How Prolific Sellers Self-Present: Dissecting the Communication Patterns of 1.6 Million Reverb Listings

A Primer on Psychology of Language Research: The Words-As-Attention Framework

• Words can mark psychological focus and attention; frequent positive-emotion terms indicate attention to positive affect, not necessarily that the writer feels positive.The framework has received substantial support across many studies.Source: A Primer on Psychology of Language… · p. 5
• Content words, such as nouns and verbs, show what people communicate and can cluster into themes; healthy-city residents wrote more about home cooking and less about alcohol (Blackburn et al., 2018).Source: A Primer on Psychology of Language… · p. 5
• Function words such as articles, prepositions, and pronouns reveal communication style; higher-performing students used more articles and prepositions but fewer storytelling words, informing the analytic thinking index (Pennebaker et al., 2014).The index represents a continuum of communicative and thinking style.Source: A Primer on Psychology of Language… · pp. 5–6
• Verbal structure captures how words and sentences are arranged, often through readability and lexical diversity; papers by scientists who committed data fraud were more complex and less readable.Source: A Primer on Psychology of Language… · pp. 6–7

• Self-presentation theory treats language in product descriptions as a way sellers strategically manage impressions and reveal information about themselves.Source: Self-Presentation, Warranting, and… · pp. 7–8
• Warranting theory holds that information is more evidentiary when communicators have less ability to manipulate it.Source: Self-Presentation, Warranting, and… · p. 8
• Instrument specifications are verifiable, while claims about sound are subjective; prolific sellers may experience this verification dynamic more intensely.Source: Self-Presentation, Warranting, and… · p. 9
• Buyers and sellers rate one another, and incomplete or inaccurate descriptions can risk future sales while reflecting on seller credibility.Source: Self-Presentation, Warranting, and… · p. 9
• The paper examines whether Reverb sellers’ verbal descriptions vary with prolific versus everyday status, independently of inventory size.Source: Self-Presentation, Warranting, and… · p. 9

• Reverb offers an at-scale self-presentation setting because sellers write open-ended descriptions that matter to buyers before purchase, and the platform tracks selling volume.Source: The Current Paper · p. 10
• The study evaluates seller language across content, communicative style, and verbal structure, following a words-as-attention framework.Warranting theory motivates the content prediction, but does not directly predict style or structure.Source: The Current Paper · p. 10
• Prolific sellers are predicted to emphasize verifiable, objective product attributes over unverifiable, subjective ones because repeated buyer scrutiny gives objective claims greater warranting value.Descriptions are checked against products on delivery and recorded in public feedback, while buyers cannot evaluate subjective impressions before purchasing.Source: The Current Paper · pp. 10–11
• Style and structural differences are examined exploratorily because prior self-presentation research does not specify their direction in this marketplace.The corresponding research questions ask how both dimensions differ between prolific and everyday sellers.Source: The Current Paper · p. 11
• The analysis covers 1.6 million Reverb listings and accounts for visual listing elements and seller characteristics to isolate language-related differences.Examples of seller characteristics include whether Reverb designates a seller as “preferred.”Source: The Current Paper · p. 11
• The study measures content with bottom-up thematic extraction, style with an analytic thinking index, and structure with readability and type-token ratio metrics.Source: The Current Paper · p. 11

• The initial collection contained 2.1 million Reverb listings, from which the API supplied descriptions, listing duration, price, category, visual metrics, and listing state.Source: Data Collection and Preprocessing · pp. 11–12
• Only live listings were retained, and non-English descriptions were removed, yielding a final sample of 1,628,876 listings.Ended and sold listings were excluded because they were infrequently available and often archived from the API.Source: Data Collection and Preprocessing · p. 12
• The final sample covered over 300 million words, averaging 201.97 words per listing (SD = 220.02; Mdn = 137; range = 1–11,460).Source: Data Collection and Preprocessing · p. 12

Automated Text Analysis

• LIWC counted words in each description as percentages of total word count across human-validated social, psychological, and part-of-speech categories.Source: Automated Text Analysis · p. 12

