Alpha |
|---|---|---|---|---|
| H1 High sPSI potential | None | ….*** | Very high | 0.716 |
| H2 Faces camera | Almost never | ….* | Virtually all the time | 0.730 |
| H2 Eye contact with camera lens | Almost never | ….* | Virtually all the time | 0.684 |
| H2 Head and shoulders mainly in shot | Almost never | ….* | Virtually all the time | 0.722 |
| H2 Alone (no other talking people) | Almost never | ….* | Virtually all the time | 0.936 |
| H2 Main focus is vlogger | Almost never | ….* | Virtually all the time | 0.842 |
| H3 Friendly, realistic chat style | Almost never | ….* | Virtually all the time | 0.552 |
| H4 Personal issues disclosed | Never | ….** | Very personal focus | 0.887 |
| Qual1 Reason for high or low PSI | Free text |
Almost never - 25% of the time - half of the time - 75% of the time - Virtually all the time.
Never - Some personal but not private issues discussed - A lot of personal but not private issues discussed -Some previously private issues discussed, but not the focus of the video - The video focuses on very personal and previously private issues.
Unlikely to engage the viewer with parasocial interaction in any of the video - Unlikely to engage the viewer with parasocial interaction in most of the video - Moderately likely to engage the viewer with parasocial interaction in most of the video - Likely to engage the viewer with parasocial interaction in most of the video - Very likely to engage the viewer with parasocial interaction in most of the video.
The effectiveness of the you pronoun proportion as an indicator of sPSI potential (H1) was assessed by calculating a 95% confidence interval for the average difference between the sPSI potential of each high you videos subtract the sPSI potential of the corresponding low you video. A confidence interval of positive values excluding zero would indicate that the you pronoun proportion is a reasonable method to assess sPSI potential for individual (successful UK female) influencer videos. The limitation of this test is that it only considers extreme cases (the highest and lowest you pronoun proportions). The focus on extreme values is necessary for statistical power (compared to random videos) to test the hypothesis.
Spearman correlations were calculated between the sPSI potential difference (high you video score minus low you video score) and PSI factor score differences (high you video score minus low you video score) for each channel. This test assesses whether the three hypothesised PSI factors tend to occur more in videos judged to have higher sPSI potential. This test may have a confirmation bias because the same coders judged all factors so they might be more likely to judge a video as having sPSI potential if they found factors known to associate with it.
To avoid confirmation bias, 95% confidence intervals were calculated for the average difference between the high and low you videos for each factor. If these confidence intervals are positive and exclude 0 then this gives statistical evidence that the factors associate with high you pronoun proportion videos. Combining this with a positive result for H1 would give evidence, untainted by confirmation bias, that the factors associate with higher sPSI potential.
The 118 influencers with at least 100 videos containing at least 100 comments were also investigated to assess whether the you proportion, as an indicator of sPSI potential, changed over time. For each influencer, their videos with at least 100 comments were arranged in order of publication date and a Spearman correlation calculated to assess whether there was a tendency for the you proportion to increase or decrease over time. The average of the Spearman correlations is an indicator of the average tendency across channels.
Spearman correlations were calculated to assess whether more popular influencers’ videos tended to elicit more strong semi-PSI, as reflected by the median proportion of you-related pronouns in comments on their videos. This test was carried out for all 118 channels in the main dataset, including the channel that was excluded from the content analysis for its missing video. The median you proportion across all videos was used to reflect the overall channel average. This test is dependent on H1 validating you pronoun proportions as an indicator of sPSI potential.
This study uses only public data and thus was exempt from ethical approval. To protect the privacy of commenters by not drawing attention to them, no commenter names or exact quotes are given (Wilkinson & Thelwall, 2011). Influencers are only named if they are prominent media figures for the same reason. The four influencers named in this article all have over a million subscribers, which seems sufficient to classify them as public figures.
