# THE WHOLE THING — Story + Evidence in One Map *Each scholar = plain-English block (did / measured / found / catch / why it matters), then THEIR PAPERS below — click each paper for question, method, caveats & fixes. Side shown in [brackets].* ## 1. THE ARGUMENT — what my thesis says *The pitch → the problem → the 5 defended claims → the stakes* ### THE PITCH IN ONE SENTENCE ### THE PROBLEM — the gap nobody filled ### MY 5 CLAIMS — each with plan → risk → fix → proof *Each claim below opens into: what I will actually DO → what could go WRONG → how I HEDGE → which paper backs me* #### CLAIM 1 — "Judges hold a yardstick" (the rules in judges’ heads) - **What I do:** define the judges' yardstick before collecting data — write down what a "proper venture" looks like in France/Italy (the reference framework) - **What could go wrong:** the yardstick differs across contexts → France ≠ Italy - **My hedge:** treat context as a formal moderator, not a pooled blur - **Backing:** Ghods E&RD 2024 — French vs international founders hold DIFFERENT images of legitimacy; Schoon 2022 — expectations live in the dyad #### CLAIM 2 — "My module teaches the right signals" (the founder’s side) - **What I do:** the 12h module trains the 5 capabilities — Cognitive (fit a known category) · Pragmatic (show value) · Social (get vouched for) · Factual (show milestones) · Regulative (show compliance) - **What could go wrong:** a trained signal could backfire — the audience might find a strategy improper - **My hedge:** teach consistency + audience-appropriateness, not tricks — one contradiction stalls the judge - **Backing:** Siraz JMS online first (categorical fit = strongest signal) · EMR 2026 (inconsistent bundles stall) · Claes EMR 2025 (strategies can backfire) #### CLAIM 3 — "Judges respond to these signals" (the verdict) - **What I do:** nothing new to build — Siraz already measured judges' verdicts on exactly these cues, thousands of ratings - **What could go wrong:** her judges rated fake profiles, not real founders - **My hedge:** my design adds real behavior + real milestones on top of her lab evidence — lab shows the mechanism, field shows it in life - **Backing:** 4 Siraz experiments (1,866 + 3,952 + 992 + 12,496 ratings) all measured the assent side #### CLAIM 4 — "Nobody connected the two sides" (the gap) - **What I do:** the systematic claim — every scholar measured ONE side of the dyad; nobody trained conformity and measured assent - **What could go wrong:** a committee member names a paper that did it - **My hedge:** the connection was tested in vignettes/labs (Pollack, Nagy) but never in a field RCT on founders — my claim is precise: first FIELD test - **Backing:** Überbacher 2014 review (the field's own map shows the gap) + the Scholarly Network section shows every scholar on one dyad component #### CLAIM 5 — "I run the experiment" (my test) - **What I do:** randomize founders → 12h module to half → measure self-legitimacy (survey T0/T1/T4) + behavior (milestones, pitch-deck text scores) + explain it (interviews) - **What could go wrong:** dropouts, contamination, self-report inflation - **My hedge:** ITT + leaver/stayer + imputation · cohort separation + staggered rollout · objective milestones + Ghods text scores - **Backing:** Campos Science 2017 — a 3-arm founder RCT of exactly this scale is publishable in Science; McKenzie 2021 — sized to detect the effect the literature predicts #### THE CHAIN IN ONE LINE - **Expectations** (judges' yardstick — Siraz/Ghods mapped it) → **Conformity** (founder signals — MY module trains them) → **Assent** (judges' verdict — Siraz measured it) → legitimacy conferred - **The sentence for committee:** "Everyone studied one link of this chain. I test whether pulling on the middle link — training the founder — actually moves the last link — the judge's verdict." ### WHY IT MATTERS - If training works → a cheap 12-hour fix for the legitimacy gap faced by unconventional founders - If it doesn't → proves legitimacy lives outside the founder's reach → policy should target judges/platforms, not founders - Either way you learn something the field doesn't know ## 2. SCHOON — the rulebook (what legitimacy IS) - **What he did:** pure theory — read 50 years of research and asked "what IS legitimacy, exactly?" - **What he measured:** nothing — no people, no data. He built the definition everyone else must use. - **What he found:** legitimacy is never inside a person or a firm — it's the RELATIONSHIP between venture and judge: the judge's expectations + the venture's signals + the judge's approval - **The catch:** it's a definition, not a measurement → you still have to build the survey yourself - **Why it matters to you:** it's the map of your whole thesis — training improves the signals (founder side), judges deliver the verdict (evaluator side) - **His role:** every scholar below studies ONE part of his dyad — Siraz measures the judge’s verdict (assent), Bertrand measures the founder’s side (conformity), Ghods scores the whole dyad on real firms, Aldebert studies the helpers around it ### "Legitimacy lives in the RELATIONSHIP" 💡 THEORY - **Paper:** Schoon — American Sociological Review 2022 - *The question:* Everyone measured legitimacy differently — legitimacy vs reputation vs prestige blurred; no agreed operational definition existed - *The 3 variables in plain English:* - EXPECTATIONS = the judge's mental yardstick ("what should a real business look like?") - CONFORMITY = the venture's signals matching that yardstick (your 5 capabilities live here) - ASSENT = the judge's verdict ("yes, appropriate") — remove any one and legitimacy fails - *Sample:* none — pure theory synthesis of 50+ years (Weber, Suchman, Tost) - *Caveats → solutions:* - *Problem: too abstract to measure* — a "relation" can't be handed to a survey firm → *Fix:* defined the 3 conditions as separately measurable - *For you:* your umbrella — ATENA trains Conformity; Siraz measures Assent; Expectations come from context ### THE CLASSICS — the famous definitions - **One line:** Suchman = the definition · Zimmerman & Zeitz = legitimacy unlocks resources · Überbacher = the field's map #### "Legitimacy = appropriate within the system" 💡 THEORY - **Paper:** Suchman — Academy of Management Review 1995 - *The question:* Everyone used the word "legitimacy" loosely — what does it actually mean? - *What they did:* Unified strategic + institutional views; defined 3 types: pragmatic (is it useful to me?), moral (is it the right thing?), cognitive (does it look normal?) - *Caveats → solutions:* - *Problem: definition so broad it risked meaning anything* → *Fix:* the 3 types became the standard typology everyone operationalizes - *For you:* the definition on Slide 3 — everything else is a footnote to this #### "Legitimacy unlocks other resources" 💡 THEORY - **Paper:** Zimmerman & Zeitz — Academy of Management Review 2002 - *The question:* Why do new ventures die (the "liability of newness")? - *What they did:* Proposed legitimacy as the gateway to finance, customers, talent; 4 paths: conform, select, manipulate, create - *Caveats → solutions:* - *Problem: untested* → *Fix:* became the standard citation for legitimacy→resources→growth; confirmed empirically later (Delmar & Shane 2004) - *For you:* your core mechanism — training = the "conform" path founders can be taught #### "Map of the whole field" 📚 REVIEW - **Paper:** Überbacher — Journal of Management Studies 2014 - *The question:* New-venture legitimation research was scattered across journals — no review existed - *What they did:* Systematic review + research programme — organized legitimation into types, mechanisms, contexts - *For you:* the review your lit review extends — Siraz/Aldebert fill its evaluator-cognition gap ### GHODS — legitimacy pays, and it can be measured - **What he did:** tracked 67 French biotech firms for 8 years; also showed legitimacy's meaning varies by context and can be auto-scored from texts - **What he measured:** firm performance + a legitimacy score — does legitimacy make international expansion profitable? - **What he found:** yes — expansion only pays when legitimacy is high; context changes what "legitimate" means; text scores predict survival - **The catch:** small niche sample (67 biotech firms) — but replicated across 3 different methods - **Why it matters to you:** atEnA is cross-border France–Italy → Ghods proves legitimacy decides whether crossing borders works, and gives you an objective scoring tool #### "Legitimacy makes going abroad pay off" ✅ POSITIVE - **Paper:** Ghods — PhD thesis, Aix-Marseille University 2021 (supervised by Aldebert & Ricard; Bitektine on the jury) - *The question:* Expanding abroad is risky ("liability of foreignness") — does legitimacy decide whether it pays off? - *What they did:* Tracked 67 French biotech SMEs over 8 years using "fixed-effects panel" — statistics that compare each firm to its OWN past, so stable firm traits can't fake the effect - *How legitimacy was scored:* a composite index built from observable proxies — certifications, partnerships, media coverage — combined into one legitimacy score per firm per year, then tested against internationalization and performance - *Sample:* 412 firm-year observations · France · 2010–18 - *Found:* legitimacy moderates — internationalization only helps performance when legitimacy is high - *Caveats → solutions:* - *Problem: small niche sample* — 67 firms, biotech only → *Fix:* 412 observations over time + fixed effects absorb unobserved firm differences - *For you:* atEnA is cross-border France–Italy — legitimacy is what makes border-crossing pay off #### "What 'legitimate' means changes by country" ✅ POSITIVE - **Paper:** Ghods, Ricard & Aldebert — E&RD 2024 - *The question:* Do founders in different contexts picture legitimacy differently? Everyone assumed universal — never tested - *What they did:* 150 French entrepreneurs wrote the first words that come to mind for "legitimate entrepreneur" + ranked importance; a matrix (how often × how important) separates core beliefs from peripheral ones - *Sample:* 75 domestic vs 75 international entrepreneurs · France · 2019–21 - *Found:* domestic and international founders hold different mental models of legitimacy - *Caveats → solutions:* - *Problem: grouping free words is subjective* — different researchers could group differently → *Fix:* lemmatization (grouping same-meaning words by rule, not opinion) + two independent coders agreeing (kappa > .85 = strong agreement) - *For you:* France vs Italy must be a formal moderator, not pooled data #### "Legitimacy can be scored from texts" ✅ POSITIVE — predicts failure - **Paper:** Ghods, Ricard, Aldebert & Schoon — British Journal of Management 2026 - *The question:* Legitimacy was only measurable via surveys — can an algorithm score it from venture texts at scale? - *What they did:* NLP (a language algorithm) scored 524 venture press releases against legitimacy dictionaries; checked whether scores predict real 5-year survival - *Found:* yes — text scores predict which ventures die - *Caveats → solutions:* - *Problem: the algorithm may misread promotional language* → *Fix:* human experts calibrated the dictionary + scores validated against actual firm death - *For