How Live Streaming Commerce Platforms Drive Consumer Purchase Intention: Dual Mediation of Trust and Perceived Value in a UK Sample

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for submitting your manuscript. This study examines how live streaming commerce platforms influence consumers’ purchase intention, with a particular focus on the mediating roles of two variables: consumer trust and perceived value. However, there are several issues that should be further clarified and improved, as follows:

1. The Introduction should more clearly highlight the platform-specific research gap to demonstrate that this study is not merely focused on consumer behavior, but also addresses the functioning and mechanisms of platforms.
2. Please specify the data collection period in the Methodology section.
3. The authors should add the HTMT ratio to further confirm discriminant validity, as the Fornell-Larcker criterion and MSV alone may not be sufficient according to current SEM reporting practices.
4. Some factor loadings are below .70, such as items measuring streamer credibility and perceived value. The authors should provide further justification for retaining these items in the model.
5. In Table 2, please clarify what the abbreviation “SL” stands for.
6. To improve transparency and reproducibility, the authors may consider adding a table or appendix presenting the full measurement items for each construct, together with the relevant source references.
7. In Table 5, there is a remaining comment: “Please add a table caption after its first citation.” Please remove this comment before resubmission.
8. The Discussion section should be strengthened by linking the findings more explicitly to platform studies, such as platform design, platform affordances, trust-building mechanisms, and platform-based value creation.
9. Please carefully check the table captions, particularly Table 5, to ensure that they are complete, accurate, and consistent with their citation in the text.
10. The authors should consider additional studies related to e-commerce live streaming to strengthen the literature review and discussion, such as:

• How can hesitation in hotel live-streaming payment be overcome?: Examine the role of entrepreneurial performance and viewers’ personality traits
• Exploring the influence of live streaming on consumer purchase intention: A structural equation modeling approach in the Chinese E-commerce sector
• Understanding Consumer Online Impulse Buying in Live Streaming E-Commerce: A Stimulus-Organism-Response Framework

Author Response

In addition to the point-by-point responses below, I have renumbered the references into their order of first appearance in the text to follow the journal's style, and I have made minor wording and table-presentation improvements throughout. None of these changes affect the data, analyses, results, or conclusions.

Comment 1: "The Introduction should more clearly highlight the platform-specific research gap to demonstrate that this study is not merely focused on consumer behavior, but also addresses the functioning and mechanisms of platforms"

Response: Thank you for this comment. I agree the platform side needed to be clearer. I have rewritten the research-gap paragraph in the Introduction so that the three characteristics now read as features the platform designs and controls, rather than as free-standing consumer cues, and the gap is stated as a platform question, namely which platform features move consumers towards purchase and through which internal mechanisms, not only as a question about consumer behaviour. I have also adjusted the first stated contribution so that it names these characteristics as part of platform design. The changes are in Section 1 (Introduction), in the second paragraph and the contributions paragraph.

 

Comment 2: "Please specify the data collection period in the Methodology section"

Response: Thank you. I have added the data collection period to the Methodology. The survey was administered in January 2026. This is now stated in Section 3.2.

 

Comment 3: "The authors should add the HTMT ratio to further confirm discriminant validity, as the Fornell-Larcker criterion and MSV alone may not be sufficient according to current SEM reporting practices"

Response: Thank you for this helpful suggestion. I agree. I have calculated the heterotrait-monotrait (HTMT) ratio for all construct pairs and added the values to a revised Table 3, alongside the Fornell-Larcker correlations. All HTMT values range from .301 to .646, which is below the conservative .85 threshold (Henseler et al., 2015), so discriminant validity is now supported by a third criterion. The new analysis is reported in Section 4.2 and Table 3, and the Henseler et al. (2015) reference has been added to the reference list.

