1claim_id,claim,stream,theory_or_empirical,supporting_keys,what_they_show,limitations_for_WP38,wp38_position2C01,"Economic activity is spatially concentrated well beyond what a random (dartboard) allocation would produce, at several spatial scales",1,empirical,ellison1997; duranton2005; rosenthal2001; mori2005; baldwin2008,"US Census/County Business Patterns (Ellison-Glaeser), UK micro-geographic plant data with distance-based K-density tests and counterfactual nulls (Duranton-Overman), KL-divergence statistic (Mori et al.), Canadian plant microdata (Baldwin et al.): most manufacturing industries are localised; concentration is heterogeneous across industries","All measure employment/plants in production industries; WP38 observes listings of online catalogues whose location is the merchant's declared address, with 17 coarse regions rather than points",WP38 re-uses the dartboard logic (merchant-count-preserving permutation nulls for RAA and for the bipartite network) but on listing counts; no claim about production localisation3C02,"Agglomeration economies arise from sharing, matching and learning; density elasticities are modest (≈2–8%) and hard to identify",1,theory+empirical,duranton2004; rosenthal2004; combes2015; duranton2020; marshall1920,Theoretical micro-foundations (Duranton-Puga) and surveys of wage/productivity-density elasticities with instrumenting strategies (Combes-Gobillon; Duranton-Puga 2020),"Outcomes are productivity/wages; WP38 has no productivity, employment or sales outcome — only catalogue composition","WP38 only tests whether assortment diversity correlates with population/scale (H2); explicitly descriptive, no elasticity interpretation as agglomeration economy"4C03,Whether specialisation (MAR) or diversity (Jacobs) favours local growth is unresolved; related variety matters more than unrelated variety,1,empirical,glaeser1992; henderson1995; frenken2007; content2016,"City-industry employment growth regressions (US 1956–87; Dutch regions 1996–2002 with entropy decomposition of variety); related variety associated with employment growth, unrelated variety with lower unemployment",Growth outcomes needed; WP38 is a single cross-section without growth outcomes,WP38 borrows the entropy decomposition (category/subcategory) as a descriptive diversity measure; cannot test growth effects5C04,"In Canada, distance to large metropolitan areas and the urban hierarchy still structure industrial location; peripheral regions (incl. Quebec's) show persistent decline",1,empirical,polese2004; polese2006; shearmur2007; shearmur2008; baldwin2008,Canadian census-division employment data 1971–2001; location quotients and growth regressions by distance class and urban size; high-order services concentrate in the urban hierarchy,Employment-based; pre-2001 data; no e-commerce dimension,WP38 provides a 2026 online-catalogue counterpart: does the metropolitan/peripheral gradient documented by Polèse-Shearmur also appear in online assortment breadth? (descriptive comparison only)6C05,"The number of distinct economic activities present in a city scales sub-linearly with population, and activities enter in a nested, predictable order",1,empirical,youn2016; hong2020; bettencourt2007; sinai2004,"NAICS business-establishment counts across US metropolitan areas (Youn et al.: diversity ~ log population; Hong et al.: nestedness/ordering of industries), urban scaling laws (Bettencourt), local online content scales with city size (Sinai-Waldfogel)",Cities not administrative regions; establishment counts not product listings; US data,"Closest precedent for H2: WP38 tests whether category diversity of online assortments scales with regional population with the same functional forms, and whether the region-category matrix is nested (ties to H7)"7C06,Larger markets sustain more varieties (love-of-variety); comparative advantage of cities is systematic in skill-intensity,1,theory+empirical,krugman1980; krugman1991; fujita1999; davis2020,Monopolistic-competition trade models; Davis-Dingel: US city-sector employment shares with population-size ordering (log-supermodularity test),Theory concerns production and trade; online catalogues mix local production and resale,Provides the motivation for an RCA-like index at the regional level; WP38 does not estimate structural parameters8C07,"The Balassa RCA index has a skewed, unstable distribution; symmetric/normalised variants and share-based alternatives mitigate this",2,empirical/methodological,balassa1965; hinloopen2001; yu2009; laursen2015,Trade data: distribution of RCA is right-skewed with unstable upper tail; NRCA and symmetric RCA proposed and compared,"Designed for export flows; WP38 applies the same algebra to listing counts (RAA), a