you:* objective check on participant pitch decks — kills the "it's all self-report" critique ### ⤷ WHO SITS WHERE — every scholar on Schoon’s map *Uniform 4-line structure: 1. Research Question → 2. Core Finding / Verdict → 3. Methodology & Epistemic Base → 4. How You Use It in Thesis* #### 0. THE 3 PIECES — the yardstick · the signals · the verdict *Schoon's 3 dyad variables organize the entire field: every scholar studies ONE side of the relationship* - **EXPECTATIONS (judge side — the yardstick, upstream):** What evaluators believe a proper venture should look like - → Suchman 1995 (the norms/values basis of the yardstick) - → Siraz EMR 2026 (evaluator values filter cues) - → Ghods E&RD 2024 (yardstick varies by context: domestic vs international) - → Ghods 2021 PhD (yardstick in foreign markets) - **CONFORMITY (founder side — the signals = YOUR 5 CAPABILITIES):** What the venture emits to match the yardstick - → Cognitive (positioning & category framing) · Pragmatic (value demonstration) · Social (networks & vouching) · Factual (milestones & credentials) · Regulative (compliance beyond minimums) - → Suddaby & Vaara (legitimation strategies founders deploy) - → Fisher 2020 (new venture signaling repertoire) - → Aldebert (accelerators teach/boost conformity signals) - **ASSENT (judge side — the verdict, downstream):** The evaluator's approval judgment itself - → Tost 2011 (propriety = individual assent; validity = collective assent) - → Bitektine & Haack 2015 (how individual assent scales to institutional validity) - → Siraz JMS online first / JBE 2026 / AOM 2020 / EMR 2025 (propriety measured via conjoint) - → Bertrand HAL 2025 (self-assent = the founder's internal mirror of the dyad) - **THE CHAIN:** EXPECTATIONS set by context → CONFORMITY trained by ATENA → ASSENT measured by Siraz → legitimacy conferred #### 1. THE CLASSICS — what a "proper" venture looks like ##### Mark Suchman 1995 — "legitimacy = appropriate within a social system" - **1. Question:** What is organizational legitimacy, and what distinct forms does it take when managed strategically? - **2. Finding / Verdict:** 💡 THEORETICAL FOUNDATION (~22,000 citations) — Defined legitimacy as a generalized perception of appropriateness; established the Pragmatic, Moral, and Cognitive tripartite typology - **3. Methodology:** Deductive conceptual synthesis bridging organizational institutionalism and strategic management - **4. How you use it:** Foundational definition for Slide 3; directly maps onto the 5 capabilities in your 12h training module ##### Michael Tost 2011 — one judge’s verdict vs the crowd’s verdict - **1. Question:** How do individuals actually form legitimacy judgments, and how does micro-judgment relate to macro-validity? - **2. Finding / Verdict:** 💡 THEORETICAL MECHANISM — Formulated the critical distinction between **Propriety** (individual evaluator belief) and **Validity** (collective social fact) - **3. Methodology:** Dual-process cognitive psychology synthesis (System 1 automatic vs System 2 deliberative evaluation) - **4. How you use it:** Provides the precise conceptual mechanism explaining how individual evaluator judgments aggregate into collective venture standing ##### Bitektine & Haack 2015 — how one verdict becomes everyone’s verdict - **1. Question:** Through what multi-level process do individual propriety judgments scale up into institutionalized validity? - **2. Finding / Verdict:** 💡 MULTILEVEL MODEL — Demonstrated that macro-validity acts as a cognitive heuristic that suppresses active judgment, while institutional disruption re-activates active propriety evaluation - **3. Methodology:** Multilevel conceptual integration linking micro-evaluator cognitive processing to macro-institutionalization - **4. How you use it:** Theoretical basis for showing why novice founders without validity cues must undergo active, high-scrutiny evaluator judgments ##### Suddaby & Vaara — the words & stories that build legitimacy - **1. Question:** What linguistic, rhetorical, and narrative strategies do actors employ to construct, maintain, or repair legitimacy? - **2. Finding / Verdict:** ✅ THEORETICAL & EMPIRICAL FOUNDATION — Identified Authorization, Rationalization, Moralization, and Mythopoesis as core discursive legitimation repertories - **3. Methodology:** Critical discourse analysis and rhetorical analysis of institutional transformation across diverse corporate fields - **4. How you use it:** Informs the communication and storytelling capability taught in the 12h ATENA entrepreneurship module #### 2. THE JUDGES — how they form the verdict (Siraz’s team) ##### Eero Vaara — co-author of the 5-states paper - **1. Question:** How do discursive legitimation strategies map onto the 5 distinct states in the grey zone? - **2. Finding / Verdict:** 💡 PROPOSED TYPOLOGY — Co-authored the foundational 5-state framework, contributing the state-specific discursive strategy matrices (Tables II–VI) - **3. Methodology:** Conceptual rhetorical synthesis - **4. How you use it:** Citing Vaara connects your discursive training module directly to the Siraz 2023 theoretical engine ##### Anne-Sophie Claes — ran the foie-gras & cue experiments - **1. Question:** How do evaluators process cue bundles and legitimation strategies in contested industries (foie gras, fracking)? - **2. Finding / Verdict:** ✅ CONFIRMED EMPIRICALLY — Proved that cue incongruence produces grey-area stall and improper legitimation strategies backfire - **3. Methodology:** Conjoint and factorial experimental designs analyzing evaluator decisions - **4. How you use it:** Direct evidence that training must ensure signal coherence to avoid backfire effects #### 3. THE HELPERS & THE MEASURERS — Aldebert, Ghods, Bertrand, Schoon ##### Daisy Bertrand — built the self-legitimacy scale I use - **1. Question:** How do nascent entrepreneurs perceive their own internal legitimacy, and how does illegitimacy block action? - **2. Finding / Verdict:** ✅ CONFIRMED (Exploratory Scale Validation) — Validated the 4-dimensional **Self-Legitimacy Scale** linking low self-legitimacy to impostor syndrome - **3. Methodology:** Psychometric scale validation, student entrepreneur surveys, and structural equation modeling - **4. How you use it:** You use Bertrand's exact scale in your 44-item ATENA instrument as your central psychological mediator ##### Antonin Ricard — legitimacy chair, Ghods’ co-supervisor - **1. Question:** How does entrepreneurial legitimacy affect international new venture performance and social representations? - **2. Finding / Verdict:** ✅ CONFIRMED — Co-directed Ghods' PhD; proved legitimacy moderates internationalization performance and varies across domestic vs foreign contexts - **3. Methodology:** Quantitative cognitive representations and longitudinal panel econometrics - **4. How you use it:** Rationale for testing cross-border regional context as a formal moderator in your Franco-Italian sample ##### Ali Ghods — proved legitimacy pays across borders - **1. Question:** How does entrepreneurial legitimacy help startups when they expand into foreign (cross-border) markets? - **2. Finding / Verdict:** ✅ CONFIRMED (p < 0.05) — Going international is risky; startups only convert cross-border expansion into high performance if they possess strong legitimacy - **3. Methodology:** Longitudinal panel econometrics (412 observations across 67 biotech firms) + scale construction; co-directed by Aldebert & Ricard; Alex Bitektine served on his jury - **4. How you use it:** - *Franco-Italian Context:* Proves why atEnA's cross-border setting demands legitimacy training. - *Empirical Arc:* Ghods established the measurement pipeline (reconceptualize → measure → test performance); your thesis injects the missing **RCT educational intervention** into that pipeline. ##### Eric Schoon — the umbrella framework everyone sits under - **1. Question:** What are the necessary and sufficient relational conditions required to operationalize and measure legitimacy? - **2. Finding / Verdict:** 💡 PROPOSED (Adopted Standard) — Proved legitimacy exists strictly in the dyad (object + audience + relationship); co-authored 2026 BJM paper with Ghods & Aldebert - **3. Methodology:** Relational sociological theory and comparative-historical analysis - **4. How you use it:** Schoon sits ONE LEVEL ABOVE the empirical papers — his 3 variables are the umbrella under which every other study (and your 5 capabilities) is located - *EXPECTATIONS — judge side, upstream:* The evaluator's yardstick = Siraz's "reference framework" / NVBs → studied by Siraz EMR 2026 (evaluator values), Ghods E&RD 2024 (context variation France vs international) - *CONFORMITY — founder side = YOUR 5 CAPABILITIES:* The signals the venture emits to match the yardstick — exactly what your 12h module trains: - Cognitive → positioning & category framing - Pragmatic → value demonstration & client utility - Social → networks, ties & vouching - Factual → milestones, credentials & track record - Regulative → compliance beyond legal minimums - *ASSENT — judge side, downstream:* The evaluator's approval judgment = Siraz's propriety ratings → measured in her conjoint experiments (JMS online first, JBE 2026) - *The chain:* EXPECTATIONS (context) → CONFORMITY (your training's target) → ASSENT (Siraz's measurement) → legitimacy conferred #### 4. THE GAP — where my thesis sits - **1. The Evaluator Stream (Siraz):** Tested evaluator cognition and cue processing in experimental settings - **2. The Ecosystem Stream (Aldebert):** Tested institutional transfer and support structure mechanics - **3. The Measurement Stream (Schoon & Ghods):** Operationalized the relational dyad and computational text metrics - **4. YOUR ATENA THESIS:** Executes the missing **Field RCT** — testing whether an educational intervention on the founder actually alters the legitimacy judgment conferred by the evaluator ## 3. SIRAZ — the judges (who decides if you look real) - **What she did:** showed ordinary people fake startup profiles (like fake job applications) and asked "how legitimate does this look?" - **What she measured:** the legitimacy rating each judge gave each profile — one number per profile, thousands of ratings total - **What she found:** looking like a NORMAL, recognizable business beats every fancy signal (endorsements, certifications). And judges punish women / immigrant / minority founders harder after identical failures. - **The catch:** judges rated fake profiles — nobody risked real money → it measures what people SAY, not what they'd DO - **Why it matters to you:** she proved WHICH signals judges care about → your training teaches exactly those signals ### "Looking normal beats endorsements" ✅ POSITIVE — ×2, n=1,866 - **Paper:** Thaler, Sievert, Siraz, Claes & Pinz — Journal of Management Studies (online first) - *The question:* When a judge sees a new venture, which signal convinces them most — looking like a normal business, having official authorization, or having endorsements? - *What they did:* Showed judges made-up venture profiles where each signal was switched on/off - *What "normal business" consisted of (the 3 signals tested):* - CATEGORICAL FIT = description matches a known business type (a bakery selling bread) vs a hybrid that fits nothing — THIS is "looking normal" - AUTHORIZATION = official license/permit — the state says it's real - ENDORSEMENT = a respected org publicly vouches for it - *What they measured:* the legitimacy score (propriety rating) each judge gave each profile — a 1–7 scale of "is this appropriate / does this look like a real business" - *Sample:* 1,866 judges across 2 separate experiments · online panels · Germany/UK · 2022–23 - *Found:* "Fitting the category of a real business" dominated — endorsements added almost nothing on top - *Caveats → solutions:* - *Problem:* fake profiles ≠ real decisions → *Fix:* realistic pitch summaries + attention checks - *Problem:* could be a fluke → *Fix:* ran twice, same result - *For you:* teach founders to LOOK like a recognizable business first — your Cognitive capability is the strongest lever ### "Mixed signals stall judges" ✅ CONFIRMED - **Paper:** Siraz, Claes & De Castro — European Management Review 2026 - *The question:* What happens when a venture sends good AND bad signals at the same time? - *What they did:* Each judge rated 16 made-up venture profiles mixing positive and negative cues. (This design is called a "conjoint" — you show many combinations and see which signals move the rating.) - *4 cue types mixed:* REGULATORY (permits) · PRAGMATIC (revenue, clients) · NORMATIVE (values, CSR) · COGNITIVE (familiar category) — each + or −, e.g. "licensed but losing clients" - *Sample:* 247 judges × 16 profiles each = 3,952 ratings · France/Spain · 2021–23 · topic: fracking - *Found:* contradictory cues cancel each other → judgment stalls in the grey zone; judges' own green values made them discount pro-industry signals - *Caveats → solutions:* - *Problem:* judges get tired → *Fix:* randomized profiles, only 16 each - *Problem:* judges' own beliefs interfere → *Fix:* measured their values first, checked if green judges rated differently (a "moderator") - *For you:* teach consistent signal bundles — one contradiction stalls the whole judgment ### "Defending a hated product backfires" ❌ NEGATIVE - **Paper:** Claes, Siraz, De Castro & Lapeyre — European Management Review 2025 - *The question:* Producers believe their justifications work — but do audiences agree? - *What they did:* First interviewed 12 foie gras producers to collect the 5 arguments the industry actually uses; then had judges rate 32 made-up profiles built from those arguments - *5 real defenses tested:* "it's tradition" · "it creates jobs" · "animals are well treated" — judges rated the STRATEGY, not the product - *Sample:* 31 judges × 32 profiles = 992 ratings · France · 2019–22 - *Found:* strategies the industry loves were judged inappropriate → they actively HURT legitimacy - *Caveats → solutions:* - *Problem:* only 31 judges → *Fix:* 992 total ratings + real industry interviews - *Problem:* niche industry → *Fix:* triangulated with fieldwork - *For you:* never assume a trained signal works — the audience decides if it's appropriate; you must test it ### "Women & immigrants punished harder after failure" ✅ CONFIRMED — bias - **Paper:** Siraz et al. — Academy of Management / BCERC 2020 - *The question:* Same failure, different founder — does the judge's verdict change with gender or origin? - *What they did:* Judges rated failed-venture profiles where founder gender, immigrant status, education, and failure cause were shuffled randomly - *4 things varied:* GENDER · IMMIGRANT status · EDUCATION · WHY it failed — everything else identical → any difference = founder bias - *Sample:* 74 judges × 32 profiles = 2,368 ratings · Europe · 2018–19 - *Found:* women and immigrants penalized harder for identical failures; education helped them less - *Caveats → solutions:* - *Problem:* judges hide bias → *Fix:* profiles vary on 4 dimensions at once, so nobody guesses the topic - *For you:* run subgroup analyses in ATENA — does training help minority founders equally? ### "Same finances, lower score for Latinx name" ✅ CONFIRMED — bias - **Paper:** Siraz, Claes & Alvarez — Journal of Business Ethics 2026 - *The question:* Is a Latinx founder judged worse just because of the surname — even with identical finances? - *What they did:* Same conjoint trick — surname varied (Latinx vs Anglo) while debt, revenue, and track record stayed identical - *Varied per profile:* SURNAME (Latinx/Anglo) · DEBT · REVENUE · TRACK RECORD — only the name changed - *Sample:* 781 US judges × 16 profiles = 12,496 ratings · 2021–24 · half the judges were Latinx themselves - *Found:* Latinx founders scored lower even at equal/lower debt — and Latinx judges were NOT more lenient → the bias is in the system - *Caveats → solutions:* - *Problem:* surname might signal risk → *Fix:* finances literally identical, only the name changed - *For you:* your target population faces structural deficits — the motivation for training ### "Legitimacy has 5 states, not yes/no" 💡 THEORY - **Paper:** Siraz, Claes, De Castro & Vaara — Journal of Management Studies 2023 - *The question:* Legitimacy was always treated as binary (legit or illegit) — what about the grey zone in between? - *What they did:* Read 50+ years of theory and built a 5-state map + the strategies that work per state - *Sample:* none — it's a theory paper - *Proposed:* 5 states (legitimate · conditional · unknown · conditional-illegitimate · illegitimate); founders can move between them - *Caveats → solutions:* - *Problem: no data — pure theory* → high severity for empiricists → *Fix:* anchored every state to real-world cases (GMOs, medical cannabis, aviation) - *For you:* novice founders start in "Unknown–Undecided" — training is the mechanism to move them up ### ⤷ THE 5 STATES — explained in plain words *Siraz, Claes, De Castro & Vaara (2023), Journal of Management Studies* #### 1. LEGITIMACY — "Everyone just accepts it" ✅ - **In plain words:** Nobody even questions it. It looks so normal that judges don't stop to evaluate it. - **Everyday example:** A bakery opening on your street. Nobody asks "is this a real business?" — it just IS one. - **Technical:** favorable judgments · strong congruence with norms · passive (taken-for-granted) processing - **Strategies:** Framing, Mythopoesis, Normalization — keep the buffer, don't draw attention - **Warning:** Defending a fully-legitimate venture too loudly can BACKFIRE — it makes people look for problems #### 2. CONDITIONAL LEGITIMACY — "Legitimate IF you follow the rules" ⚠️ - **In plain words:** "Yes, BUT..." — judges accept you only if you meet specific conditions. - **Everyday example:** GMO food is accepted ONLY with labeling. A drone startup is fine ONLY with flight permits. Break the condition → acceptance collapses. - **Technical:** acceptance conditional on compliance · opposition is ACTIVE, ready to pounce if you breach · evaluators scrutinize you - **Strategies:** Authorization, Rationalization, Self-Regulation, Commitment Narratives #### 3. UNKNOWN LEGITIMACY — "The judge can't decide" ❓ ← WHERE YOUR FOUNDERS START *The judge has no stable verdict — but WHY can't they decide? Three different reasons = three sub-types:* ##### A. UNDETERMINED — "Nobody is even looking" - **In plain words:** The venture is invisible. Judges haven't thought about it at all — not because they dislike it, they just never noticed. - **Everyday example:** A founder working from their bedroom that nobody has heard of. No opinion exists yet — good or bad. - **The problem to fix:** ZERO ATTENTION. You need people to notice you exist. - **Strategy:** emotional framing, high-visibility storytelling → build baseline salience (get on the radar) ##### B. UNDECIDED — "They look, but can't make up their mind" ← YOUR ATENA FOUNDERS - **In plain words:** Judges DO look, but the picture is blurry — missing information, mixed signals, or no clear category. They literally cannot judge yet. - **Everyday example:** A woman who left a marketing job to launch a "wellness-tech" venture — is it a health company? a tech company? a coaching service? The judge sees it but can't place it. - **The problem to fix:** CONFUSION, not invisibility. The judge needs a recognizable frame. - **Strategy:** analogical reasoning — "it's like X but for Y" (frame the unknown as familiar) · role models · authorized intermediaries (a trusted program vouching) - **Direct connection to you:** ATENA's 12h module IS the analogical-framing mechanism — it teaches founders to give judges a familiar category to latch onto ##### C. POLARIZED — "Half love it, half hate it" - **In plain words:** The audience itself is split into two camps with opposite verdicts, roughly balanced. - **Everyday example:** Crypto startups — some judges see "the future of finance," others see "a scam." Same venture, two stable opposite judgments. - **The problem to fix:** A DIVIDED audience — you can't win everyone with one story. - **Strategy:** bridging narratives, dual-framing — different framing for each audience segment #### 4. CONDITIONAL ILLEGITIMACY — "Illegitimate, but tolerated anyway" ⚠️ - **In plain words:** Judges think it's improper IN PRINCIPLE — but allow it anyway because it serves a necessary purpose. - **Everyday example:** Medical cannabis. Morally contested, but tolerated because patients need it. Or a payday lender — distasteful, but legal and filling a gap. - **Technical:** fundamentally incongruent BUT saved by pragmatic necessity or a regulatory carve-out · opposition actively fights the exception - **Strategies:** Authorized Exception (formal dispensation), Higher-Good Moralization ("yes it's ugly, but it serves a greater good") #### 5. ILLEGITIMACY — "Actively rejected" ❌ - **In plain words:** Judges have decided: this should not exist. It's rejected, sanctioned, or stigmatized. - **Everyday example:** A scam "investment scheme" — not just distrusted, but shut down and condemned. - **Technical:** unfavorable, sticky judgments · severe incongruence with societal norms · rejection is ritualized/passive — it no longer needs justification - **Strategies:** impression management, decoupling, normalization accounts, or relocate to a favorable jurisdiction #### HOW STATES FLIP (when the rules change) - **Trigger:** Venture enters a new environment or market - **Determinants:** Stability of Reference Framework × Presence of Type Conflicts × Strength of Societal NVBs - **Key Insight:** A venture's legitimacy state can flip from legitimate to contested without any change in its own behavior, solely because the societal yardstick shifted ## 4. FOUNDERS — inside the founder’s head - **What she did:** built the first "self-legitimacy" questionnaire and surveyed 122 students/founders at Aix-Marseille - **What she measured:** how legitimate founders feel INSIDE — "do I have the right to call myself an entrepreneur?" - **What she found:** founders who feel illegitimate get impostor syndrome → they stop acting - **The catch:** measured once, at one moment → shows correlation, can't prove cause - **Why it matters to you:** her scale is your psychological mediator — the thing inside the founder that your training should move first ### "Feeling like a fraud freezes founders" ⚠️ POSITIVE — weak test - **Paper:** Bertrand, Benamara & Aldebert — 2025 (HAL working paper) - *The question:* Does an internal feeling of illegitimacy stop a founder from