 

Comment 4: "Some factor loadings are below .70, such as items measuring streamer credibility and perceived value. The authors should provide further justification for retaining these items in the model"

Response: Thank you for raising this. I agree it needed a clearer justification. Four loadings fell just below .70 (SC2 = .674, SC4 = .699, PDQ2 = .696, PV3 = .680). I have explained in Section 4.2 that these items were kept following Hair et al. (2019), who advise retaining indicators with loadings between .40 and .70 unless removing them would raise composite reliability or AVE above the required levels. Because composite reliability (.837 to .885) and AVE (.565 to .661) already passed their thresholds for every construct, dropping the items was not warranted. This explanation is now in Section 4.2.

 

Comment 5: "In Table 2, please clarify what the abbreviation 'SL' stands for"

Response: Thank you for catching this. I have defined the abbreviation in the note to Table 2. "SL" stands for standardised factor loading. This is now stated in the Table 2 note.

 

Comment 6: "…the authors may consider adding a table or appendix presenting the full measurement items for each construct, together with the relevant source references"

Response: Thank you for this suggestion, which improves transparency. I have added Appendix A, which lists every construct, its full set of items, and the source each item was adapted from. The items themselves were already shown in Table 2, but Appendix A (right after the "Conflicts of Interest" statement and before the "References" heading) now presents them together with their sources in one place for reproducibility.

 

Comment 7: "In Table 5, there is a remaining comment: 'Please add a table caption after its first citation.' Please remove this comment before resubmission"

Response: Thank you for noting this. This comment does not appear in my manuscript file, so I believe it was an editorial annotation added during processing rather than part of the text, and there was no comment for me to delete on my side. I have, however, addressed the underlying point by adding a caption to Table 5, which now reads "Table 5: Mediation analysis (standardised indirect effects)" (Section 4.4).

 

Comment 8: "The Discussion section should be strengthened by linking the findings more explicitly to platform studies, such as platform design, platform affordances, trust-building mechanisms, and platform-based value creation"

Response: Thank you, I agree the platform link could be more explicit. I have added a fourth theoretical implication to Section 5.2 that situates the findings in platform research. It frames the three antecedents as platform affordances (credibility resting on how the platform verifies and ranks sellers, interactivity on its chat and real-time reply features, and demonstration quality on its camera and display tools), connects the trust mechanism to Mattathil et al.'s [10] account of platform trust as an engineered behavioural architecture, and makes the point that value here is produced within the stream rather than fixed in the product, so value creation is itself platform-mediated. The new material also states what the findings add to that account, namely that trust is the specific conversion point at which a platform's credibility signals become value. This sits alongside the platform-design and trust-signal guidance in Section 5.3, which I have also strengthened. The new paragraph is in Section 5.2.

 

Comment 9: "Please carefully check the table captions, particularly Table 5, to ensure that they are complete, accurate, and consistent with their citation in the text"

Response: Thank you. I have checked all the table captions. Table 5 was missing its caption, which I have now added, and I confirmed that Tables 1 to 5 are each cited in order in the text and that every caption matches its table. This is reflected across Sections 3 and 4.

 

Comment 10: "The authors should consider additional studies related to e-commerce live streaming to strengthen the literature review and discussion, such as: How can hesitation in hotel live-streaming payment be overcome? Examine the role of entrepreneurial performance and viewers' personality traits; Exploring the influence of live streaming on consumer purchase intention: A structural equation modeling approach in the Chinese E-commerce sector; Understanding Consumer Online Impulse Buying in Live Streaming E-Commerce: A Stimulus-Organism-Response Framework"

Response: Thank you for these recommendations. I read all three papers in full and assessed each one against the focus of the study, following the journal's guidance to include a suggested reference only where it genuinely adds to the paper.

I have added two of them. Li, Wang and Cao [81], an S-O-R study of impulse buying in live streaming, now supports the statement in Section 5.3 that TikTok Shop trades in lower-involvement goods bought on impulse. It fits well, because it uses the same S-O-R framework as this paper and shows that impulse buying in live streaming works through the viewer's emotional state, which matches the affective route discussed in the study. Yang et al. [36], a recent and large Chinese e-commerce sample, is now cited in Section 2.1. It confirms that trust mediates the effects of the streamer, the product and the platform on purchase intention, but it tests its two mediators in parallel and does not include perceived value or a serial trust-to-value path. That serial path, tested in a UK rather than a Chinese sample, is what this study adds, so the paper helps me make the gap clearer.

