Data Analysis & Validity (jbv-data-analysis)
When to trigger
- Data are collected and it is time to estimate and report
- You are unsure the estimator matches an entrepreneurial-data structure (venture survival, founding choice, nested funding events)
- Reviewers will probe survivorship, selection into founding, or endogeneity
- A handling editor says "the analysis does not support the inference about entrepreneurship"
Match the estimator to the entrepreneurial data structure
JBV is methodologically pluralistic, so the right tool depends on the claim. Common patterns in new-venture data:
| Data structure / claim |
Estimator |
| Time-to-exit / IPO / failure |
Survival / event-history (Cox, parametric AFT, competing risks) |
| Choice to found / endogenous selection |
Heckman / Roy selection; control function |
| Venture panel with unit heterogeneity |
Fixed/random effects; cluster-robust SE (reghdfe, fixest) |
| Policy / ecosystem / financing shock |
DiD / event study / staggered-adoption estimators |
| Counts (patents, funding rounds, ventures) |
Poisson / negative binomial; zero-inflated as fits |
| Binary outcomes (funded, survived) |
Logit / probit; rare-events corrections where outcomes are rare |
| Manipulated entrepreneurial judgment |
ANOVA/regression with manipulation & attention checks |
| Inductive process / theory-building |
Gioia data structure, audit trail, representative quotations |
Cluster standard errors to the sampling/nesting structure (e.g., by cohort, region, accelerator, or industry).
Handle entrepreneurship-specific threats
- Survivorship: report how failed/exited ventures are retained or how their absence is bounded; an analysis on survivors only must say so and qualify the inference.
- Selection into founding: model the founding decision or use a design-based identification; do not interpret survivor associations as antecedents of venture creation.
- Attrition in nascent panels (PSED/KFS-style): document attrition, test for differential attrition, and use FIML/multiple imputation rather than listwise deletion by default.
- Endogeneity of resources/financing/strategy: IV/2SLS, DiD, matching, or control functions, with the identifying assumption stated and probed.
Robustness expected by JBV reviewers
- Alternative specifications (controls in/out, alternative venture measures, subsamples by stage/region).
- Sensitivity to selection and survivorship assumptions (bounds, alternative frames).
- Rule out the leading alternative explanation for the entrepreneurial finding empirically.
- For experiments: report manipulation/attention checks, effect sizes, and pre-registration if any.
Reporting
- Report effect sizes and practical magnitude for the entrepreneurial phenomenon, not just p-values.
- For mediation, report indirect effects with bootstrap CIs; for moderation, plot simple slopes.
- For qualitative work, make the path from raw founder data to constructs traceable.
Execution bridge (StatsPAI / Stata MCP)
Run the battery, don't just enumerate it. Full map:
execution-with-mcp. JBV studies founders and ventures where selection / survivorship threatens every claim; lead with identification and selection-correction tooling.
- Many outcomes / specifications:
romano_wolf (step-down FWER) or
benjamini_hochberg — report the adjusted threshold.
- OVB sensitivity:
oster_delta / sensemakr.
- Inference:
wild_cluster_bootstrap (few clusters), twoway_cluster / conley;
multilevel data → cluster at the right level.
- Re-fit off one handle:
audit_result(result_id) lists the missing checks and the
exact suggest_function for each.
- Exhibits:
etable / did_summary_to_latex from the handle — no retyped numbers.
Keep the decisive checks in the body and the exhaustive battery in the appendix. See the
executed chain in the JF execution walkthrough.
Checklist
Referee-pushback patterns and the JBV-specific fix
| Reviewer pushback |
JBV-specific fix |
| "Your sample is survivors; failures are invisible." |
Re-draw the frame from registry/nascent data; bound the survivor bias. |
| "Founder choices are endogenous." |
Model founding selection or use a shock; report how the estimate moves. |
| "A general-management effect on a startup panel." |
Test a moderation that only makes sense for ventures (uncertainty, liability of newness). |
Worked micro-example (illustrative numbers)
A hypothetical JBV study asks whether prior startup failure raises the hazard of a founder's next venture securing Series A. Data: an illustrative panel of second-time founders from a registry frame retaining failed first ventures (so it is not survivor-only).