acting? Self-legitimacy had never been measured - *What they did:* Surveyed students and student-entrepreneurs with Bertrand's new self-legitimacy scale; clustered them into 3 profiles - *What the scale asks:* items like "I have the right to call myself an entrepreneur" / "people like me can legitimately run a venture" — the founder's own feeling of entitlement, not skill - *Sample:* 122 respondents · Aix-Marseille University · 2023–24 - *Found:* low self-legitimacy → impostor syndrome → lower entrepreneurial intention - *Caveats → solutions:* - *Problem: measured once, at one moment* ("cross-sectional") — can't prove cause → *Fix:* validated the scale psychometrically + separated active founders from non-founders - *For you:* precedent for using Bertrand's scale as your psychological mediator ### THE CLASSICS — legitimize early, survive longer - **One line:** ventures that legitimate FIRST survive longer (real data) — and different audiences judge differently #### "Legitimize first → survive longer" ✅ POSITIVE — real data - **Paper:** Delmar & Shane — Journal of Business Venturing 2004 - *The question:* Does doing legitimation activities early actually help survival — or do good ventures just do them anyway? - *What they did:* Tracked a Swedish new-venture panel for ~30 months using "event-history" models — statistics that measure WHEN ventures die and what happened before - *The legitimation activities counted:* concrete acts like registering the firm, writing a business plan, opening a bank account, seeking external funding — founders who did these BEFORE building the product/team survived more - *Sample:* Swedish new ventures, 1998 founding cohort - *Found:* ventures that legitimated BEFORE other organizing activities survived more - *Caveats → solutions:* - *Problem: self-selection* — maybe serious founders legitimate early AND would survive anyway → *Fix:* analyzed the ORDER and TIMING of activities, not just whether they happened - *For you:* the empirical backbone of "legitimacy first" — exactly what your module teaches #### "Identity must fit the life stage" 💡 THEORY - **Paper:** Fisher, Kotha & Lahiri — Academy of Management Review 2016 - *The question:* Ventures change as they grow — how do identity and legitimacy relate at each stage? - *What they did:* Built a stage-by-stage framework: entry, endorsement, competition, renewal - *Caveats → solutions:* - *Problem: pure theory* → *Fix:* wrote testable propositions for each stage - *For you:* the endorsement mechanism — how atEnA affiliation itself signals legitimacy #### "Many audiences, many thresholds" 📚 REVIEW - **Paper:** Fisher — Organization Theory 2020 - *The question:* The literature was fragmented — different audiences, dimensions, thresholds never unified - *What they did:* Integrative review organizing judgments by audience × threshold × dimension - *For you:* justifies treating banks, clients, family as separate evaluator audiences in your design ## 5. ALDEBERT — the helpers (accelerators & programs) - **What she did:** spent 2 years inside one French accelerator — interviewed managers, watched juries pick startups - **What she measured:** how legitimacy moves from the accelerator to the startup, step by step over time - **What she found:** accelerators CAN lend their reputation to startups — but only after the accelerator itself is trusted - **The catch:** she studied ONE accelerator → other programs might work differently - **Why it matters to you:** atEnA is exactly this kind of program → her work says what atEnA can and can't do for founders ### "Accelerator lends legitimacy only if trusted first" ✅ CONDITIONAL - **Paper:** Aldebert — AEI 2021/2023 - *The question:* Accelerators claim to transfer legitimacy to startups — do they actually, and when does it fail? - *What they did:* 24 months inside one French accelerator — 34 interviews + watching selection juries + internal documents - *What "lending legitimacy" looked like concretely:* the accelerator's name on the startup's materials, introductions to the accelerator's network, selection into the program acting as a public stamp of approval - *Found:* yes, but only after the accelerator built its OWN legitimacy; transfer happens in stages - *Caveats → solutions:* - *Problem: one accelerator only* — a single case can't generalize → *Fix:* real-time 2-year tracking + triangulation (interviews + jury rubrics + performance data all saying the same thing) - *For you:* your design holds atEnA's legitimacy constant → isolates the module's effect ### "Deeptech needs a corporate bridge" ✅ POSITIVE - **Paper:** Aldebert — Entrepreneurship & Innovation 2024 - *The question:* Deeptech ventures have no track record — university incubators give money but not market legitimacy; who bridges the gap? - *What they did:* Compared 2 French university incubators + 8 deeptech spinoffs; interviewed managers, tech-transfer officers, founders - *Found:* established corporate partners act as the essential legitimacy bridge - *Caveats → solutions:* - *Problem: deeptech niche* — may not apply to normal startups → *Fix:* cross-case pattern matching isolates the mechanism itself, not the industry - *For you:* atEnA's partners (CCIs, universities) = legitimacy bridges ## 6. MY METHODS — the 3 tools & how I run them *The 3 designs in plain English → the step-by-step playbook → the deep problem/fix toolbox* ### THE 3 TOOLS — in plain words *Each design = one plain block first (what it is / what you do / what you measure / the catch / why it matters), then "Details" opens the committee caveats & fixes.* #### 1. THE EXPERIMENT — a fair race (Article 1) - **What it is:** you split founders randomly into two groups — one gets the 12h legitimacy module, one doesn't → then you compare them - **What you do:** founders sign up → you flip a coin (randomization) → half attend the module, half don't → everyone answers the same surveys at T0, T1, T4 - **What you measure:** whether the trained group ends up MORE legitimate (self-report + objective milestones + text scores) than the untrained group - **The catch:** real life is messy — people drop out, groups talk to each other, and founders report their own feelings - **Why it matters:** random split = the ONLY difference between groups is the training → it's the ONLY design that can say "training CAUSED the change" #### 2. THE QUESTIONNAIRE — a thermometer (T0 → T1 → T4) - **What it is:** the same questions, same scale, asked at three moments — before training, right after, months later - **What you do:** founders rate how legitimate they feel (Bertrand's self-legitimacy scale) + how judges see them (perceived external standing) — three times - **What you measure:** does the number MOVE after training, and does it move MORE in the treated group than the control? - **The catch:** people say what sounds good, and answering the same survey repeatedly can itself change answers - **Why it matters:** it's the only way to see INSIDE the founder's head — self-legitimacy can't be measured any other way #### 3. THE INTERVIEWS — the "why" camera (Article 2) - **What it is:** you sit with founders and ask them to walk through real episodes — a pitch, a client meeting, a funding ask - **What you do:** 30–60 min conversations with high-responders AND low-responders from both arms → record → transcribe → code into themes - **What you measure:** not numbers — MEANING. Which capability actually moved, and why it worked for one founder but not another - **The catch:** YOU interpret the words — another researcher might read the same transcript differently - **Why it matters:** the experiment says IF training worked; interviews say WHY — the mechanism the committee will ask about #### THE BASELINE — training works, but modestly (McKenzie & Woodruff, World Bank) - Every business-training RCT reviewed → real but small effects (~5–10%) → a legitimacy-SPECIFIC module is the untested lever — your honest starting point ### MY 10 STEPS — what I actually do, in order *Run ATENA in order. Each step says: WHAT you do → WHICH paper you cite → WHICH fix/number you use* #### STEP 1 — Define your variables (before touching anyone) - **What you do:** write down the 3 things you measure, using Schoon's dyad as your ruler: EXPECTATIONS (the judge's yardstick — you don't train this), CONFORMITY (the founder's signals — you train this), ASSENT (the judge's verdict — your outcome) - **Cite:** Schoon 2022 (the dyad) + Bitektine 2011 (why legitimacy specifically, not reputation) - **Plain:** "I measure the founder side of the dyad that training can reach" #### STEP 2 — Choose the scales (your thermometer parts) - **What you do:** adopt Bertrand's self-legitimacy scale for the founder's inside feeling + a perceived-legitimacy scale for how they think judges see them - **Cite:** Bertrand, Benamara & Aldebert 2025 (the scale exists and was validated) + Alexiou & Wiggins (perceived legitimacy) - **Fix to apply:** psychometric validation of your translated versions before fieldwork #### STEP 3 — Calculate your sample size (BEFORE recruiting) - **What you do:** pick the effect you expect (d ≈ 0.3–0.4, because McKenzie's meta says training effects are small-to-medium) → n per group ≈ 16/d² → ~100–175 per arm - **Cite:** McKenzie 2021 (effect size expectations) + his line "not finding an effect is not finding no effect" - **Fix to apply:** pre-registered power analysis — the number is decided now, defended later #### STEP 4 — Recruit & randomize (the fair coin flip) - **What you do:** recruit ATENA founders → collect T0 baseline survey from EVERYONE → then randomly split into treatment/control → keep cohorts separated (different time slots) - **Cite:** Campos et al. 2017 Science (their 1,500-person 3-arm RCT is your design template) - **Fix to apply:** cohort separation from day 1 — it is the contamination fix, applied at setup not cleanup #### STEP 5 — Deliver the module (the treatment) - **What you do:** treatment arm attends the 12h legitimacy module covering the 5 capabilities: Cognitive (look like a recognizable business) · Pragmatic (show the value) · Social (get vouched for) · Factual (show milestones) · Regulative (show compliance) - **Cite:** Siraz JMS online first (categorical fit is the strongest signal → Cognitive is the star) + Siraz EMR 2026 (signals must be consistent — one contradiction stalls the judge) + Claes et al. EMR 2025 (a strategy can backfire → teach audience-appropriateness, not tricks) - **Plain:** the module trains CONFORMITY — Schoon's middle variable #### STEP 6 — Run the measurement waves (T0 → T1 → T4) - **What you do:** same survey to BOTH arms at all 3 time points + collect objective milestones (registrations, funding applications, revenue) + pitch decks for text scoring - **Cite:** Bertrand (the scale) + Ghods BJM 2026 (the text-scoring algorithm for decks) - **Fixes to apply:** control answers the same waves (panel-conditioning cancels) · different question formats across waves (CMV fix) · staggered rollout so control isn't denied forever #### STEP 7 — Guard against dropout (attrition watch) - **What you do:** at each wave, track who left → run the leaver/stayer table: compare quitters' T0 answers to stayers' → if they look the same, say so in the paper - **Fixes to apply:** leaver/stayer