{"id":"56ec19aa-e49b-4810-8274-3c7aa304db06","arxiv_id":"2412.00233","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":2.0,"correctness_risk":"high","formal_verification":"none","parameter_count":3,"one_line_summary":"The paper reports that roommate participation raises a student's probability of joining Double 11 shopping by 18.6 to 23.5 percentage points, with women and less experienced shoppers more affected.","lead":"This paper claims that college students are much more likely to join Double 11 online shopping when their roommates participate, based on a Bayesian Probit analysis of 200 questionnaires. A smart generalist should care because it is an example of how small, self-reported surveys can produce confident causal-sounding claims about social influence.","discovery_kind":"incremental","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'Peer Effects' regressor is an agreement dummy (roommate behavior consistent with respondent), not roommate participation; the 18.6%-23.5% effect in Table 2 is therefore not identified as a peer effect.","rationale":"The paper's central claim is a causal peer-effect estimate: roommate participation raises the probability of the respondent's participation by 18.6%-23.5%. For that claim to hold, the regressor must measure roommate participation. Table 1 shows it does not: 'Peer Effects' is coded 1 when the roommate's behavior is consistent with the respondent, an agreement indicator. Since the respondent's own participation is the dependent variable, the regressor is a function of the outcome; the coefficient can be positive even under the null of independence, as the simple calculation in my attack shows. This is a more fundamental failure than the endogeneity concern in the reader's report: even with perfectly exogenous roommate behavior, the Table 2 estimates do not identify the stated effect. The paper also promises but never displays the model equations in Section 2, so the exact marginal-effect target is unverifiable. I agree with the reader's reject verdict, but my reason is the measurement/identification mismatch rather than the Bayesian prior critique. No independent support (code, machine-checked proofs, or falsifiable out-of-sample predictions) offsets this issue. The proposed concrete test is decisive: if the raw roommate-participation variable exists and produces the same result, the concern would be resolved; if not, the headline claim fails.","tokens_in":6499,"tokens_out":7011,"duration_ms":66353,"concrete_test":"Using the original questionnaire, build R_i = 1 if the respondent reports that the roommate actually participated in Double 11 (not whether their behaviors agree) and re-estimate the six specifications in Table 2 with R_i replacing the agreement dummy. If R_i is unavailable in the data or its marginal effect is not in the 18.6%-23.5% range, the headline causal claim is not identified.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3 Table 1 defines the peer-effects variable as 'whether the roommate's behavior is consistent with the respondent', i.e., D_i = 1{Y_i = R_i}, not R_i = roommate participation. Section 4.1 Table 2 then interprets the coefficient on D_i as 'participation of a roommate' raising the probability of participation by 18.6%-23.5%. This interpretation is unsupported: because D_i is partly determined by the outcome Y_i, a positive coefficient can be purely mechanical. To see this, use the reported means P(Y=1)=0.834 and P(D=1)=0.811. Under the null of no peer effect (Y independent of R), these means imply P(R=1) is about 0.966; then P(Y=1|D=1) is about 0.993 and P(Y=1|D=0) is about 0.152, so a probit on D alone produces a large positive coefficient with no causal effect. The paper's promised model equations in Section 2 are also absent, so the estimand cannot be checked. The central claim therefore rests on a regressor that does not measure what the conclusion says it measures.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper uses questionnaire data from university students to estimate a Bayesian Probit model of participation in China's \"Double 11\" shopping festival. It reports that a 'peer effects' variable raises the probability of participation by 18.6% to 23.5%, and that gender, online shopping experience, and fashion consciousness are also significant. The paper frames the analysis as an information-cascade or herd-behavior study and derives marketing and policy implications.","tokens_in":6752,"tokens_out":6953,"duration_ms":57605,"significance":"The topic is relevant to consumer behavior and e-commerce marketing. The paper is transparent in reporting variable definitions and in presenting both classical and Bayesian estimates, and it includes a wide set of control variables. However, the central coefficient is not identified as a peer effect because the regressor is defined as agreement between the roommate's and the respondent's behavior, not as the roommate's participation. The descriptive statistics are internally inconsistent, the promised model equations are absent, and the conclusion claims an instrumental-variable procedure that is not performed. These problems undermine the paper's headline claims, so the manuscript in its current form does not provide a reliable empirical