Videos with the highest you proportion tended to have higher sPSI potential, as judged by the coders (Figure 2), with the difference being on average 1.18 (95% CI: 1.02 to 1.35; sd=.90). Moreover, in 98 out of the 117 channels, the coders judged the high you video to have higher sPSI potential than the low you video, with only 10 reverse cases. Thus, you proportions are reasonably effective at differentiating between high and low sPSI potential videos. As a corollary, you proportions can be used as an indicator of sPSI potential (because they correlate) and it seems reasonable to use them also as indicators of stronger semi-PSI, as defined above (because the correlate with the related sPSI potential concept, and are calculated from comment interactions).
Figure 2
The difference between the sPSI potential of the highest you video and the lowest you video by channel (e.g., a channel would score 3 if the highest you video had sPSI potential 5 and the lowest you video had sPSI potential 2), in the 117 channels analysed.
The ten cases where the video with a lower you proportion had a higher sPSI potential score were examined for possible causes, as an extra validation step. In one case, a low PSI video included a request by the influencer to comment with a specific phrase to enter a competition. This phrase contained “you”, which inflated the you proportion in a non-sPSI way. In three cases the high you proportion video with low sPSI potential discussed the channel itself, with commenters making suggestions. This seems to be an example of genuine interactions. Three other videos had small segments discussing personal issues upsetting the influencer, although the focus of the video was something else, suggesting that even brief mentions of personal issues can generate sPSIs. The reason for the discrepancy in the other two videos was unclear.
Based on the average coder scores, within each channel, videos with higher sPSI potential were more likely to have stronger degrees of all individual factors classified: influencer facing the camera, making eye contact with the camera, shooting mainly their head and shoulders, presenting alone, focusing on themselves, talking in a friendly and realistic style, and mentioning personal issues (Table 2). The Spearman correlations supporting this conclusion are for differences between the score in each dimension for the high and low you videos for each channel.
Spearman correlations between channel differences: high you proportion video subtract low you proportion video (n=117).
| Spearman's rho* | Eye contact | Head and shoulders | Alone | Influencer focus | Friendly, realistic | Personal issues | sPSI potential |
|---|---|---|---|---|---|---|---|
| Faces camera | .537** | .482** | .328** | .365** | .235* | 0.166 | .224* |
| Eye contact | 1 | .457** | .513** | .487** | .342** | .286** | .405** |
| Head & shoulders | 1 | .460** | .274** | .231* | 0.054 | .262** | |
| Alone | 1 | .395** | .191* | 0.159 | .302** | ||
| Influencer focus | 1 | .449** | .760** | .657** | |||
| Friendly, realistic | 1 | .450** | .506** | ||||
| Personal issues | 1 | .712** |
Correlation significant at p=0.05;
Correlation significant at p=0.01
The reason for the weaker associations between some factors and sPSI potential may be that the characteristic was either nearly universal or rare. The influencer was almost always facing the camera in almost all videos, for example, and was also usually alone (Table 3). In this context, the weakest association may be for the head and shoulders shot style, since this is much less universal (3.4 instead of 4.4) but has only a slightly stronger correlation with sPSI potential (0.262 instead of 0.224).
Average coder scores for the attributes classified (n=234).
| Factor | Min. | Max. | Mean | Std. dev. |
|---|---|---|---|---|
| Faces camera | 1 | 5 | 4.4 | 0.80 |
| Eye contact with camera lens | 1 | 5 | 3.2 | 0.87 |
| Head and shoulders mainly in shot | 1 | 5 | 3.4 | 1.51 |
| Alone | 1 | 5 | 3.7 | 1.54 |
| Main focus | 1 | 5 | 3.4 | 1.11 |
| Friendly, realistic chat style | 1 | 5 | 3.4 | 0.78 |
| Personal issues disclosed | 1 | 5 | 2.8 | 1.29 |
| High parasocial interaction potential | 1 | 5 | 3.1 | 0.89 |
Recalling that the correlation results above may be influenced by confirmation bias since sPSI potential was judged by the same people that judged the other factors, comparing between high and low you videos avoids this possibility, however, and has been validated by H1 being accepted. The 95% confidence intervals for the difference between high and low you videos for each factor are all positive and exclude 0, confirming the correlation results above (Figure 3).