comparison (the check) + multiple imputation (the repair) + ITT analysis (everyone counts by original assignment) - **Cite:** standard field-RCT practice — McKenzie 2021 argues most studies were underpowered, so your attrition margins matter double #### STEP 8 — Analyze the experiment (the numbers) - **What you do:** (a) ITT = compare everyone offered vs not offered; (b) TOT = ITT ÷ attendance rate for the dose effect; (c) DiD = (treated change) − (control change); (d) report effect size + CI, not just p-values - **Cite:** the math cheat sheet in EXTERNAL AMMUNITION — Cohen's d, Wald, DiD, kappa - **Plain:** three numbers to report — the offer effect, the attendance effect, and the beyond-time effect #### STEP 9 — Run the interviews (the "why" camera) - **What you do:** pick HIGH responders and LOW responders on purpose from both arms → ask concrete episodes ("walk me through your last pitch") → transcribe → code with Gioia: quote → 1st-order code → theme → dimension → second coder, compute kappa - **Cite:** Aldebert 2021/23 (34 interviews / 24 months in this exact ecosystem = your model) - **Fixes to apply:** anonymized transcripts + neutral wording (social desirability) · triangulate claims vs survey + milestones (recall bias) #### STEP 10 — Write it as the missing piece - **What you do:** position each chapter: Intro (the puzzle — judges decide, training changes founders) → Theory (Schoon umbrella, Siraz assent, Bertrand self-assent) → Methods (this playbook) → Results (Step 8 numbers + Step 9 mechanisms) → Contribution (the first field RCT on legitimacy training) - **Cite:** THESIS ARGUMENT MAP section for the paper-per-chapter wiring - **The closing line:** Siraz proved judges respond to signals; Campos proved training changes behavior; the link between them was never tested — that's you ### THE TOOLBOX — every problem → its fix *Click any method → each problem is its own branch: WHAT the problem is → HOW BIG → each FIX explained* #### PART A — MY 3 TOOLS (the designs I run) ##### A1. FIELD EXPERIMENT — proves the cause (Article 1) - **1. Who & What:** Nascent entrepreneurs randomly split into Treatment (12h module) vs Control. You measure the founder-side of the dyad + objective milestones. - **2. How:** Random assignment → T0 baseline → 12h module → T1 post-test → T4 follow-up. - *The steps, in order ↓* - Step a — Recruit ATENA founders → collect T0 baseline from EVERYONE before anyone knows their group - ⚠ can go wrong: people who signed up are already motivated → Fix: report baseline table for both groups to show they started equal - Step b — Randomize into treatment / control (the fair coin flip) - ⚠ can go wrong: someone swaps or complains → Fix: assignment is locked before the module is announced - Step c — Separate the cohorts (different slots / rooms) - ⚠ can go wrong: they meet anyway at the program → Fix: measure the leak — ask control "what did you hear about the module?" - Step d — Deliver the 12h module to treatment only - ⚠ can go wrong: some treated don't attend → Fix: ITT keeps them counted anyway; attendance rate feeds the TOT number - Step e — Survey both arms at T1 and T4 + collect milestones - ⚠ can go wrong: founders drop out → Fix: leaver/stayer table + imputation - Step f — Analyze: ITT → TOT → DiD → effect size + CI - **Problem: small sample / power** - What it means: too few founders → random noise can hide a real effect → "no effect" even if training works - How big: HIGH — the #1 attack on field experiments - Fix 1 — power analysis - What it is: a calculation you do BEFORE the study that answers "how many founders do I need?" - How it works: you guess the effect size you expect (e.g., medium) → plug in how sure you want to be (usually 80%) → the formula returns the minimum n → you only start once you have it - What it solves: turns "your sample is too small" from a debate into a number you already defended upfront - Fix 2 — ITT (intention-to-treat) - What it is: a rule for who counts in your results - How it works: everyone who was randomly assigned stays in their group for analysis — even the founders who skipped the training. You compare "everyone offered training" vs "everyone not offered" - Why it works: if you only counted attendees, motivated people would pile up on one side → the comparison breaks. Keeping everyone in their original group preserves the fair coin flip - What it solves: dropouts and skippers can't silently break your randomization - Fix 3 — effect size + confidence interval - What it is: instead of just "yes/no significant", you report how BIG the change is and the range of plausible values - How it works: effect size = the gap between groups in plain units (e.g., +0.4 on a 7-point scale); the CI = the honest band around it ("the true effect is probably between X and Y") - What it solves: with a small sample you can still show a meaningful effect and be honest about uncertainty — instead of a bare "p > .05, nothing found" - **Problem: attrition (founders drop out)** - What it means: some founders leave between T0→T1→T4 → if quitters differ from stayers, the groups stop being comparable → silent bias - How big: HIGH — most common field-RCT killer - Fix 1 — leaver/stayer comparison - What it is: a check you run, not a fix that prevents dropout - How it works: take the T0 baseline answers of everyone who later quit → compare them to the stayers → if they look the same (age, confidence, venture stage), dropout was random noise, not systematic - What it solves: turns "your dropouts ruined it" into "I showed the leavers looked identical to stayers" - Fix 2 — multiple imputation - What it is: a way to handle the empty cells instead of deleting those founders - How it works: for a founder who quit at T4, the model uses everything known about them (T0 answers, group, progress) to generate several plausible T4 answers → runs the analysis on each → combines results - What it solves: keeps everyone in the analysis AND keeps honest uncertainty — better than either deleting people or pretending nothing is missing - What remains: imputation assumes missingness is explainable by what you already observed — you state this assumption openly - **Problem: contamination (control learns the content)** - What it means: treated founders share module content with control peers → the control group gets secretly treated → gap shrinks artificially - How big: MEDIUM-HIGH — same program = they talk - Fix 1 — cohort separation - What it is: keep the two groups physically apart - How it works: treatment and control run in different time slots / cohorts → they never meet → no channel for the content to leak - What it solves: removes the leak path entirely rather than hoping nobody talks - Fix 2 — staggered rollout - What it is: control isn't "no training" — it's "training LATER" - How it works: control founders get the module AFTER T4 → their T0–T4 answers are still clean → and you keep everyone happy (nobody is denied training forever) - What it solves: gives a clean untreated comparison period without creating angry "never-treated" founders - Fix 3 — measure the leak directly - What it is: ask, don't guess - How it works: at T1/T4 ask control founders "did you hear about the module content? what did you hear?" → you quantify how much spillover happened instead of assuming zero - What it solves: if some leak happened anyway, you can report its size instead of hiding it - **Problem: self-report only** - What it means: asking "do you feel legitimate?" measures confidence inside the head — not what judges outside actually think - How big: MEDIUM - Fix 1 — objective milestones - What it is: count real-world behavior, not just feelings - How it works: track things founders actually DO — registering the business, submitting funding applications, getting first revenue — these are hard to fake on a survey - What it solves: proves the training changed behavior, not just self-image - Fix 2 — pitch-deck text scores - What it is: an external algorithm scores the legitimacy of their actual pitch texts - How it works: founders submit real pitch decks → Ghods BJM 2026 algorithm scores them on the legitimacy dimensions → reproducible score, no self-report involved - What it solves: an outside-the-head legitimacy measure → kills the "it's all just self-report" critique - **4. How You Apply It:** Article 1 — the causal test the literature has never run. ##### A2. INTERVIEWS — explains the why (Article 2) - **1. Who & What:** You interview founders from both arms — high vs low responders. ~30–60 min each. - **2. How:** Guide built on the 5 capabilities → record → transcribe → Gioia coding. Anchor: Aldebert ran 34 interviews / 24 months in this ecosystem. - *The steps, in order ↓* - Step a — Pick interviewees on purpose: HIGH responders + LOW responders from both arms - ⚠ can go wrong: only talking to fans → Fix: low responders have no reason to flatter you - Step b — Interview 30–60 min each with a guide built on the 5 capabilities - ⚠ can go wrong: vague "it was nice" answers → Fix: ask concrete episodes — "walk me through your last pitch" - Step c — Transcribe and anonymize - ⚠ can go wrong: people self-censor if identifiable → Fix: names stripped before analysis - Step d — Code with Gioia: quote → 1st-order code → 2nd-order theme → aggregate dimension - ⚠ can go wrong: you read your theory into it → Fix: the data-structure table keeps every claim traceable to a quote - Step e — Second coder + kappa on the same transcripts - ⚠ can go wrong: still just two opinions → Fix: keep raw quotes in the appendix so readers audit you - Step f — Triangulate: do interview claims match the survey scores and milestones? - **Problem: coding subjectivity** - What it means: YOU turn quotes into themes — committee could say you read your own theory into the answers - How big: HIGH — the default attack on all qualitative work - Fix 1 — Gioia data structure - What it is: a table that shows the full chain from quote to conclusion - How it works: raw founder quote → 1st-order code (their words) → 2nd-order theme (your grouping) → aggregate dimension (the big claim). Every claim in your paper can be traced back down the chain to an actual quote - What it solves: "you just made this up" → "here is the quote, the code, the theme — audit it" - Fix 2 — dual coders + kappa - What it is: a second researcher independently codes the same transcripts - How it works: they code without seeing your labels → you compare → kappa = the agreement score (> .80 = strong agreement) - What it solves: if two people independently see the same themes, it's not just your opinion - What remains: kappa shows AGREEMENT, not truth — two biased coders could still agree → keep raw quotes visible so readers judge for themselves - **Problem: recall bias** - What it means: months later, founders retell a cleaned-up version — memory edits reality - How big: MEDIUM - Fix 1 — concrete episodes - What it is: ask about a specific event, not an opinion - How it works: "walk me through your last pitch — what did you say, what did they ask?" forces real details instead of a polished story - What it solves: details are hard to invent; vague feelings are easy - Fix 2 — triangulation - What it is: check interview claims against your other two data sources - How it works: if a founder says "I now know how to present myself" → check whether their survey self-legitimacy score actually rose and whether they hit real milestones - What it solves: a claim that survives 3 different data types is much harder to dismiss - **Problem: social desirability** - What it means: founders tell you training "worked" because they want to please you / the program - How big: MEDIUM - Fix — anonymized transcripts + neutral framing + interview LOW responders too - What it is: remove the reason to lie - How it works: names stripped before analysis + ask neutral questions ("what changed, if anything?") + deliberately interview founders the training DIDN'T help → they have no incentive to praise it - What it solves: low responders saying "it didn't help me" proves people aren't just flattering you - **Problem: small n** - What it means: 20–30 interviews can't estimate how common something is in the population - How big: LOW — because it's not the goal - Fix — purposive (extreme-case) sampling - What it is: you choose interviewees on purpose — high responders AND low responders - How it works: comparing "it worked for me" founders against "nothing changed" founders reveals the MECHANISM — what the training does differently inside different people - What it solves: interviews answer WHY, not HOW MANY — and comparing extremes is the design, not a weakness - **4. How You Apply It:** Article 2 — explains HOW/WHY the treatment works inside the founder. ##### A3. SURVEY — tracks the change over time (T0→T1→T4) - **1. Who & What:** Founders report self-legitimacy + perceived standing at 3+ time points. - **2. How:** Same validated scales every wave; treatment × time interaction = the effect. - *The steps, in order ↓* - Step a — T0: everyone answers BEFORE randomization - ⚠ can go wrong: groups start unequal → Fix: T0 is your comparability proof — keep it safe - Step b — T1: same scales right after the module - ⚠ can go wrong: same-sitting inflation (CMV) → Fix: different question formats, separated constructs - Step c — T4: follow-up months later - ⚠ can go wrong: attrition → Fix: incentives per wave + leaver/stayer check - Step d — Anchor to objective markers each wave (milestones + pitch-deck text scores) - ⚠ can go wrong: pure self-report critique → Fix: outside-the-head measures in the same model - Step e — Analyze: treatment × time interaction = DiD logic — the EXTRA change is the effect - **Problem: CMV (common-method variance)** - What it means: you measure cause AND effect on the same questionnaire in the same sitting → the two correlate because of HOW you asked (same scale, same mood, same person), not because of reality - Everyday version: ask someone "do you feel confident?" then "do you feel legitimate?" in one breath — a cheerful person says yes to both → fake correlation - How big: MEDIUM - Fix 1 — temporal separation - What it is: measure different things at different times - How it works: self-legitimacy at T1, outcomes at T4 → the same sitting can't fake a link measured weeks apart - What it solves: the correlation can't be just a moment's mood - Fix 2 — different question formats - What it is: mix up scales and wording so "answering style" can't drive the result - How it works: if every question uses the same 1–7 agree/disagree format, people slip into a pattern → varying it breaks the pattern - What it solves: correlation reflects real difference, not a habit of clicking "4" - Fix 3 — objective markers - What it is: anchor answers to things outside the questionnaire - How it works: compare survey answers against real milestones + the pitch-deck text score → a self-report backed by observable behavior can't be pure answering-style - What it solves: gives the survey an outside reality check - **Problem: panel conditioning** - What it means: filling in the same survey repeatedly makes founders MORE aware of legitimacy — the measurement itself becomes a mini-treatment - Everyday version: weigh yourself daily → you start dieting → the scale changed the behavior it was only supposed to measure - How big: MEDIUM - Fix — control group fills the same survey - What it is: the comparison group also answers at T0, T1, T4 - How it works: if the survey itself teaches awareness, it teaches BOTH groups equally → when you compare groups, the survey effect is identical on both sides → it cancels out - What it solves: whatever the survey does to people, it does to everyone → the training difference stays clean - **Problem: maturation** - What it means: founders improve naturally over time — business skills grow with experience → any change isn't necessarily caused by training - How big: HIGH without a control group, near-zero with it - Fix — randomized control group - What it is: a group that goes through the same months WITHOUT the module - How it works: both groups mature at the same natural rate → control group's improvement = "what time alone does" → treated group's EXTRA improvement beyond that = the training effect - What it solves: "they just got better with time" → "the treated group improved MORE than time alone explains - **4. How You Apply It:** The measurement backbone of Article 1. #### PART B — OTHER PEOPLE’S METHODS (what the literature uses) ##### B1. CONJOINT — judges rate fake profiles (Siraz's tool) - **1. Who & What:** Human judges rate made-up venture profiles — "how legitimate is this?" - **2. How:** Software generates random cue bundles; each judge rates 16–32 profiles. - **Problem: rating fatigue** - What it means: after ~16+ profiles judges skim → noisy data - Fix — "orthogonal" design: profiles randomized so every combination is still tested while each judge sees few - **Problem: non-independence** - What it means: one judge's 16 ratings correlate → statistics overconfident - Fix — multilevel model: ratings nested inside judges, errors clustered per judge - **Problem: hypothetical bias** - What it means: rating a paragraph ≠ risking real money - Fix — realistic pitch vignettes + attention checks - **4. For you:** Planned Article 4 — real bank officers rating atEnA pitch profiles. ##### B2. TEXT SCORING — an algorithm reads pitch decks - **1. Who & What:** An NLP algorithm scores venture texts (websites, pitch decks) against legitimacy dictionaries. - **2. How:** Text → per-dimension legitimacy scores → composite score per firm. - **Problem: nuance blindness** - What it means: algorithms misread sarcasm/context - Fix — human experts calibrate the dictionary - **Problem: validity** - What it means: does the score reflect reality? - Fix — validate against real venture survival (Ghods BJM did this) - **4. For you:** Objective check on pitch decks — kills "it's all self-report." ##### B3. CASE STUDY — embed inside a program (Aldebert) - **1. Who & What:** Researcher embeds inside an accelerator/incubator for months. - **2. How:** Interviews + observation + documents → coded over time. - **Problem: recall bias** — founders misremember → Fix: triangulate with real-time records - **Problem: researcher subjectivity** — your lens shapes findings → Fix: dual coders + audit trail - **Problem: one case can't generalize** → Fix: multi-case comparison (worked vs stalled) - **4. For you:** The model for Article 2's interview design. ##### B4. SURVEY — founders rate themselves - **1. Who & What:** Founders rate themselves on multi-item scales. - **2. How:** Structured questionnaire → factor/cluster analysis. - **Problem: CMV** — same-survey correlations inflated → Fix: temporal spacing + different formats - **Problem: social desirability** — founders overstate → Fix: anonymized + indirect wording - **Problem: no causality** — correlation only → Fix: anchor inside your RCT - **4. For you:** The instrument for self-legitimacy (Bertrand) + perceived legitimacy. ##### B5. WORD-MAPPING — "first 5 words" survey (Ghods) - **1. Who & What:** Entrepreneurs write the first 5 words that come to mind for "legitimate entrepreneur" + rank them. - **2. How:** Frequency × importance matrix separates core beliefs from peripheral ones. - **Problem: word ambiguity** — synonyms fragment data → Fix: lemmatization (rule-based grouping) + dual coders - **Problem: cross-language** — words don't translate 1:1 → Fix: compare groups separately - **4. For you:** Rationale for France vs Italy as distinct reference frameworks. ## 7. DEFENSE PACK — proof, answers, paper map *Outside evidence with real numbers → the math cheat sheet → the 12 hard questions → which paper goes in which chapter* ### OUTSIDE PROOF — numbers from studies nobody can call mine *Studies OUTSIDE your committee's circle — independent proof that the chain "training → founder behavior → legitimacy → resources" is real* #### TRAINING WORKS — and mindset-training works best ##### Campos, Frese, Goldstein et al. (2017, Science) — the Togo RCT - **What they did:** 1,500 micro-enterprise owners, randomly split into 3 groups: control (500) · traditional business training (500) · personal-initiative psychology training (500) - **The numbers:** psychology-based training → **profits +30%** ($60/month) and **sales +17%**, both statistically significant; traditional training → +11% profits, NOT significant - **Why it's in your favor:** it PROVES a short targeted behavioral training beats classic content training → your 12h legitimacy module is exactly this type (behavior/signals, not accounting) - **Bonus:** cost-effective — paid for itself within 1 year → answers "is 12h worth funding?" ##### McKenzie (2021, Oxford Review of Economic Policy) — meta-analysis of all training RCTs - **The numbers:** across all randomized trials of business training → average **+10.1% profits** (95% CI: 4.1% to 16.1%) and **+4.7% sales**; effect on employment ≈ 0 - **His key line (use it verbatim in committee):** *"not finding a significant effect of training is not the same as finding that training has no effect"* — most studies were underpowered - **Why it's in your favor:** (1) training DOES work on average; (2) it defends your power-analysis fix — you sized ATENA to detect exactly this magnitude; (3) training creates profit, not jobs → don't overclaim employment effects #### LEGITIMACY → MONEY — the link is already proven ##### Pollack, Rutherford & Nagy (2012, ETP) — Shark Tank / Dragons' Den - **What they did:** coded 113 REAL funded pitches from TV pitch shows - **Found:** preparedness (a TRAINABLE behavior) → raised perceived cognitive legitimacy → raised the FUNDING AMOUNT — a full mediation chain - **Why it's in your favor:** this is literally your thesis chain on real money data — trainable behavior → legitimacy judgment → resources ##### Nagy, Pollack, Rutherford & Lohrke (2012, ETP) — impression-management experiment - **Found:** founders' credentials AND impression-management behaviors → higher legitimacy perceptions (experimental, so causal) - **Why it's in your favor:** founder behavior moves judges' legitimacy ratings — the exact conformity→assent link ATENA trains ##### Tornikoski & Newbert (2007, JBV) — PSED, n=830 nascent entrepreneurs - **Found:** legitimation ACTIONS predicted whether a venture actually emerged MORE than founder characteristics did — what you DO beats what you ARE - **Why it's in your favor:** legitimacy is built through learnable