different base with its own skewness (large merchants)",WP38 reports RAA alongside symmetric/normalised variants and merchant-weighted versions; explicitly distinguishes RAA from export RCA9C08,Regions/countries diversify preferentially into activities related to their existing portfolio (principle of relatedness),2,empirical,hidalgo2007; neffke2011; boschma2017; hidalgo2018; rigby2015; kogler2013; balland2019,"Country-product export panels (proximity from co-export), Swedish plant-level industry panels, US city patent classes: entry probability increases with relatedness density; mostly panel regressions with fixed effects",All rely on multi-period panels (decades); WP38 has essentially one snapshot (weak panel at best),H4 is exploratory: WP38 can compute a category relatedness matrix and relatedness density but can at most describe cross-sectional co-occurrence; any entry regression is flagged as exploratory and onboarding-confounded10C09,"Economic complexity indices (ECI/PCI, fitness) summarise the structure of a bipartite country-product matrix and correlate with income and growth",2,empirical/methodological,hidalgo2009; hausmann2011; hausmann2014; tacchella2012; cristelli2013; hidalgo2021; balland2022; hausmann2007,Method of reflections / eigenvector formulation on binary RCA matrices; nonlinear fitness-complexity iteration; cross-country growth regressions,Country-level export data with thousands of products and >100 countries; WP38 has 17 regions × 20 categories (very small matrix) and listings not exports,"H7 builds a small region-category bipartite network; given the dimension, WP38 must validate any eigenvector against null models and report instability (bootstrap by merchant) rather than a headline ECI"11C10,"ECI is mathematically a spectral-clustering/dimension-reduction of a similarity network and is sensitive to algorithm choice, thresholds and sample composition",2,methodological,mealy2019; morrison2017; sciarra2020; servedio2018,Mealy et al. show ECI = Fiedler vector of a normalised similarity matrix; Morrison et al. and Sciarra et al. show rankings differ substantially across ECI vs fitness and across variants; Servedio et al. document non-convergence of the fitness iteration,Critiques formulated for large trade matrices; small-matrix behaviour is less documented,"WP38 adopts the spectral interpretation (clustering of regions by assortment similarity) as the honest reading of its indicator, reports several variants and never interprets an arbitrary eigenvector as 'complexity' without validation"12C11,"Complexity/relatedness tools transfer to non-export bases: patents, industries, occupations, firms at city/regional level",2,empirical,balland2017; kogler2013; muneepeerakul2013; gao2018; davis2020,"US city patent data (knowledge complexity), occupation data for US metros, Chinese prefecture firm/industry data: regional complexity computed from location quotients on non-export counts","Bases are production-side (patents, employment); none uses retail listings or product catalogues","WP38's base (observed online product listings) is new in this literature; contribution is the data base and its validation, not a new method"13C12,Validating structure in bipartite matrices requires degree-preserving null models; nestedness and specialisation must be benchmarked against them,2,methodological,saracco2015; hausmann2011; ellison1997,Maximum-entropy bipartite configuration models for the world trade web; nestedness of country-product matrix compared with randomised matrices,Methods developed for trade networks; computation trivial at WP38's scale,"WP38 commits to merchant-count-preserving permutations and bipartite configuration-model nulls for RAA, nestedness and the H7 indicator"14C13,"Online channels shift demand toward niche products (long tail), but hits still capture a large share and the tail may be heavier in listings than in sales",3,empirical,brynjolfsson2003; brynjolfsson2011; anderson2006; elberse2008; zentner2013; tan2017; barisaac2012; waldfogel2017,"Book, apparel, video and music sales data (online vs catalogue/store), consumer-level panels; Gini/Pareto measures of sales concentration; theory of search costs and market structure","All concern sales/demand concentration; WP38 observes listing counts per merchant (supply-side catalogue breadth), not sales","H3 describes concentration of listings across merchants (rank-size, Gini, Pareto tails) and explicitly distinguishes catalogue-size concentration from sales concentration"15C14,E-commerce reallocates market share toward larger retailers/platforms; platform owners and large complementors dominate,3,empirical,goldmanis2010; zhu2018; einav2016,US county retail establishment counts vs internet