Content Patterns: The Meaning Extraction Method

• The Meaning Extraction Method removed function and low-base-rate words, retained content words, and coded each word’s presence or absence in each description (Chung & Pennebaker, 2008; Markowitz, 2021).Source: Content Patterns: The Meaning… · p. 13
• The method used unigrams, bigrams, and trigrams as textual units for extracting meaning.Source: Content Patterns: The Meaning… · p. 13
• Principal Component Analysis with varimax rotation clustered content words into themes, retained for variance explained, eigenvalues above one, and face-valid interpretability (Chung & Pennebaker, 2008; Foxman et al., 2021; Markowitz, 2021).Source: Content Patterns: The Meaning… · p. 13
• The extraction process retained 8 themes.Source: Content Patterns: The Meaning… · p. 13
• The themes were saved as standardized regression weights (M = 0, SD = 1) for later inclusion in statistical models.Source: Content Patterns: The Meaning… · p. 13

Style Patterns: Analytic Thinking

• The analytic-thinking index is a standardized composite of eight function-word categories, with positive loadings for articles and prepositions and negative loadings for six other categories.The negatively loaded categories are auxiliary verbs, adverbs, personal pronouns, impersonal pronouns, negations, and conjunctions.Source: Style Patterns: Analytic Thinking · p. 13
• Low analytic-thinking scores indicate dynamic, storytelling, informal communication, whereas high scores indicate hierarchical, formal communication; scores around 50 are middle of the road.The scores are standardized against comparison corpora.Source: Style Patterns: Analytic Thinking · pp. 13–14

Structural Patterns: Readability and Type-Token Ratio

• Flesch Reading Ease uses words per sentence and syllables per word, with higher scores indicating more readable text.The metric is Flesch Reading Ease.Source: Structural Patterns: Readability and… · p. 14
• Type-token ratio measures lexical diversity: high values indicate more unique and varied terms, while low values indicate greater consistency or term reuse.Source: Structural Patterns: Readability and… · p. 14
• Both readability and type-token ratio were measured with the quanteda.textstats package in R.The cited package reference is Benoit et al. (2021).Source: Structural Patterns: Readability and… · p. 14

Prolific Seller Identification

• Seller type was defined by feedback, with the top 10% classified as prolific and the rest as everyday sellers.The analysis excluded one-off transactions by restricting the sample to sellers with at least five active listings.Source: Prolific Seller Identification · p. 14
• 1,484 prolific sellers accounted for 1,131,989 listings (69.5% of the sample), while 13,343 everyday sellers accounted for 496,887 listings.The analysis was restricted to sellers with at least five active listings; prolific status meant falling in the top decile (10%).Source: Prolific Seller Identification · p. 14

Analytic Plan and Covariates

• Language-pattern outcomes were modeled with prolific seller status against everyday sellers as the reference group, using seller-clustered standard errors.The authors chose this approach over a linear mixed model because unobserved seller characteristics could correlate with feedback volume and violate the random-effects assumption; key-outcome intraclass correlations ranged from 0.13 to 0.63.Source: Analytic Plan and Covariates · p. 15
• Models adjusted for price, seller inventory, photo and video counts, listing age, product category and condition, preferred-seller status, and listing views, offers, and saves.The active-days fixed effects enabled comparisons among listings of the same age.Source: Analytic Plan and Covariates · pp. 15–16
• Price was logged and standardized within product category, inventory size was logged, and views, offers, and saves were log-transformed and standardized before averaging into an engagement index.These transformations are described for price, inventory, and the engagement components, respectively.Source: Analytic Plan and Covariates · pp. 15, 17
• .299 < rs < .725 for the positive, statistically significant intercorrelations among the three log-transformed engagement metrics.These metrics were views, saves, and offers.Source: Analytic Plan and Covariates · pp. 16–17

Results

• Table 1 presents descriptive statistics for all variables and their intercorrelations.Source: Results · p. 17