I decided not to add the third paper, on payment hesitation in hotel live-streaming (Jattamart, Nusawat and Kwangsawad). I read it carefully and it is a sound study, but its outcome is the viewer's reluctance to pay after a hotel booking, and its organism constructs are personality traits from the Five-Factor Model together with dissatisfaction, financial concern and privacy concern. These sit outside the model tested here, which runs from streamer credibility, perceived interactivity and product demonstration quality through consumer trust and perceived value to purchase intention. Adding it would not have strengthened the argument, so I have left it out.

The two new references are [36] (in Section 2.1) and [81] (in Section 5.3).

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript presents a rigorous, well-executed empirical study. With a strong consumer sample, the analysis is methodologically well structured. However, a few critical conceptual, methodological, and contextual refinements are necessary to enhance clarity.

1. Theoretical Revisions

• The authors split Ohanian's traditional three-dimensional source credibility model to keep only competence-based parameters (expertise and trustworthiness), routing "attractiveness" into the perceived interactivity construct. While the authors cite Zhou & Lou (2022) to defend this cognitive/affective separation, they must state explicitly how physical or relational attractiveness conceptually interfaces with "interactivity" (which Liu defines as active control, synchronicity, and two-way exchange). Please add 2–3 sentences in Section 2.2 explaining this exact analytical bridging to prevent readers from viewing this exclusion as arbitrary.

• The authors rely on a global unidimensional give-get approach for perceived value rather than a multidimensional architecture like PERVAL (Sweeney & Soutar, 2001). Given that live streaming is intensely hedonic and social, treating value as global strips away our view of whether these stimuli drive emotional value versus monetary/price value. While acceptable, the authors should explicitly justify in Section 2.5 why the unidimensional approach was preferred for the SOR framework, and list this as a distinct item under the limitations section.

2. Methodological Clarifications

• The paper references a 6-month live commerce purchasing filter. Given that the respondents were recalling historical streams across an open mix of platform architectures (TikTok Shop, Whatnot, Amazon Live, etc.), how did the authors control for the systemic differences between these environments? A consumer evaluating a live auction stream on Whatnot likely rates "interactivity" fundamentally differently than someone watching a polished corporate show on Amazon Live. Please discuss how this variance was statistically checked or conceptualized, or specify if a specific "last viewed" anchor prompt was given to stabilize the indicators.

• In Table 2, item 2 for Streamer Credibility (0.674) and item 3 for Perceived Value (0.680) load slightly below the widely accepted 0.70 threshold. While the authors correctly note that they retained these because the structural fit and overall construct AVE were preserved, it would be beneficial to state this defense in the text alongside a reference to Hair et al. 

3. Discussion & Practical Implications

• The core structural finding of this paper is brilliant: Streamer Credibility has zero direct path to Perceived Value, but exerts its weight wholly through Consumer Trust. This means when it comes to human sources, value is a gate locked by trust. The authors must expand on this under Section 5.1 (Theoretical Implications). This asymmetry provides an outstanding response to Adaba et al. (2023) and highlights the evolutionary architecture of Western live-streaming setups.

• To better engage the journal's readership, translate the structural findings into clear feature recommendations. For instance, because trust must precede value evaluations for human actors, platforms (like TikTok Shop or Whatnot) should visually sequence trust-building badges (e.g., historical review consistency, third-party authentications, platform verification milestones) before or higher than flash-promotional value cues (e.g., ticking countdown timers or discount coupons) within the UI/UX pipeline.

Author Response

In addition to the point-by-point responses below, I have renumbered the references into their order of first appearance in the text to follow the journal's style, and I have made minor wording and table-presentation improvements throughout. None of these changes affect the data, analyses, results, or conclusions.

