Mir, I. A. (2026). Influencer’s physical attractiveness and content aesthetics: Conscious and preconscious determinants of fashion-branded content engagement on Instagram. Journal of Creative Communications, 21(2), 201–219. https://doi.org/10.1177/09732586241288672
“The authors invoke an unproven perception–behavior mechanism to explain causation that was never observed. A direct path in a cross-sectional SEM establishes neither causation nor preconscious processing.”
Mir (2026) examines whether fashion influencers’ physical attractiveness and the aesthetics of their branded content predict followers’ engagement on Instagram. The study uses survey responses from 300 followers of 15 fashion influencers in Pakistan and analyzes the proposed relationships with structural equation modeling and mediation analyses. The main empirical finding is straightforward: followers who rate influencers and their content more positively also report more favorable attitudes and greater engagement. The methodological problem is that the article draws causal and psychological-process conclusions that the design cannot support.
The most serious problem is that all variables were measured in a single cross-sectional self-report survey. Participants simultaneously reported how attractive they considered the influencer, how aesthetically pleasing they considered the content, their attitude toward that content, and how often they viewed, liked, commented on, and shared it. Nothing was manipulated, and there was no temporal ordering of the variables. Nevertheless, the article repeatedly describes attractiveness and aesthetics as factors that “cause,” “trigger,” “stimulate,” or “activate” engagement. Those causal statements do not follow from the design.
For example, the proposed model assumes
attractiveness → attitude → engagement.
But the same covariance pattern is compatible with numerous alternatives. Followers who engage frequently with an influencer may develop more positive attitudes and subsequently rate that influencer as more attractive. A general liking or identification with the influencer could simultaneously increase attractiveness ratings, content-aesthetic ratings, attitudes, and engagement. The structural equation model cannot distinguish among these explanations.
The sampling procedure makes this problem particularly important. Participants had to have followed one of the selected influencers for more than six months. Thus, the sample is already conditioned on sustained interest in the influencer. People who disliked the influencer, found the content unattractive, or disengaged from it are systematically less likely to appear in the sample. The resulting correlations describe differences among an already selected group of followers; they provide weak evidence for the article’s practical recommendation that firms should hire physically attractive influencers because attractiveness causes engagement.
The article’s central methodological error is even more fundamental. It claims to distinguish a “conscious” route from a “preconscious” route. The mediated path
attractiveness/aesthetics → attitude → engagement
is interpreted as conscious influence, whereas a remaining direct path from attractiveness or aesthetics to engagement is interpreted as evidence of a preconscious perception–behavior process.
A direct regression coefficient is not a measure of unconscious processing.
If attractiveness predicts engagement after statistical adjustment for an explicit attitude measure, this merely shows residual covariance between those variables. That residual association could reflect measurement error in attitude, omitted mediators, stable preferences, common response tendencies, reverse causation, or numerous other processes. Nothing in the study measures awareness, intention, automaticity, processing speed, or participants’ ability to report the causes of their behavior. Consequently, the data provide no evidence that engagement was “preconscious” or unintentional.
For the same reason, the indirect path through an explicit attitude measure does not establish a conscious causal mechanism. Participants consciously completed the attitude questionnaire, but that does not mean the psychological process producing their engagement operated consciously. Statistical mediation and conscious psychological mediation are different concepts.
This problem is especially consequential because the “preconscious” interpretation is one of the article’s principal theoretical contributions. The authors explicitly invoke the perception–behavior literature, including Bargh et al. (1996), to justify the claim that a significant direct path demonstrates automatic behavior. But the current research contains none of the experimental procedures that would be required to test an automatic perception–behavior effect. The SEM therefore cannot adjudicate between conscious and unconscious processes.