Figure 3
95% confidence intervals for the difference between the coder score of the highest you video and the lowest you video within a channel (n=117).
As mentioned above, low values for the confidence intervals can occur either because a factor has little effect or because there is little variation in the factor. An example of a factor with little variation is Faces camera (Figure 4), for which the high and low you videos in 45 of the 117 channels had the same score. Taking this into account, in all cases it is rare for channels to have higher scores on the low you video than on the high you video for any factor (Figure 4).
Figure 4
The difference between the coder score of the highest you video and the lowest you video (e.g., a channel difference would be -4 if the highest you video scored 1 and the lowest you video scored 5) by channel (n=117).
A qualitative examination of the high you videos that did not have the typical sPSI potential factors suggested that content focusing on the channel itself or requesting viewer feedback may generate strong semi-PSIs, or at least a high proportion of you pronoun comments. The exceptions included videos with brief introductions to personal issues, suggesting that even this small amount can be effective at generating strong semi-PSIs.
The proportion of comments containing “you” was correlated against publication date for channels with at least 100 videos with at least 100 comments each (n=118). The median Spearman correlation was −0.033, indicating a slight tendency for semi-PSI, as indicated by you proportions, to decrease over time. This tendency occurred for a small majority of influencers (56%), although this is not statistically significantly different from 50% (Binomial test). Thus, there is insufficient evidence to conclude that the sPSI potential of videos from an influencer tends to increase or decrease over time (publication date).
The two thresholds were relatively arbitrary, so alternative values were tried as a robustness check. Varying the two thresholds (100, 50, or 10 comments, 100, 50, or 10 videos) had little effect on the Spearman correlation (always between −0.02 and −0.05) but the median correlation was statistically significantly different from 0 in two of the nine cases examined (Table 4). Thus, it is possible that there would be statistical evidence for a tendency for sPSI potential to decrease over time within a UK female influencer channel for some conceptualizations of the problem, but the correlations seem too small to be of practical interest.
Median Spearman correlation between video age and you pronoun proportion within a channel for different thresholds.
| Spearman's rho | ≥ 10 videos | ≥ 50 videos | ≥ 100 videos |
|---|---|---|---|
| ≥ 100 comments per video | −0.033 (n=118) | −0.041 (n=148) | −0.021 (n=194) |
| ≥ 50 comments per video | −0.037* (n=151) | −0.028 (n=185) | −0.030 (n=236) |
| ≥ 10 comments per video | −0.049** (n=211) | −0.029 (n=247) | −0.021 (n=275) |
Rho significant at the 0.05 level (2-tailed).
Rho significant at the 0.01 level (2-tailed).
There was a statistically significant weak tendency for more popular influencers to attract
Spearman correlations between the median proportion of you-related pronouns in comments on videos and the total number of subscribers or video views, for different minimum numbers of qualifying videos. One channel with hidden channel subscription counts was excluded.
| Qualification | Total channel views | Channel subscribers | No. of channels |
|---|---|---|---|
| ≥ 100 videos with ≥ 100 comments | −0.260** | −0.229* | 117 |
| ≥ 50 videos with ≥ 100 comments | −0.216** | −0.193* | 147 |
| ≥ 10 videos with ≥ 100 comments | −0.252** | −0.250** | 192 |
Significant at the 0.05 level (2-tailed).
Significant at the 0.01 level (2-tailed).
A limitation of this study is its focus on individual PSIs of various types rather than longer-term PSRs, which may be more important. Although previous survey-based studies have found a moderate positive correlation between the two (Tukachinsky, Walter, & Saucier, 2020), it should not be assumed that the same is true for the types of PSI examined here. It also does not consider relationship building over time (Tukachinsky & Stever, 2019) or other channels used by the influencer, such as Instagram, which may help to build or to deepen PSRs (Giles, 2002; Tukachinsky & Stever, 2019). In addition, a few comments are not in English and may contain pronouns in other languages. Some uses of the you pronoun do not refer to the influencer, and there are other ways of addressing them (e.g., by name), so the data is incomplete.