Comment 1: "The authors split Ohanian's traditional three-dimensional source credibility model to keep only competence-based parameters (expertise and trustworthiness), routing 'attractiveness' into the perceived interactivity construct. While the authors cite Zhou & Lou (2022) to defend this cognitive/affective separation, they must state explicitly how physical or relational attractiveness conceptually interfaces with 'interactivity' (which Liu defines as active control, synchronicity, and two-way exchange). Please add 2–3 sentences in Section 2.2 explaining this exact analytical bridging to prevent readers from viewing this exclusion as arbitrary"

Response: Thank you for this comment. I agree the bridge needed to be set out explicitly, and I have added three sentences to Section 2.2. I make clear that attractiveness and perceived interactivity are different constructs and that the interactivity items do not measure attractiveness. The link between them is the route they act through, not their items. In Zhou and Lou's study, attractiveness predicted the affective route to purchase (perceived pleasure) but did not predict product trust, while expertise and trustworthiness predicted trust, which is why attractiveness sits apart from the competence-based credibility construct. The bridge I then draw is that the affective response attractiveness provides in a recorded endorsement is, in a live stream, produced instead by responsive real-time interaction, the two-way replies, the viewer's control over what is shown, and the synchronous exchange that Liu treats as interactivity, which live-streaming research links to social presence and a more enjoyable experience. Perceived interactivity therefore carries this affective pathway, so excluding attractiveness removes a static personal cue rather than the affective route itself. The new text is in Section 2.2, in the paragraph that defines the credibility construct, just after the sentence stating that the affective influences are captured through perceived interactivity.

 

Comment 2: "The authors rely on a global unidimensional give-get approach for perceived value rather than a multidimensional architecture like PERVAL (Sweeney & Soutar, 2001). Given that live streaming is intensely hedonic and social, treating value as global strips away our view of whether these stimuli drive emotional value versus monetary/price value. While acceptable, the authors should explicitly justify in Section 2.5 why the unidimensional approach was preferred for the SOR framework, and list this as a distinct item under the limitations section"

Response: Thank you for this fair point, and for accepting the global approach. I have done both things you ask. In Section 2.5 I now set out why a single give-get construct fits the S-O-R model. Value occupies the organism stage as the consumer's overall judgement of the exchange, and the emotional, social, and functional sources of that value already enter the model at the stimulus stage, through perceived interactivity and through product demonstration quality and streamer credibility. Decomposing value into those components would place the same influences on both sides of the model, whereas a single give-get judgement keeps value as the downstream state through which the stimuli reach purchase intention. In Section 5.4 I have added this as a distinct limitation, noting that a single construct cannot show whether the stimuli act mainly through emotional or through monetary value, and that future work using dimension-specific value items would resolve this. The additions are in Section 2.5 (the paragraph defining perceived value) and Section 5.4 (limitations).

 

Comment 3: "…how did the authors control for the systemic differences between these environments? A consumer evaluating a live auction stream on Whatnot likely rates 'interactivity' fundamentally differently than someone watching a polished corporate show on Amazon Live. Please discuss how this variance was statistically checked or conceptualized, or specify if a specific 'last viewed' anchor prompt was given to stabilize the indicators"

Response: Thank you for this important point, which has been addressed in two ways. First, the survey did include a "last viewed" anchor. Respondents were asked to recall their most recent live streaming shopping experience and to rate all items with that single session in mind, rather than rating a platform in the abstract. This has now been made explicit in Section 3.2. Because every construct is measured as the consumer's own perception of one concrete stream, differences in objective platform features are absorbed into the individual ratings rather than left uncontrolled. Second, this was checked empirically. I compared construct scores across the platforms respondents used most often using one-way analyses of variance. None of the six constructs differed significantly across platforms (Streamer Credibility F(6,471) = 1.28, p = .264; Perceived Interactivity F = 1.12, p = .348; Product Demonstration Quality F = 0.26, p = .955; Consumer Trust F = 0.60, p = .734; Perceived Value F = 0.88, p = .513; Purchase Intention F = 0.48, p = .826), and platform explained under 2% of the variance in every case.