Content Patterns

• Table 2 describes eight extracted themes, grouped into three higher-order collections.Source: Content Patterns · p. 17
• The product-specifications collection includes guitar body (Component 1), drums (Component 3), electronics (Component 5), response latency (Component 7), and power supply information (Component 8).Source: Content Patterns · p. 17
• The other collections are post-purchase information (shipping, Component 2, and customer service) and acoustics (tone, Component 6).Source: Content Patterns · p. 17
• At p = .001, compared with everyday sellers, prolific sellers focused more on guitar-body details such as frets and scale length, and guitar-bridge details; top component words described functional, structural, or playability features, not look or feel.Source: Content Patterns · pp. 17–18
• At p = .039, prolific sellers used fewer tone-related terms, such as rich, warm, smooth, and clarity, than everyday sellers, partially supporting H1.The difference concerned subjective sound-related claims about an instrument.Source: Content Patterns · p. 18

Style Patterns

• Prolific sellers’ lower analytic-thinking scores corresponded to a more informal, dynamic, narrative style, accompanied by more personal pronouns than everyday sellers.Source: Style Patterns · p. 18

Structural Patterns

• Readability did not differ by seller type, and the type-token-ratio difference became nonsignificant after controlling for listing length.Source: Structural Patterns · pp. 18–19
• Prolific sellers’ descriptions contained approximately 16% more text than everyday sellers’ listings.Source: Structural Patterns · p. 19

Discussion

• The discussion frames seller presentation as multidimensional: prolific sellers paired technical, specs-focused content with less-analytic, narrative style and elevated personal-pronoun use.Source: Discussion · p. 20
• The discussion interprets prolific sellers’ preference for objective claims as consistent with warranting logic, as repeated buyer verification favors claims that withstand scrutiny.Source: Discussion · pp. 21–22
• Further research is needed to establish whether the patterns identified at scale on Reverb or similar marketplaces hold across other online marketplaces and self-presentation settings.Source: Discussion · p. 22

Practical Implications

• The findings provide practical evidence about how more experienced sellers communicate compared with less experienced sellers, while remaining noncausal and small in magnitude.They empirically describe what prolific sellers communicate and how they communicate it relative to everyday sellers.Source: Practical Implications · p. 22
• Platforms could advise sellers to emphasize verifiable specifications and objective details so buyers can assess product claims before purchase.The results suggest prolific sellers favor verifiable content and specifications over speculation; if objective details signal expertise, platform advice could encourage such communication.Source: Practical Implications · p. 23
• The tone finding suggests novice sellers may rely on subjective claims when they have few concrete product details to offer.This is a suggested interpretation of the tone result, not a causal conclusion.Source: Practical Implications · p. 23

Limitations and Future Directions

• Effect sizes were small, and feedback-based prolific-seller status conflated transaction volume with platform tenure; transaction-level data could separate experience from sales output.The study used accumulated feedback rather than direct sales volume, and future transaction-level data could clarify how experience and actual sales output relate to language.Source: Limitations and Future Directions · p. 23
• Because the study was observational and excluded already-sold listings, it cannot establish causality or determine whether seller language differences translate into sales.Reverb does not systematically index sold listings through its API for analysis.Source: Limitations and Future Directions · p. 23
• The English-only corpus from one specialized marketplace leaves open whether the observed content, style, and structural patterns generalize to marketplaces with less expert buyers.Reverb buyers may already have domain knowledge, potentially making specifications more valuable there than in general-purpose marketplaces.Source: Limitations and Future Directions · p. 23

Intercorrelations Between Variables

• The C1–C8 intercorrelations were zero because the components were fully orthogonal under varimax rotation.C1–C8 are detailed in Table 2.Source: Intercorrelations Between Variables · p. 32, Table 2 · p. 33
• Prolific seller status predicted lower tone (B = -0.057, p = .039), while video count predicted higher tone (B = 0.105, p < .001).Both effects are reported in the fixed-effect content results.Source: Table 3 · p. 34, Intercorrelations Between Variables · p. 35
• Photo count was positively associated with customer service, response latency, and power; price and video count were positively associated with response latency.The reported associations were statistically significant (p < .001 for customer-service photo count and latency price; p = .003 for latency video count and p < .001 for power photo count).Source: Intercorrelations Between Variables · pp. 34–35
• Prolific seller status was not significantly associated with electronics or response latency, and its association with power was also nonsignificant.The power estimate was marginal (p = .061); the electronics and response-latency estimates had p = .679 and p = .993, respectively.Source: Intercorrelations Between Variables · p. 35