contribution.","major_comments":[{"comment":"The 'Peer Effects' regressor is defined in Table 1 as 'whether the roommate's behavior is consistent with the respondent,' i.e., D_i = 1{Y_i = R_i}, not as the roommate's participation R_i. The text in Section 4.1 interprets the coefficient on this variable as the causal effect of a roommate's participation on the respondent's participation. This interpretation is invalid: because D_i is constructed from the respondent's own outcome Y_i, a positive coefficient can be purely mechanical. Using the reported means P(Y_i=1)=0.834 and P(D_i=1)=0.811, under the null that Y_i is independent of R_i the implied P(R_i=1) is about 0.966, and a probit of Y_i on D_i alone would produce a large positive marginal effect even when no true peer effect exists. The paper's causal claims therefore rest on a regressor that does not measure what the conclusions say it measures.","section":"Section 3, Table 1; Section 4.1, Table 2"},{"comment":"The abstract and Section 3.2 state that the survey collected 200 valid responses, but every column in Table 2 reports 204 observations. The paper does not explain this discrepancy, and it compromises the descriptive statistics in Table 1 if the means are computed over a different sample than the regressions.","section":"Section 3.2 and Table 2"},{"comment":"The conditional participation shares reported in this section are logically inconsistent. If 81.1% of respondents with peer effects and 18.9% of respondents without peer effects participated, then with P(peer effects)=0.811 the overall participation rate would be 0.811*0.811 + 0.189*0.189 = 0.693, not the reported 83.4%. Conversely, if 81.1% and 18.9% are shares of participants and nonparticipants with peer effects, the implied P(peer effects) is 0.708, not 0.811. The text cannot reconcile the numbers under either reading, so the descriptive claim about peer effects is not supported by the data as presented.","section":"Section 4.2.2"},{"comment":"The two subsections under 'Specific Relationship Formulas' promise equations for the consumer decision process and the Bayesian equilibrium, with the text stating 'The specific relationship formula is as follows' and referring to 'i*' and to 'α and β'. No equations are actually displayed in the manuscript. As a result, the likelihood, the priors, and the equilibrium condition are not defined, and the reported marginal effects cannot be traced to a precise estimand.","section":"Section 2.2"},{"comment":"The first paragraph of the Conclusion claims that the result 'remains robust even after rigorous statistical corrections using instrumental variables.' No instrumental-variable estimation appears anywhere in the paper; Table 2 reports only traditional and Bayesian Probit estimates. This assertion is unsupported and appears to describe a method that was not carried out.","section":"Section 5.1"}],"minor_comments":[{"comment":"The significance note under Table 2 reads '***p < 0.01, *p < 0.05' but the table also uses '**' for Fashion Consciousness in column (1); the note should be completed to include the '**' level.","section":"Table 2 note"},{"comment":"The bar chart labels are in Chinese ('有同伴效应' and '无同伴效应') with no English equivalents; for an English-language manuscript, the labels and the figure title should be translated.","section":"Section 4.2.2, chart"},{"comment":"References [6]-[9] concern deep learning, plant disease detection, potato production, and neural radiance fields; they are not connected to the peer-effects or consumer-behavior analysis, and the sentence citing them at the end of Section 5.2 does not motivate their inclusion. These references should be removed or properly integrated.","section":"References [6]-[9]"},{"comment":"The list of variables in the text includes 'grade' with range 0 to 5 and 'income' with range 1 to 5, but the table notation '0' for freshman is inconsistent with the common coding 1=freshman; please clarify the coding to match the definitions.","section":"Section 3.1"}],"recommendation":"reject","confidential_remarks":"The manuscript appears to be a rough translation with missing equations and an unrelated reference list. The measurement problem in the key regressor is fundamental; even if the roommate-participation variable could be reconstructed from the survey items, the paper would need to be rewritten and re-estimated. The unsupported instrumental-variable claim and the sample-size discrepancy further indicate that the manuscript is not ready for publication. I recommend rejection rather than major revision because the current analysis does not support any of its headline conclusions, and a sound version would require substantial new analysis rather than local corrections."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear [name], quick take: the headline effect is not a peer effect. The 'peer effects' variable is defined as whether the roommate's behavior is consistent with the respondent (Table 1), so it is partly determined by the respondent's own participation. The stress-test note is right: with P(Y=1)=0.834 and P(D=1)=0.811, a mechanical correlation is almost forced. Table 2's coefficient on this agreement dummy cannot be interpreted as 'participating roommate raises participation by 18.6-23.5%.' That kills the paper's central claim.