penetration (Goldmanis et al.); Amazon product-level entry data (Zhu-Liu); survey of peer-to-peer markets,Market structure of retail industries measured by establishments or platform entries; no regional catalogue data,WP38 quantifies how much of each region's visible assortment comes from a few merchants (coverage/concentration component of QDACI) — descriptive16C15,Online purchasing is higher where local offline supply is thin and for 'preference minorities'; local tastes are heterogeneous across markets,3,empirical,forman2009; choi2011; quan2018; brynjolfsson2009,"Amazon sales rank by location vs store openings (Forman et al.), online purchase share by local preference minority status (Choi-Bell), footwear sales across US markets with demand estimation (Quan-Williams), cross-channel competition data (Brynjolfsson et al.)",Demand-side data with transactions; WP38 has no demand data at all,WP38 is the supply-side mirror: regional heterogeneity in what is offered online by local merchants. Cannot speak to demand heterogeneity17C16,"Distance and borders still reduce online trade (gravity holds online, with smaller distance elasticities)",3,empirical,hortacsu2009; blum2006; lendle2016,eBay/MercadoLibre transaction data with gravity models; digital-goods consumption across countries; eBay cross-border vs offline trade,Transaction flows between buyer and seller locations; WP38 observes only seller location,WP38 cannot estimate gravity; it can only describe the spatial distribution of sellers and listings18C17,"Gains from e-commerce are larger for consumers far from stores and in rural areas, and e-commerce changes relative prices across space",3,empirical,dolfen2023; couture2021; jo2019; lieber2012; goldfarb2019,"Visa card transactions (Dolfen et al.), RCT of e-commerce terminals in rural Chinese villages (Couture et al.), Japanese scanner/online price data (Jo et al.), surveys of digital economics",Causal designs use transactions and experiments; WP38 has neither,"WP38 documents the rural/urban distribution of online supply in Quebec (merchants and listings per capita) as context, without welfare claims"19C18,Online price dispersion for identical goods is persistent and shaped by platform search design and obfuscation,3+5,empirical,dinerstein2018; ellison2009; baye2004; brynjolfsson2000; clay2001; baye2006,"eBay search redesign (Dinerstein et al.), price-comparison site data with obfuscation (Ellison-Ellison), Shopper.com listings (Baye et al.), online books/CDs (Brynjolfsson-Smith; Clay et al.)",Single-platform or price-comparison settings; WP38 covers independent stores across many platforms with no search-design variation,"H5 measures cross-merchant dispersion of matched products with product fixed effects; dispersion is interpreted descriptively (region/merchant correlates), not causally"20C19,"Internet diffusion and its economic benefits have been geographically uneven: urban leadership in adoption, benefits concentrated in already advantaged places; broadband effects on rural growth are modest/ambiguous",4,empirical,forman2005; forman2012; forman2018; sinai2004; kolko2012; whitacre2014; haller2015; haller2019; goldfarb2008; haight2014,"US county/establishment adoption surveys, county wage growth 1995–2000, broadband expansion and employment (US, Irish firm panels), Canadian Internet Use Survey microdata (Haight et al.)","Mostly US/Irish; Canadian evidence is on household usage (Haight et al.), not merchant e-commerce","WP38 adds a Quebec regional map of online merchant presence (Fabri-Ka coverage) and checks it against CIUS/SDTIU-type adoption patterns; selection of merchants into the aggregator is a confounder, flagged"21C20,"Official Canadian/Quebec statistics measure e-commerce adoption by households and enterprises, with provincial and urban/rural breakdowns",4+7,institutional,statcan_cius; statcan2023cius; statcan_sdtiu; oecd2019; atn_netendances,Survey programmes (CIUS household survey; SDTIU enterprise survey; NETendances Quebec panel); OECD review of e-commerce measurement and business models,Surveys do not identify individual merchants or catalogues; regional detail below province is limited,"WP38 uses these as external benchmarks for coverage/representativeness, never as substitutes for merchant-level data"22C21,"Scraped online prices are a valid but selected source: online and offline prices are similar in most retailers; scraped data have specific properties (stockouts, missing variants, frequency)",5+6,empirical,cavallo2017; cavallo2018; cavallo2016; gorodnichenko2017; gorodnichenko2018; cavallo2023; konny2022,Billion Prices Project multi-country scraped panels vs in-store checks; price-comparison platform micro data; BLS experience with web-scraped