In a direct comparison of the auction-led platform (Whatnot) against the others, only streamer credibility differed at the uncorrected level (p = .019), and this did not survive Bonferroni correction for six tests; no other construct differed. I have added a short sentence reporting this check in Section 4.3.

 

Comment 4: "In Table 2, item 2 for Streamer Credibility (0.674) and item 3 for Perceived Value (0.680) load slightly below the widely accepted 0.70 threshold… it would be beneficial to state this defense in the text alongside a reference to Hair et al"

Response: Thank you for this helpful and precise comment. I agree. I have added a defence of the retained items to Section 4.2, citing Hair et al. (2019). The text now names the items below .70 and explains that they were kept because composite reliability and AVE for every construct already exceeded the recommended thresholds, so their removal would not have improved the measurement model.

 

Comment 5: "The core structural finding of this paper is brilliant: Streamer Credibility has zero direct path to Perceived Value, but exerts its weight wholly through Consumer Trust. This means when it comes to human sources, value is a gate locked by trust. The authors must expand on this under Section 5.1 (Theoretical Implications). This asymmetry provides an outstanding response to Adaba et al. (2023) and highlights the evolutionary architecture of Western live-streaming setups"

Response: Thank you, I am glad the finding came through, and I agree it deserved more space. I have expanded the discussion of the asymmetry in Section 5.1 and drawn the contrast with Adaba et al. more sharply. The expanded passage makes the point that trust does opposite work in the two studies. In their influencer setting the influencer-to-trust path was strong but did not convert, with trust relating negatively to purchase intention, which they explain as overtrust and persuasion knowledge. In the present live-commerce model trust is the only route by which competence-based credibility reaches perceived value, it also supports purchase intention directly, and the chain completes rather than breaking. I suggest the difference is not that UK consumers trust more, but that the live, demonstration-led setting attaches trust to a concrete exchange the viewer can appraise immediately and channels it into a give-get judgement about that exchange, rather than to a persona standing at one remove from any single purchase. The expanded text is in Section 5.1.

 

Comment 6: "To better engage the journal's readership, translate the structural findings into clear feature recommendations. For instance, because trust must precede value evaluations for human actors, platforms (like TikTok Shop or Whatnot) should visually sequence trust-building badges (e.g., historical review consistency, third-party authentications, platform verification milestones) before or higher than flash-promotional value cues (e.g., ticking countdown timers or discount coupons) within the UI/UX pipeline"

Response: Thank you for this concrete suggestion, which I have taken up almost exactly. In Section 5.3 I have turned the sequencing point into interface guidance. Because trust gates value, the cues that build trust (verification badges, third-party authentication, and visible seller history or review consistency) should be encountered before, or placed above, flash-promotional devices such as countdown timers and discount coupons, rather than left to a buried profile tab. I have framed these trust cues as the structural heuristics that Mattathil et al. [10] describe, so the recommendation is anchored in their platform-trust account. The revised text is in Section 5.3, in the paragraph on the asymmetric investment logic.

Reviewer 3 Report

Comments and Suggestions for Authors

I have read the manuscript with great interest. The study offers a solid application of the S-O-R framework to the UK live-streaming market, effectively untangling the dual mediating roles of trust and perceived value. The asymmetric pathway findings regarding streamer credibility are particularly interesting and well-argued. The methodology is sound and the paper is generally well-written. Just provide some minor suggestions:

First, while the focus on a non-Chinese sample is clearly a major strength of this paper, the theoretical motivation for this could be slightly sharper. I suggest adding a brief discussion early in the introduction or background section about the specific characteristics of UK consumers—such as differing privacy concerns, platform habits, or skepticism toward influencers compared to mature Asian markets. This will better justify why testing these specific trust pathways is particularly relevant in the UK context.

Second, regarding the methodology, a couple of minor reporting clarifications are needed. Please explicitly state the timeframe for data collection in the methodology section. Additionally, since PLS-SEM is highly prevalent in current e-commerce literature, it would be helpful to include a brief sentence justifying the choice of covariance-based SEM (AMOS), perhaps simply emphasizing the theory-confirmation nature of your study.