\n\nWhat's genuinely here: new survey data from Beijing university students on Double 11, and a Bayesian Probit application to a binary outcome, which is a reasonable modeling choice. The variable definitions in Table 1 are mostly clear, and the prose is readable. That's about it. The paper does not articulate a new theoretical result or a new mechanism; its own reference [5] (Xi and Wu, 2020) is the same model, same festival, same question. The new questionnaire does not create a new contribution.\n\nThe soft spots are multiple and load-bearing. Section 2 promises model equations but none are shown; the 'Bayesian equilibrium' is a verbal restatement of the information-cascade assumption. The conclusion claims results 'remain robust after rigorous statistical corrections using instrumental variables' and mentions interaction analyses, but no IV or interaction results appear in the paper. The numbers don't line up either: abstract says 200 valid responses, Table 2 reports 204 observations; Section 4.2.2 gives 81.1% vs 18.9% participation conditional on 'peer effects present/absent,' which cannot be reconciled with the overall 83.4% participation rate. These are not typos that a good referee would fix; they undermine confidence in the whole data handling.\n\nMy verdict: this paper is not ready for peer review. The central estimand is mislabeled, the supporting analysis is internally inconsistent, and the novelty relative to [5] is not articulated. A serious referee would spend the whole report on the regressor definition and missing analyses. I would desk-reject, and tell the authors to rework the design so that the peer variable is actual roommate participation, not an agreement dummy, and to reconcile every number before resubmitting.\n\nHard to find a reader who gets much from this as-is. Skip it.","headline":"The paper's headline peer effect is an artifact of using an agreement dummy as the regressor; without that, there is little new beyond a re-run of reference [5].","tokens_in":7260,"tokens_out":2336,"would_cite":false,"duration_ms":21502,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that a roommate's participation in the Double 11 shopping festival raises a student's probability of participating by 18.6 to 23.5 percentage points.","keywords":["peer effects","herd behavior","Double 11 shopping festival","Bayesian Probit model","online shopping","consumer behavior"],"falsifier":"Estimate the same participation equation on data in which roommates are exogenously assigned, or instrument roommate participation with an administrative dorm-assignment rule, and test whether the marginal effect of roommate participation still falls in the 18.6–23.5% range after controlling for shared dorm-level unobservables; if it shrinks toward zero, the causal claim fails.","tokens_in":6293,"feed_emoji":"🛍️","tokens_out":10805,"duration_ms":93584,"temperature":0.7,"pith_summary":"This paper asks whether college students shop on Double 11 because their roommates do. Using a questionnaire of 200 university students and a Bayesian Probit model, it reports that when peer effects are present, measured as the roommate's behavior matching the respondent's, the probability of participating rises by 18.6% to 23.5%, significant at the 1% level. The paper also reports that gender, prior online shopping experience, and fashion consciousness shape herd behavior, with female students about 10% more likely to conform under peer influence. If the effect is real, the basic unit of consumer behavior is not the isolated shopper but the peer group, which would make dormitories a natural target for marketing, regulation, and consumer education.","feed_headline":"Roommate effect raises Double 11 participation by 18.6-23.5%","feed_subtitle":"Bayesian Probit analysis of 200 students ties roommate behavior to Double 11 participation.","key_machinery":"The engine of the paper is a Bayesian Probit model on a binary outcome: participation in Double 11 (1/0). The main regressor is a self-reported peer-effects dummy indicating whether the roommate's behavior matched the respondent's. The paper describes the decision process as an information cascade in which later consumers imitate earlier ones, and it formalizes the trade-off as a Bayesian equilibrium between trust in personal information ($\\alpha$) and trust in peer behavior ($\\beta$). The model's role is to turn the survey data into marginal effects, giving the headline 18.6%–23.5% estimate while using priors to stabilize inference with 200 observations.","core_discovery":"The paper's central claim is that offline peer effects directly cause herd behavior in the Double 11 shopping festival. Using both a traditional Probit and a Bayesian Probit model on survey data, it finds a marginal effect of the peer-effects variable between 18.6% and 23.5%, all significant at the 1% level, meaning a student whose roommate participates is roughly one-fifth more likely to participate. The paper further claims that online shopping experience, gender, and fashion consciousness are significant drivers of conformity, and that female consumers exhibit about 10% greater conformity under peer influence than males. The author states this as a direct, causal effect of roommate behavior on consumer