The results support H1, providing evidence that you pronoun proportions in comments is a valid sPSI indicator for UK female YouTube influencer videos. The statistical tests add to the analytic argument above that referring to an influencer as ‘you’ indicates a stronger type of interaction than usual semi-PSI YouTube comments. This is the first study to investigate this indicator. It seems likely that the you pronoun proportion indicator would be valid for female lifestyle influencers in English-speaking nations and similar indicators may work in other languages. The extent to which this indicator works for other types of influencer or video content is unclear, however.
The you pronoun indicator seems likely to reflect the four stages of PSR formation to different extents, with stronger associations with later stages. This is based on the hypothesis that the stronger the PSR, the more likely the viewer would be to comment directly to the performer. Some “you” comments were negative, criticising the influencer for boring content. These could be evidence of PSR breakdown (i.e., late stage PSR: Tukachinsky & Stever, 2019) from previously satisfied viewers or evidence of blocked PSR from unimpressed first-time viewers. These comments seemed to be a small minority, so they were not singled out for removal.
The results confirm that three factors known to elicit PSIs or PSRs in other contexts also apply to female YouTube influencers: talking directly to the camera (assessed in different ways) (Dibble, Hartmann, & Rosaen, 2016; Hartmann & Goldhoorn, 2011; Oliver et al., 2019), revealing private and personal information (Giles, 2002; Ko & Wu, 2017) and (although not validated) using an informal friendly style (Giles, 2002; Ru; Tukachinsky, Walter, & Saucier, 2020). Of these, revealing private and personal information seems to be the strongest marker of sPSI potential, as found both by the ‘you’ proportion confidence intervals and the classification correlations. This is partly because influencers seem to vary this factor more than the others examined: not all videos can reveal new personal issues. Personal issues were sometimes discussed briefly at the start of a video, before switching to the usual style. Two of the coders thought that the emotionality was false in some videos, but it is not clear whether typical viewers would agree.
Influencers displayed emotions when discussing personal issues. These emotions sometimes manifested in slight changes in tone of voice and expression (as is normal: Van den Stock, Righart, & de Gelder, 2007) but also sometimes in (appropriate) tearfulness. It has previously been remarked that crying influencers can create intimacy with their audiences and perhaps also authenticity (Berryman & Kavka, 2018). It is therefore not clear whether the personal information or the emotions elicited PSIs, although a combination of the two may be the most powerful and most authentic.
The results suggest no sPSI potential change over time (publication date) for the videos of an influencer, or a small decrease over time. PSRs should strengthen over time for individual audience members (Tukachinsky & Stever, 2019), so it might be expected that sPSIs would also be more common for later videos. Nevertheless, even if videos are mainly watched when they are recent and are watched chronologically, time series comment statistics partly reflect loyal repeat visitors and partly reflect new visitors. Thus, it is possible that sPSIs are more common for successful channels for repeat visitors, but that this is counterbalanced by the increased number of new visitors hearing about the channel. The results are therefore not directly comparable to prior research.
The negative relationship between influencer popularity and the you pronoun sPSI potential indicator suggests that more popular influencers do not produce videos with higher sPSI potential. Whilst this has not been tested before in prior research, it is unexpected because more popular influencers elicit stronger PSRs (Rasmussen, 2018; Rihl & Wegener, 2019). A possible explanation is that you pronoun commenting is less likely for popular influencers because their viewers would face more competition for attention from the larger audience. Thus, more popular influencers may generate stronger PSRs and more sPSIs but weaker semi-PSIs. Alternatively, the fame of popular influencers may attract a larger share of new viewers, with these interacting less than long term viewers with PSRs.
This study has made the following theoretical, methodological, and empirical contributions to the literature. They are briefly contextualised in terms of prior knowledge here, and their practical implications are suggested.