The measures themselves also create substantial interpretive problems. On page 209, physical attractiveness is measured with “stylish,” “good looking,” “sexy,” and “elegant.” Content aesthetics is measured with “striking,” “wonderful,” “fascinating,” and “lovely,” while attitude is measured with “pleasant,” “good,” “likeable,” and “my favourite.” These constructs are conceptually and evaluatively intertwined. “Wonderful” and “lovely,” for example, are not narrowly aesthetic judgments, while “stylish” and “elegant” are not purely measures of physical attractiveness. Much of the model may therefore reflect a broad positive-evaluation factor rather than distinct psychological constructs connected by causal pathways.
The observed correlations are consistent with this concern. Physical attractiveness correlates .64 with content aesthetics and .64 with attitude; content aesthetics correlates .66 with attitude and .68 with reported content consumption. Demonstrating discriminant validity with the Fornell–Larcker criterion does not eliminate the possibility that halo effects or general liking strongly influence all of these ratings.
The attempt to dismiss common-method bias is also inadequate. All predictors, mediator variables, and outcomes were obtained from the same respondent at the same time using similar rating formats. The authors test whether several sets of items can be represented by a single latent factor and conclude that poor single-factor fit shows that common-method bias is not important. That conclusion does not follow. Common-method variance does not require every item to load on one factor. Several distinguishable constructs can coexist while correlations among them are inflated by shared method, evaluative consistency, acquiescence, or halo effects.
The proposed “snowball effect” suffers from the same causal problem. The authors find that self-reported consumption behaviors—viewing, reading comments, and liking—predict contribution behaviors such as commenting and sharing, and conclude that consumption gradually causes users to progress toward more active participation. Yet consumption and contribution were measured simultaneously. Someone who frequently comments and shares influencer content almost necessarily also views and consumes that content. A positive cross-sectional association therefore does not demonstrate a temporal progression from passive to active engagement. Testing a snowball process would require longitudinal evidence showing that earlier consumption predicts subsequent increases in contribution.
There may also be an unmodeled dependence problem. The 300 respondents followed one of only 15 macro-influencers. Followers of the same influencer are not necessarily independent observations. Influencers may differ systematically in appearance, production quality, follower demographics, posting frequency, and baseline engagement. Those influencer-level characteristics could generate correlations among respondent ratings. The reported SEM appears to treat all 300 followers as independent rather than accounting for clustering by influencer.
Another reporting issue concerns the engagement scale. The response categories are described as 1 = “very often,” 2 = “often,” 3 = “sometimes,” and 4 = “never.” Thus, larger numerical values indicate less engagement. Yet positive path coefficients are consistently interpreted as greater attractiveness and aesthetics producing greater engagement. The article does not clearly state in the reported method that these scores were reverse-coded. If they were reversed before analysis, that transformation should have been explicitly documented. If they were not, the substantive interpretation of the coefficients would be reversed.
Finally, the statistical success of the model should not be confused with strong evidence for the hypotheses. All five proposed hypotheses are supported, including the weakest direct attractiveness effect, β = .12, t = 2.19. The study was not preregistered, and the sample-size justification consists largely of the statement that N = 300 is sufficient for purposive sampling and structural equation modeling rather than an a priori power analysis tied to the focal effects. Good model-fit indices demonstrate that a specified covariance model can reproduce the observed covariance matrix; they do not establish that the arrows in Figure 2 represent the true causal processes.
The study therefore supports a much narrower conclusion than the article claims. Among existing long-term followers of fashion influencers, positive ratings of influencer attractiveness and content aesthetics are associated with positive attitudes and greater self-reported engagement. That descriptive association is plausible and potentially useful.
The study does not establish that physical attractiveness or content aesthetics cause engagement, that attitude mediates these effects causally, that any residual direct relationship reflects a preconscious perception–behavior mechanism, or that passive engagement develops over time into active contribution.
A suitable experimental design would manipulate influencer attractiveness and content aesthetics independently, randomly assign participants to conditions, measure actual engagement behavior, and include measures capable of testing awareness or automaticity. A longitudinal design would be required to test the proposed consumption-to-contribution “snowball” process. Without such evidence, the article’s strongest psychological claims are interpretations imposed on cross-sectional correlations rather than findings produced by the research design.