Finally, a bit of housekeeping is needed in the results and discussion sections. The measurement model reporting is thorough, but the manuscript currently lacks the Means and Standard Deviations for the main constructs. Please add these descriptive statistics, either by integrating them into Table 3 or providing a separate brief table. Furthermore, your discussion brings up the interesting contrast between platforms like Whatnot (auction-led) and TikTok Shop (entertainment-led). Expanding the practical implications by just a few sentences to offer tailored advice for sellers on these distinct types of platforms would make your conclusion much stronger and more actionable.

Author Response

In addition to the point-by-point responses below, I have renumbered the references into their order of first appearance in the text to follow the journal's style, and I have made minor wording and table-presentation improvements throughout. None of these changes affect the data, analyses, results, or conclusions.

Comment 1: "First, while the focus on a non-Chinese sample is clearly a major strength of this paper, the theoretical motivation for this could be slightly sharper. I suggest adding a brief discussion early in the introduction or background section about the specific characteristics of UK consumers—such as differing privacy concerns, platform habits, or skepticism toward influencers compared to mature Asian markets. This will better justify why testing these specific trust pathways is particularly relevant in the UK context"

Response: Thank you for this suggestion. I agree the motivation for the UK sample could be sharper. I have added a short passage early in the Introduction that sets out how UK consumers differ from shoppers in the mature Asian markets where live commerce grew, covering the three points you raise: a greater readiness to read commercial intent into endorsements, different platform habits, and stronger data-protection expectations. These points are supported by references, some of which are new to the paper. The passage is in Section 1, in the paragraph that begins "The UK live-commerce environment makes such a test particularly pointed".

 

Comment 2: "Please explicitly state the timeframe for data collection in the methodology section"

Response: Thank you for noting this. I have added the timeframe. Data were collected in January 2026, as now stated in Section 3.2.

 

Comment 3: "…since PLS-SEM is highly prevalent in current e-commerce literature, it would be helpful to include a brief sentence justifying the choice of covariance-based SEM (AMOS), perhaps simply emphasizing the theory-confirmation nature of your study"

Response: Thank you for this useful suggestion. I agree. I have added a short justification to Section 3.5, where I explain that covariance-based SEM was chosen because the study tests an established theoretical model for confirmation rather than prediction, and because covariance-based SEM provides global fit indices suited to this aim. This addition can be found in Section 3.5.

 

Comment 4: "…the manuscript currently lacks the Means and Standard Deviations for the main constructs. Please add these descriptive statistics, either by integrating them into Table 3 or providing a separate brief table"

Response: Thank you for pointing this out. I agree. I have added the mean and standard deviation of each construct to Table 3, as you suggested, so the descriptive statistics now sit together with the correlation and discriminant validity information. This can be found in Section 4.2 and Table 3.

 

Comment 5: "Furthermore, your discussion brings up the interesting contrast between platforms like Whatnot (auction-led) and TikTok Shop (entertainment-led). Expanding the practical implications by just a few sentences to offer tailored advice for sellers on these distinct types of platforms would make your conclusion much stronger and more actionable"

Response: Thank you, this is a helpful suggestion and I agree it makes the implications more actionable. I have added a short paragraph to Section 5.3 giving sellers tailored advice for the two formats. On Whatnot, where auctions are fast and trade in high-value, often non-returnable collectibles, the binding constraint is trust in authenticity, so the priorities for a seller are a visible track record, third-party authentication or grading, and close-up demonstration of condition, all established before the bidding and its urgency take over. On TikTok Shop, which leads with entertainment and discovery and often features unfamiliar sellers and impulse goods, the risk is the non-conversion Adaba et al. [25] document, so the task is to carry credibility and substantive demonstration into the entertaining content rather than leaning on promotion alone. I close by noting that the model points the same way on both, trust before value, but the cue in shortest supply differs, authentication on Whatnot and credible demonstration on TikTok Shop. The new paragraph is in Section 5.3.