Introduced three new concepts that were implicit, overlooked or underdeveloped in prior research:
○ ○ ○
Figure 1
Figure 2
Figure 3
Figure 4
Spearman correlations between channel differences: high you proportion video subtract low you proportion video (n=117).
| Spearman's rho |
Eye contact | Head and shoulders | Alone | Influencer focus | Friendly, realistic | Personal issues | sPSI potential |
|---|---|---|---|---|---|---|---|
| Faces camera | .537 |
.482 |
.328 |
.365 |
.235 |
0.166 | .224 |
| Eye contact | 1 | .457 |
.513 |
.487 |
.342 |
.286 |
.405 |
| Head & shoulders | 1 | .460 |
.274 |
.231 |
0.054 | .262 |
|
| Alone | 1 | .395 |
.191 |
0.159 | .302 |
||
| Influencer focus | 1 | .449 |
.760 |
.657 |
|||
| Friendly, realistic | 1 | .450 |
.506 |
||||
| Personal issues | 1 | .712 |
Median Spearman correlation between video age and you pronoun proportion within a channel for different thresholds.
| Spearman's rho | ≥ 10 videos | ≥ 50 videos | ≥ 100 videos |
|---|---|---|---|
| ≥ 100 comments per video | −0.033 (n=118) | −0.041 (n=148) | −0.021 (n=194) |
| ≥ 50 comments per video | −0.037 |
−0.028 (n=185) | −0.030 (n=236) |
| ≥ 10 comments per video | −0.049 |
−0.029 (n=247) | −0.021 (n=275) |
Inter-coder consistency scores for facets of UK female influencer videos Krippendorff's alpha at the difference between high you and low you videos within a channel level (n=117). Interval alpha is used for the middle six and ordinal alpha for the other two.
| Facet | 1 | 2,3,4 | 5 | Alpha |
|---|---|---|---|---|
| H1 High sPSI potential | None | …. |
Very high | 0.716 |
| H2 Faces camera | Almost never | …. |
Virtually all the time | 0.730 |
| H2 Eye contact with camera lens | Almost never | …. |
Virtually all the time | 0.684 |
| H2 Head and shoulders mainly in shot | Almost never | …. |
Virtually all the time | 0.722 |
| H2 Alone (no other talking people) | Almost never | …. |
Virtually all the time | 0.936 |
| H2 Main focus is vlogger | Almost never | …. |
Virtually all the time | 0.842 |
| H3 Friendly, realistic chat style | Almost never | …. |
Virtually all the time | 0.552 |
| H4 Personal issues disclosed | Never | …. |
Very personal focus | 0.887 |
| Qual1 Reason for high or low PSI | Free text |
Average coder scores for the attributes classified (n=234).
| Factor | Min. | Max. | Mean | Std. dev. |
|---|---|---|---|---|
| Faces camera | 1 | 5 | 4.4 | 0.80 |
| Eye contact with camera lens | 1 | 5 | 3.2 | 0.87 |
| Head and shoulders mainly in shot | 1 | 5 | 3.4 | 1.51 |
| Alone | 1 | 5 | 3.7 | 1.54 |
| Main focus | 1 | 5 | 3.4 | 1.11 |
| Friendly, realistic chat style | 1 | 5 | 3.4 | 0.78 |
| Personal issues disclosed | 1 | 5 | 2.8 | 1.29 |
| High parasocial interaction potential | 1 | 5 | 3.1 | 0.89 |
Spearman correlations between the median proportion of you-related pronouns in comments on videos and the total number of subscribers or video views, for different minimum numbers of qualifying videos. One channel with hidden channel subscription counts was excluded.
| Qualification | Total channel views | Channel subscribers | No. of channels |
|---|---|---|---|
| ≥ 100 videos with ≥ 100 comments | −0.260 |
−0.229 |
117 |
| ≥ 50 videos with ≥ 100 comments | −0.216 |
−0.193 |
147 |
| ≥ 10 videos with ≥ 100 comments | −0.252 |
−0.250 |
192 |
Editorial board publication strategy and acceptance rates in Turkish national journals Research misconduct in hospitals is spreading: A bibliometric analysis of retracted papers from Chinese university-affiliated hospitals The need to develop tailored tools for improving the quality of thematic bibliometric analyses: Evidence from papers published in Sustainability and Scientometrics The notion of dominant terminology in bibliometric research