and alternative data in the CPI; stockouts in online data,"Price-focused; WP38's primary object is listings/categories, with prices only in H5","WP38 documents the analogous properties of a catalogue aggregator (duplicates, variants, missing prices, availability) in its data-quality report and treats the snapshot as a non-probability sample"23C22,Most price dispersion is relative (store×good) and persistent; chains price uniformly across locations; spatial price gaps grow with distance,5,empirical,kaplan2015; kaplan2019; dellavigna2019; cavallo2014; atkin2015; handbury2015; handbury2021,"Nielsen Homescan/retail scanner data (variance decompositions, uniform pricing across stores of the same chain), online prices across currency unions, barcode-level prices across Indian/US locations, city price indices with variety","Transaction or in-store posted prices; WP38 has posted online prices from independent merchants, where merchant and region effects are nearly collinear","H5 decomposes dispersion into product, merchant and region components, acknowledging that region effects are identified only from merchants of the same region selling the same matched product"24C23,Hedonic regressions with product characteristics or product fixed effects isolate quality-adjusted price differences,5,theory+empirical,rosen1974; pakes2003; bajari2005; jaravel2019,Hedonic equilibrium theory; PC price indices with characteristic drift; hedonic demand with unobserved characteristics; barcode-level product entry data,"Characteristics in WP38 are sparse (title, category, brand); fixed effects require matched products across merchants",WP38 relies on matched-product fixed effects (not characteristics regressions) and reports match quality thresholds as a sensitivity dimension25C24,Probabilistic record linkage provides a principled framework for matching products across sources,5,methodological,fellegi1969; christen2012,"Fellegi-Sunter decision rule on agreement patterns; survey of blocking, comparison functions and evaluation for entity resolution",Generic; product titles across merchants are noisier than administrative records,WP38 applies blocking (brand/category) + string similarity with manual validation of a labelled sample and reports precision/recall26C25,"Non-probability 'big data' samples can be badly biased: the data-defect correlation, not sample size, governs error; valid inference requires modelling selection",6,theory+empirical,meng2018; bradley2021; elliott2017; buelens2018; daas2015; beaumont2020; rao2021; lazer2014; groves2006; little2019; heckman1979,"Formal decomposition of estimation error (Meng), application to vaccine-uptake surveys (Bradley et al.), quasi-randomisation and superpopulation approaches (Elliott-Valliant; Buelens et al.), official-statistics perspective (Daas; Beaumont; Rao), classic selection models",Frameworks assume a target population and some auxiliary data; WP38's target population (all Quebec online merchants/listings) is itself not enumerated,"WP38 treats Fabri-Ka as a non-probability sample, audits coverage against Business Counts and CIUS/SDTIU, reports merchant-weighted vs listing-weighted results and does not generalise to 'Quebec production'"27C26,Measurement error in regressors/outcomes and the choice of spatial units (MAUP) can change estimates substantially,6,methodological,hausman2001; bound2001; briant2010,Attenuation and non-classical error analysis; French employment data re-aggregated into alternative zonings show modest but non-trivial sensitivity for agglomeration regressions,WP38's 17 regions are fixed by administration; alternative zonings (MRC) possible only if merchant addresses allow,"WP38 reports region-assignment uncertainty (geocoding quality) and, where feasible, MRC-level sensitivity"28C27,Canadian Business Counts come from the Business Register (payroll/$30k revenue/incorporation rules; counted by location; not a time series); Quebec origin certifications set explicit cost/transformation thresholds,7,institutional,statcan2026cbc; statcan_brdef; statcan_br; isq2024; quebec_regions; quebec_aliments; alimentsqc; produitsqc,Official methodological pages and releases; Les Produits du Québec: 'Produit du Québec' ≥85% of direct production costs in Quebec + final substantial transformation; 'Fabriqué au Québec' = final substantial transformation; 'Conçu au Québec' = design workforce in Quebec; Aliments du Québec marks for food,Business Counts include non-online firms and exclude many micro-sellers; certification coverage among Fabri-Ka merchants unknown,WP38 uses Business Counts as the denominator for coverage ratios and treats certification marks as the only admissible origin evidence for H6 (otherwise H6 is marked untestable)29