decisions.","pith_inferences":["A testable extension: randomly assign a small participation incentive to one student in each dormitory room and measure the spillover onto roommates; the paper's estimates imply a within-room multiplier that a field experiment could quantify.","A caveat beyond the paper's claims: because the peer-effect variable is the respondent's own report of behavioral consistency, the reported estimate probably captures selection into rooms and shared unobservables as well as influence, so the causal reading is stronger than the displayed identification.","A further extension: separate the two channels the paper conflates—informational influence (learning that roommates shop) and social utility (wanting to match roommates)—by collecting independent reports from both roommates rather than one self-reported alignment measure.","The conclusion states that instrumental-variable corrections confirm the effect, but no IV regression appears in the methods or results sections; readers should treat the causal language as an assertion rather than a displayed result."],"forward_implications":["Dormitories and other offline peer groups become viable marketing units: nudging one roommate can raise group participation by roughly one-fifth.","E-commerce promotions should treat offline word-of-mouth and dorm-level peer dynamics as complements to online reviews, not substitutes.","Because participation is driven partly by conformity, a negative signal about the event can spread through the same peer channel, supporting the paper's 'tide-out effect' concern.","Gender-specific communication and campaign design may be effective, since the paper reports female students are about 10 percentage points more likely to conform under peer influence.","The stability of the marginal effect across traditional and Bayesian Probit specifications suggests the qualitative result is not an artifact of estimator choice within the paper's model class."],"supporting_citations":[{"why":"Documents university-student conformity psychology in Taobao Double 11 shopping, the behavioral pattern this paper extends.","marker":"[1]"},{"why":"Provides earlier Double 11 conformity-effect evidence that motivates the paper's peer-effects hypothesis.","marker":"[2]"},{"why":"Supplies the female-consumer impulsive-buying finding behind the paper's gender-conformity claim.","marker":"[4]"},{"why":"Introduces the Bayesian Probit estimation strategy that produces the paper's headline marginal effects.","marker":"[5]"}],"fun_headline_variants":["Roommate effect ups Double 11 odds by 18.6-23.5%","Peer pressure boosts Double 11 participation: ~20% increase","Bayesian Probit: Roommates raise your Double 11 chances by ~1/5","Herd behavior: Roommate participation lifts shopping odds 18.6-23.5%","Roommate's Double 11 habit boosts your odds by 18.6-23.5%"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the self-reported consistency between the respondent's and the roommate's behavior is an exogenous peer effect rather than a shared dorm-level cause; the paper's conclusion asserts IV corrections support this, but the methods and results sections present no IV specification or formal exogeneity test.","fun_headline_variants_meta":{"raw":{"variants":["Roommate effect ups Double 11 odds by 18.6-23.5%","Peer pressure boosts Double 11 participation: ~20% increase","Bayesian Probit: Roommates raise your Double 11 chances by ~1/5","Herd behavior: Roommate participation lifts shopping odds 18.6-23.5%","Roommate's Double 11 habit boosts your odds by 18.6-23.5%"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001103,"raw_usage":{"total_tokens":4528,"prompt_tokens":803,"completion_tokens":3725,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":419,"completion_tokens_details":{"reasoning_tokens":3622}},"tokens_in":419,"tokens_out":3725,"duration_ms":25320,"temperature":1.0,"reasoning_tokens":3622,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T05:34:51.904040+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Estimate the same participation equation on data in which roommates are exogenously assigned, or instrument roommate participation with an administrative dorm-assignment rule, and test whether the marginal effect of roommate participation still falls in the 18.6–23.5% range after controlling for shared dorm-level unobservables; if it shrinks toward zero, the causal claim fails.","supporting_citations":[{"cited_title":"Double 11","cited_arxiv_id":null,"evidence_quote":"Documents university-student conformity psychology in Taobao Double 11 shopping, the behavioral pattern this paper extends."},{"cited_title":"One of its significant advantages is its ability to integrate prior information, providing a more flexible and precise model specification","cited_arxiv_id":null,"evidence_quote":"Provides earlier Double 11 conformity-effect evidence that motivates the paper's peer-effects hypothesis."},{"cited_title":"Double 11","cited_arxiv_id":null,"evidence_quote":"Supplies the female-consumer impulsive-buying finding behind the paper's gender-conformity claim."},{"cited_title":"Double 11","cited_arxiv_id":null,"evidence_quote":"Introduces the Bayesian Probit estimation strategy that produces the paper's headline marginal effects."}],"review_version":1}