actions → therefore trainable → therefore a legitimate thesis premise - **Local note:** Tornikoski's PhD was at Aix-Marseille — the legitimacy-emergence tradition lives in your own institution ##### Tornikoski (2005 PhD, Aix-Marseille; Tremplin Entreprises contest) - **Found:** legitimation model explained ~50% of variance in venture emergence - **Why it's in your favor:** half of whether a venture emerges is explained by legitimacy-side factors — huge effect for a "soft" variable #### THE MATH CHEAT SHEET — formulas for the board *Each = formula + plain-English meaning + the number that matters for you* ##### Cohen's d — the effect size - **Formula:** d = (mean_treatment − mean_control) / pooled standard deviation - **Plain:** "how many standard-deviations apart are the two groups?" — d=0.2 small, 0.5 medium, 0.8 large - **For you:** McKenzie's 10% profit effect ≈ small-to-medium d → you need power for d≈0.3–0.4 ##### Power analysis — the sample-size formula - **Rule of thumb:** n per group ≈ 16 / d² for 80% power at p<.05 - **Plain:** medium effect (d=0.4) → ~100 founders/group; small effect (d=0.2) → ~400/group - **For you:** this is why the power analysis must run BEFORE data collection — the formula sets the recruitment target ##### ITT vs TOT — the Wald estimator - **Formula:** TOT = ITT / compliance rate - **Plain:** if ITT effect = +0.10 and only 70% of treated actually attended → the per-attendee effect (TOT) ≈ 0.14 - **For you:** lets you report BOTH the policy effect (ITT: offering training) and the dose effect (TOT: attending training) ##### Difference-in-Differences (DiD) - **Formula:** (T1−T0 of treatment) − (T1−T0 of control) - **Plain:** "extra change beyond what time alone does" — subtracts maturation automatically - **For you:** your T0→T1→T4 design is built exactly for this subtraction ##### Cohen's kappa — inter-rater agreement - **Formula:** κ = (observed agreement − chance agreement) / (1 − chance agreement) - **Plain:** agreement AFTER removing what random guessing would produce — κ>.80 = strong, >.60 = acceptable - **For you:** the number that turns "you coded it yourself" into "two independent coders agreed at κ=?" ##### AMCE — how conjoint effects are read (Siraz's method) - **Plain:** the average change in the rating when ONE cue switches on while all others stay randomized — lets her rank signals (categorical fit >> endorsement) without confounds - **For you:** your Article 4 pitch-experiment will report the same AMCEs — same grammar as Siraz, real bank officers as judges #### BACKUP CITES — if the committee pushes harder - **Bitektine (2011, AMR):** legitimacy, status, reputation and trust are DIFFERENT judgment types → defends why you study legitimacy specifically, not "reputation" - **Suddaby, Bitektine & Haack (2017, AMR):** the legitimacy review — supports the multi-level propriety→validity logic - **Navis & Glynn (2010, AMJ):** satellite-radio case — legitimation strategies for radically new categories → backs your cognitive-framing capability - **Tracey, Dalpiaz & Phillips (2018, Org Science):** "fish out of water" — ventures legitimate themselves by TRANSLATING identity claims for each audience → different audiences need different signals - **Martin, McNally & Kay (2013, JBV) + Bae et al. (2014, ETP):** meta-analyses — entrepreneurship education → entrepreneurial intentions/actions, r ≈ .14–.16, small but consistent → education changes founder behavior at the population level ### FRESH AMMUNITION — 49 papers (2024–26) + the 3 validated scales *Two sources: P01–P30 = most-cited recent papers citing the field's canon (OpenAlex, Sep 2026) · P31–P46 = your Scopus export · P47–P49 = the validated instruments · [cit.] = citation count of that source.* *REORDER TOOL (you only): every paper is ONE line — cut/paste single lines or whole #### blocks to reorder. IDs are stable keys you can cite from anywhere ("see P13"). solves: = the problem that paper fixed.* *Format: FORMULA first (what constructs into what), names second — grouped by level on Schoon's dyad: THE JUDGE → THE CONNECTION → THE OBJECT → THE FIELD. Most new work sits in THE CONNECTION — exactly where this thesis lives.* #### L1. THE JUDGE — evaluator side - **[P16] verdicts → collective validity (masks low consensus)** — Haack et al. '25 JMS · 14 · *solves: validity/agreement blur→multilevel split* - **[P17] societal acceptance ≠ institutional legitimacy** — Alsheimer et al. '24 RSER · 14 · *solves: separate literatures→integration* - **[P18] norms → investor yardstick → IPO diversity↑** — Kenney et al. '24 SBE · 14 · *solves: norm-shift mechanism→30-yr panel* - **[P20] judgment = variance + prediction + wisdom** — Tsoukas et al. '24 AMA · 12 · *solves: loose term→3 perspectives* - **[P21] CEO humor × schema-fit → social evaluation** — König et al. '24 AMR · 12 · *solves: schema filtering→humor×fit model* - **[P40] validity cues × categorical fit → propriety (fit dominates)** — Thaler, Sievert, Siraz & Pinz '26 JMS · 2 · *solves: evaluator propriety→factorial survey n=1,866* - **[P42] text scoring → legitimacy per dyad at scale** — Ghods, Ricard, Aldebert & Schoon '26 BJM · 0 · *solves: uni-dim measures→multi-dim dyads (our instrument)* - **[P43] multi-source evals → validity↑ → action** — Du et al. '26 BJM · 0 · *solves: single-source→multi-source* - **[P38] content endorsed + agent rejected → cooptive rejection** — Bourgoin et al. '26 JMS · 2 · *solves: object-agent tie→dissociation* - **[P37] misconduct × stakeholder → defence varies** — Oberhauser et al. '25 CSR · 4 · *solves: one-size repair→differentiated defence* #### L2. THE CONNECTION — founder↔evaluator boundary - **[P09] hype → expectations↑ → resources↑** — Rady et al. '25 JBV · 20 · *solves: hype≠always good→contingent value* - **[P10] founder updates ↔ backer sentiment** — Wood et al. '24 JBV · 19 · *solves: hype decay→longitudinal pairs* - **[P11] internal challenges → 8 inward strategies** — Loos & Spraul '24 JMS · 19 · *solves: external-only focus→internal repertoire* - **[P13] hype → legitimacy↑ then ↓↓** — Spivack et al. '25 JBV · 17 · *solves: hyped crashes→3-phase model* - **[P15] local legitimacy × brokerage → diffusion** — Kansheba et al. '24 IJEBR · 15 · *solves: local-only→diffusion beyond ecosystem* - **[P19] crisis comms → crowdfunding outcome** — McKenny et al. '24 ETP · 14 · *solves: comms black box→dims + vignette* - **[P24] rebranding → host-market legitimacy** — Ramos da Silva et al. '24 IMR · 9 · *solves: untested cross-country→6-country case* - **[P26] make-or-break intermediary → 3 pathways** — Neuberger et al. '24 AMJ · 7 · *solves: unstudied→accelerator ethnography* - **[P29] secrecy → 7 discursive strategies** — Clarke et al. '24 JBE · 7 · *solves: secrecy legitimation→repertoire* - **[P28] eWOM → de-stigmatization → legitimation** — Villers et al. '24 JSR · 7 · *solves: stigma mechanism→eWOM model* - **[P32] non-Western ↔ Western investors → tensions** — Jones & Murray '26 JMS · 7 · *solves: universalism→tension lens* - **[P33] challenge×strategy configs → deals** — Yang, Zhang & Wu '25 MD · 7 · *solves: single-strategy→fsQCA 328* - **[P36] wedging+alignment → revised assessments** — Dahan & Leca '25 JMS · 4 · *solves: hybridity liability→reframing* - **[P41] political affinity → endorsements → IPO legitimacy** — Song, Hu & Xia '25 JMS · 1 · *solves: transfer assumed→pathways* - **[P44] contradiction → myths → internal legitimacy** — Howard & Brown '26 JMS · 0 · *solves: contradiction→myth model* - **[P39] crisis narratives → legitimacy managed** — de Castilho Rossoni & Lima '26 CSR · 2 · *solves: disclosure→CDA+ethics* #### L3. THE OBJECT — venture side - **[P02] green innovation → legitimacy → export↑** — Shu et al. '24 MOR · 35 · *solves: EMF barriers→innovation vehicle* - **[P04] rival alliances → ecosystem defence** — Chen et al. '24 JMS · 27 · *solves: venture vulnerability→rival-alliance defence* - **[P05] resourcefulness → venture process** — Williams et al. '24 JMS · 23 · *solves: fragmented lit→integrated model* - **[P07] category density → legitimacy spillover** — Li, Cai & Bosma '24 SBE · 21 · *solves: new forms legitimize→spillover* - **[P27] category reconfiguration → new-category legitimacy** — Rhee, Lo & Durand '25 OrgSci · 7 · *solves: challenger legitimation→3 strategies* - **[P31] awareness+motive+weak capability → decoupling** — Xu et al. '25 JMS · 20 · *solves: signal w/o substance→config drivers* - **[P34] responsible ent. → social legitimacy** — Adomako & Tran '25 CSR · 7 · *solves: boundary conditions→time-lag survey* - **[P35] EO → substitutive legitimacy → underconformity** — Askarzadeh et al. '25 JMS · 6 · *solves: underconformity→substitutive mech* #### L4. THE FIELD — macro environment - **[P01] strategies×positions×foundations×time×arenas → legitimation** — Vaara et al. '24 JoM · 68 · *solves: fragmented→5-element framework* - **[P03] legitimacy = create→maintain→lose** — Meyer & Tse '25 IBR · 30 · *solves: snapshots→process view* - **[P06] illegitimate actors → institutional repair** — Rintamäki et al. '24 AMR · 22 · *solves: threat framing→repair role* - **[P08] intensive mentoring → accelerator effect** — Rechter & Avnimelech '24 SBE · 21 · *solves: active ingredient→mentoring* - **[P12] university incubator → social intrapreneurship** — Nafari et al. '24 Technovation · 18 · *solves: econ-only→social value* - **[P14] racial identity → lived experience** — Santos et al. '24 SBE · 17 · *solves: race as control→identity-centered* - **[P22] underexploited opportunity → inequality** — Lewis et al. '24 AMR · 12 · *solves: equalizer myth→inequality* - **[P23] outcome = institution + cohort** — Avnimelech et al. '24 SMJ · 12 · *solves: confound→Bayesian decomposition n=1,350* - **[P25] 3R strategies → marginalized legitimation** — Spanuth & Urbano '24 AMP · 8 · *solves: marginalization→3R strategies* - **[P30] accelerator × ecosystem = complement/substitute** — Qin '25 JIBS · 3 · *solves: uniform effect→ecosystem moderator* - **[P45] invoked past → evaluative authorization** — Markman et al. '26 JMS · 0 · *solves: legacy divergence→authorization theory* - **[P46] digital forces → evaluation assumptions break** — Clemente et al. '26 JMS · 0 · *solves: stable audiences→rethink* #### S · SCALE ANCHORS — the validated instruments (any year) - **[P47] self-legitimacy scale (4 dims) → impostor syndrome → intention↓** — Bertrand, Benamara & Aldebert '25 HAL · wp · *solves: self-legitimacy never measured → validated scale n=122* - **[P48] individual perceptions → pragmatic/moral/cognitive legitimacy scale** — Alexiou & Wiggins '19 Strategic Organization · 138 · *solves: legitimacy measured only collectively → individual-perception scale (7 studies)* - **[P49] entrepreneurial self-efficacy scale → the benchmark you discriminate against** — Chen, Greene & Crick '98 JBV · ~4k · *solves: no ESE measure → validated scale (Q6's foil)* ### 12 TOUGH QUESTIONS — with my answers *Structured for rapid 30-second mastery: Model Answer (Core Claim + Two-Link Logic) → Backing Literature* #### A. THE GAP — "isn’t it just nobody did this before?" ##### Q1: "Isn't your gap merely an absence of literature ('nobody did an RCT')?" - **Model Answer:** No. It is a fundamental theoretical tension. Legitimacy is defined as an evaluation that resides strictly in the external stakeholder's mind, whereas entrepreneurship training operates strictly inside the founder. Existing theory cannot predict whether altering a founder's internal knowledge and signals will successfully bridge the boundary and change an external evaluator's proprietary judgment. - **Backing Literature:** Mark Suchman (1995 AMR); Eric Schoon (2022 ASR); Sonia Siraz et al. (2023 JMS). ##### Q2: "Why do you assume your novice founders start in 'Unknown' legitimacy rather than Illegitimacy?" - **Model Answer:** Following Siraz et al. (2023), unknown legitimacy includes the *undecided* sub-state, where evaluators lack sufficient information to judge. Novice, unconventional founders possess no negative track record or stigma—they simply suffer from a lack of visibility and recognizable category fit. We do not assume this as a proven fact; we treat it as a theoretical expectation and empirically characterize baseline distributions at T0. - **Backing Literature:** Siraz, Claes, De Castro & Vaara (2023 JMS, Table I); Thaler, Sievert, Siraz et al. (JMS Online First). ##### Q3: "What if your experiment finds a null result—does your whole thesis fail?" - **Model Answer:** Absolutely not; a null result is theoretically profound. It identifies an empirical boundary condition for entrepreneurship policy: proving that founder-side capability training cannot cross the evaluator boundary. As Claes et al. (2025) and Siraz et al. (2026) show, evaluator bias and perceived strategy appropriateness can override founder behavior. If training fails to move external judgments, it proves public policy must shift from founder training to evaluator-facing certification platforms. - **Backing Literature:** Claes, Siraz et al. (2025 EMR); Siraz, Claes & Alvarez (2026 JBE); Bitektine & Haack (2015 AMJ). ##### Q4: "What are your 5 training dimensions, and what does the intervention train?" - **Model Answer:** Our 12-hour intervention trains **5 distinct legitimacy dimensions** (as established in our training design): - **1. Cognitive:** Clear positioning and category framing of the venture - **2. Pragmatic:** Demonstrating concrete value creation and client utility - **3. Social:** Building networks, ties, and social ease/vouching - **4. Factual:** Documenting expertise, track records, and operational milestones - **5. Regulative:** Regulatory compliance that goes beyond baseline legal minimums - *Theoretical Link:* These 5 trained dimensions operationalize Suchman's (1995) classical macro-dimensions by equipping founders with the specific behavioral signaling repertoires needed to generate external validity cues. - **Backing Literature:** Mark Suchman (1995 AMR); Zimmerman & Zeitz (2002 AMR); Greg Fisher (2020 Org Theory); Alex Bitektine (2011 AMR). #### B. MEASUREMENT — "isn’t it all self-report?" ##### Q5: "Your dependent variable relies on founder self-report. Isn't that circular?" - **Model Answer:** We explicitly acknowledge that our survey captures the founder-side node of Schoon's (2022) relational dyad. To ensure methodological rigor, we include objective stakeholder validation milestones, and our design incorporates two direct validation extensions: applying Ghods et al.'s (2026) computational text-analysis scoring to participants' written pitch decks, and executing a Siraz-style evaluator conjoint experiment in Article 4. - **Backing Literature:** Eric Schoon (2022 ASR); Ghods, Ricard, Aldebert & Schoon (2026 BJM); Daisy Bertrand et al. (2025 HAL). ##### Q6: "Why use Bertrand's Self-Legitimacy scale instead of standard General Self-Efficacy?" - **Model Answer:** General self-efficacy measures a generalized belief in one's capability to execute tasks. In contrast, Bertrand's self-legitimacy instrument measures the entrepreneur's internalized feeling of social rightness, appropriateness, and entitlement to operate as a venture founder. As Bertrand, Benamara & Aldebert (2025) prove [P47], self-illegitimacy specifically triggers impostor syndrome and suppresses action — making it the precise mediator. Bandura's (1997) general self-efficacy and Chen, Greene & Crick's (1998) ESE scale [P49] measure capability beliefs, not social rightness; Alexiou & Wiggins (2019) [P48] cover perceived external legitimacy. Three different constructs — I measure all sides. - **Backing Literature:** Daisy Bertrand, Benamara & Aldebert (2025 HAL); Alex Bitektine (2011 AMR); Michael Tost (2011 AMR). ##### Q7: "How do you defend your construct validity against Common-Method Variance (CMV)?" - **Model Answer:** We employ procedural and statistical remedies: (1) Temporal separation between Baseline (T0), Post-treatment (T1), and Follow-up (T4); (2) Separating self-legitimacy items from external stakeholder validation scales; and (3) Triangulating self-reports against objective textual artifacts scored via Ghods et al.'s (2026) multi-dimensional NLP method. - **Backing Literature:** Podsakoff et al. (2003 JAP); Ghods, Ricard, Aldebert & Schoon (2026 BJM). #### C. ECOSYSTEM & BIAS — "what if bias wins anyway?" ##### Q8: "Is atEnA itself legitimate enough to confer legitimacy to its participants?" - **Model Answer:** Aldebert's (2021, 2023) research demonstrates that support structures must establish their own institutional legitimacy before they can confer it. In our field RCT, atEnA's program-level standing and institutional backing (CCIs, regional universities) are held strictly constant between treatment and control cohorts, allowing us to isolate the specific incremental value-add of the 12h educational curriculum. - **Backing Literature:** Bénédicte Aldebert (2021, 2023 AEI); Aldebert (2024 E&I). ##### Q9: "How do you know your 5 capability categories won't backfire?" - **Model Answer:** We do not assume they will succeed ex-ante. Claes, Siraz et al. (2025) empirically demonstrated that legitimation strategies favored by an industry can be perceived as inappropriate by evaluators and actively backfire. That is precisely why our study is designed as an empirical field experiment: to test whether training these capabilities produces positive, neutral, or negative effects on evaluator judgments. - **Backing Literature:** Claes, Siraz, De Castro & Lapeyre (2025 EMR); Siraz, Claes & De Castro (2026 EMR). ##### Q10: "Won't structural bias (gender, ethnicity) override any training effect?" - **Model Answer:** Siraz et al. (2020, 2026) demonstrate that women, immigrants, and ethnic minorities face persistent baseline legitimacy penalties. This is why our empirical analysis plan incorporates formal subgroup moderation: to test whether capability training can partially compensate for structural identity deficits, or whether evaluator bias imposes an insurmountable ceiling. - **Backing Literature:** Sonia Siraz et al. (2020 AOM); Siraz, Claes & Alvarez (2026 JBE). ##### Q11: "How do you account for institutional differences between France and Italy?" - **Model Answer:** Ghods, Ricard & Aldebert (2024) show that social representations of the 'legitimate entrepreneur' vary across operating environments. Because norms, values, and beliefs (NVBs) differ across the Franco-Italian border, we model geographic site as a formal multi-group moderator rather than pooling cross-border data into a single homogeneous bucket. - **Backing Literature:** Ghods, Ricard & Aldebert (2024 E&RD); Siraz et al. (2023 JMS). ##### Q12: "Why did you choose a 12-hour module instead of a multi-month incubator program?" - **Model Answer:** A focused 12-hour intervention provides high experimental fidelity and clean causal identification. In long, multi-month incubation programs, treatments become confounded with peer effects, ad-hoc mentoring, and external market shocks. A 12-hour structured curriculum allows us to cleanly isolate the direct causal impact of codified legitimation training on entrepreneurial self-perception and signaling behavior. - **Backing Literature:** McKenzie & Woodruff (2014 World Bank); Anderson et al. (2018 Management Science); Greg Fisher (2020 Org Theory). ### PAPER → CHAPTER MAP — which paper goes where #### 1. INTRODUCTION & MOTIVATION - **Core Argument:** Unconventional and novice founders face a structural, non-meritocratic legitimacy deficit that paralyzes action - Siraz et al. (JBE 2026): Proves ethnicity penalty persists holding debt and financials constant - Siraz et al. (AOM 2020): Shows credentials compensate less for minority/immigrant founders - Bertrand, Benamara & Aldebert (2025): Proves feeling illegitimate suppresses entrepreneurial intention - **Key Phrasing:** *"The legitimacy hurdle is documented to be structural rather than merit-based..."* #### 2. THEORETICAL FRAMEWORK - **Core Argument:** Legitimacy is a multi-state relational judgment; training represents a transition from Unknown/Undecided to Conditional/Full Legitimacy - Siraz et al. (JMS 2023): 5-state framework (Cite as theoretical/proposed) - Schoon (ASR 2022): Relational dyad requiring expectations, assent, and conformity - Suchman (AMR 1995): Foundational pragmatic, moral, and cognitive dimensions - **Key Phrasing:** *"We conceptualize novice ventures as occupying unknown legitimacy, specifically the undecided sub-state..."* #### 3. MECHANISMS & HYPOTHESES - **Core Argument:** Evaluators judge via cues where categorical fit dominates; training improves cue emission and bundle coherence - Thaler, Siraz et al. (JMS Online First): Categorical fit is the #1 predictor of propriety - Siraz et al. (EMR 2026): Incongruent cue bundles stall evaluators in intermediate states - Aldebert (AEI 2021/2023): Support structures transfer legitimacy sequentially - **Key Phrasing:** *"Training targets categorical fit while ensuring cue coherence across capability dimensions..."* #### 4. COUNTER-ARGUMENTS & BOUNDARY CONDITIONS - **Core Argument:** Legitimation attempts are themselves evaluated; inappropriate strategies backfire, and structural bias may cap training effects - Claes, Siraz et al. (EMR 2025): Legitimation strategies backfire when perceived as improper - Siraz et al. (JBE 2026 / AOM 2020): Bias embedded in reference frameworks creates potential ceiling effects - **Key Phrasing:** *"A null or heterogeneous treatment effect is theoretically informative rather than an empirical failure..."* #### 5. METHODOLOGY & MEASUREMENT DEFENSE - **Core Argument:** Perceptual scales capture the founder node of the dyad; text analysis and conjoint designs represent natural validation extensions - Schoon (ASR 2022): The relational dyad as the proper unit of analysis - Ghods, Ricard, Aldebert & Schoon (BJM 2026): Computational text scoring of venture communications - Ghods, Ricard & Aldebert (E&RD 2024): Contextual variation in legitimacy representations - **Key Phrasing:** *"Our instrument captures the founder-side node of Schoon's relational dyad..."* #### 6. THE CORE GAP & CONTRIBUTION - **Core Argument:** The literature has tested evaluator cognition (Siraz) and ecosystem transfer (Aldebert), but has never tested whether the founder can be trained via an RCT - **Papers to Cite:** The complete network synthesis + Ghods (2021 PhD template) - **Key Phrasing:** *"What remains untested is whether training the founder crosses the boundary into the evaluator's judgment..."*