- Estimator: a Cox model for time-to-Series-A, clustered by accelerator cohort. (Illustrative) HR = 1.34, 95% CI [1.08, 1.66].
- Survivorship guard: a funded-only frame inflates the HR to ≈ 1.71 — a bias illustration only.
- Selection guard: a first-stage probit for "founds again" yields an inverse Mills ratio that moves the HR to 1.28 — still positive, so the inference holds.
- Effect size: HR 1.28 implies reaching Series A ~4–5 months sooner at the median, tied to learning-from-failure theory rather than a bare star.
Calibration anchors (hedged)
- The bar is identification in service of an entrepreneurship mechanism; a flawless instrument with no venture-theory payoff still risks a "theory-thin" reject.
- Reviewers weight process and micro-foundations: show the entrepreneurial actor's decision in the data, not a reduced-form association alone.
- Robustness norms are pluralistic. Treat any "required" battery as guidance and confirm against the journal's current author guidelines.
Anti-patterns
- OLS on time-to-event data instead of survival models.
- Survivor-only inference read as antecedents of venture creation.
- Ignoring founding self-selection in observational venture data.
- p-values with no effect size or practical entrepreneurial meaning.
Output format
【Estimator】survival / selection / panel-FE / DiD / experiment / qual ...
【Survivorship & selection】how handled ...
【Attrition / missing data】...
【Endogeneity】strategy + identifying assumption ...
【Robustness】alt specs, bounds, alternative explanation ...
【Effect sizes】magnitude for the phenomenon ...
【Open issues for reviewers】...
【Next step】jbv-contribution-framing
1---2name: jbv-data-analysis3description: Use when running and reporting the analysis for a Journal of Business Venturing (JBV) manuscript — choosing estimators that fit entrepreneurial data (survival/event-history, selection models, panels, experiments, qualitative trustworthiness), handling attrition and endogenous founding, and reporting robustness. Executes and reports the analysis; it does not design the study (jbv-methods) or frame the contribution (jbv-contribution-framing).4---56# Data Analysis & Validity (jbv-data-analysis)78## When to trigger910- Data are collected and it is time to estimate and report11- You are unsure the estimator matches an entrepreneurial-data structure (venture survival, founding choice, nested funding events)12- Reviewers will probe survivorship, selection into founding, or endogeneity13- A handling editor says "the analysis does not support the inference about entrepreneurship"1415## Match the estimator to the entrepreneurial data structure1617JBV is methodologically pluralistic, so the right tool depends on the claim. Common patterns in new-venture data:1819| Data structure / claim | Estimator |20|---------------------------------------------------------|------------------------------------------------------------------|21| Time-to-exit / IPO / failure | Survival / event-history (Cox, parametric AFT, competing risks) |22| Choice to found / endogenous selection | Heckman / Roy selection; control function |23| Venture panel with unit heterogeneity | Fixed/random effects; cluster-robust SE (`reghdfe`, `fixest`) |24| Policy / ecosystem / financing shock | DiD / event study / staggered-adoption estimators |25| Counts (patents, funding rounds, ventures) | Poisson / negative binomial; zero-inflated as fits |26| Binary outcomes (funded, survived) | Logit / probit; rare-events corrections where outcomes are rare |27| Manipulated entrepreneurial judgment | ANOVA/regression with manipulation & attention checks |28| Inductive process / theory-building | Gioia data structure, audit trail, representative quotations |2930Cluster standard errors to the sampling/nesting structure (e.g., by cohort, region, accelerator, or industry).3132## Handle entrepreneurship-specific threats3334- **Survivorship**: report how failed/exited ventures are retained or how their absence is bounded; an analysis on survivors only must say so and qualify the inference.35- **Selection into founding**: model the founding decision or use a design-based identification; do not interpret survivor associations as antecedents of venture creation.36- **Attrition** in nascent panels (PSED/KFS-style): document attrition, test for differential attrition, and use FIML/multiple imputation rather than listwise deletion by default.37- **Endogeneity** of resources/financing/strategy: IV/2SLS, DiD, matching, or control functions, with the identifying assumption stated and probed.3839## Robustness expected by JBV reviewers4041- Alternative specifications (controls in/out, alternative venture measures, subsamples by stage/region).42- Sensitivity to selection and survivorship assumptions (bounds, alternative frames).43- Rule out the leading alternative explanation for the entrepreneurial finding empirically.44- For experiments: report manipulation/attention checks, effect sizes, and pre-registration if any.4546## Reporting4748- Report effect sizes and practical magnitude for the entrepreneurial phenomenon, not just p-values.49- For mediation, report indirect effects with bootstrap CIs; for moderation, plot simple slopes.50- For qualitative work, make the path from raw founder data to constructs traceable.5152## Execution bridge (StatsPAI / Stata MCP)5354Run the battery, don't just enumerate it. Full map:55[`execution-with-mcp`](../../../shared-resources/empirical-methods/execution-with-mcp.md). JBV studies founders and ventures where selection / survivorship threatens every claim; lead with identification and selection-correction tooling.5657- **Many outcomes / specifications:** `romano_wolf` (step-down FWER) or58 `benjamini_hochberg` — report the adjusted threshold.59- **OVB sensitivity:** `oster_delta` / `sensemakr`.60- **Inference:** `wild_cluster_bootstrap` (few clusters), `twoway_cluster` / `conley`;61 multilevel data → cluster at the right level.62- **Re-fit off one handle:** `audit_result(result_id)` lists the missing checks and the63 exact `suggest_function` for each.64- **Exhibits:** `etable` / `did_summary_to_latex` from the handle — no retyped numbers.6566Keep the decisive checks in the body and the exhaustive battery in the appendix. See the67executed chain in the [JF execution walkthrough](../../../Journal-of-Finance-Skills/resources/worked-examples/02-execution-walkthrough.md).68## Checklist6970- [ ] Estimator matches the structure (survival/selection/panel/experiment/qual)71- [ ] Survivorship and selection into founding addressed, not assumed away72- [ ] Attrition documented; principled missing-data handling73- [ ] Endogeneity strategy executed and identifying assumption discussed74- [ ] SEs clustered to the entrepreneurial sampling structure75- [ ] Robustness + leading-alternative-explanation tests reported76- [ ] Effect sizes and practical magnitude interpreted7778## Referee-pushback patterns and the JBV-specific fix7980| Reviewer pushback | JBV-specific fix |81|-----------------------------------------------------------|-------------------------------------------------------------------------------------|82| "Your sample is survivors; failures are invisible." | Re-draw the frame from registry/nascent data; bound the survivor bias. |83| "Founder choices are endogenous." | Model founding selection or use a shock; report how the estimate moves. |84| "A general-management effect on a startup panel." | Test a moderation that only makes sense for ventures (uncertainty, liability of newness).|8586## Worked micro-example (illustrative numbers)8788A hypothetical JBV study asks whether prior startup failure raises the *hazard* of a founder's next venture securing Series A. Data: an illustrative panel of second-time founders from a registry frame retaining failed first ventures (so it is not survivor-only).8990- **Estimator**: a Cox model for time-to-Series-A, clustered by accelerator cohort. (Illustrative) HR = 1.34, 95% CI [1.08, 1.66].91- **Survivorship guard**: a funded-only frame inflates the HR to ≈ 1.71 — a bias illustration only.92- **Selection guard**: a first-stage probit for "founds again" yields an inverse Mills ratio that moves the HR to 1.28 — still positive, so the inference holds.93- **Effect size**: HR 1.28 implies reaching Series A ~4–5 months sooner at the median, tied to learning-from-failure theory rather than a bare star.9495## Calibration anchors (hedged)9697- The bar is **identification in service of an entrepreneurship mechanism**; a flawless instrument with no venture-theory payoff still risks a "theory-thin" reject.98- Reviewers weight **process and micro-foundations**: show the entrepreneurial actor's decision in the data, not a reduced-form association alone.99- Robustness norms are pluralistic. Treat any "required" battery as guidance and **confirm against the journal's current author guidelines**.100101## Anti-patterns102103- **OLS on time-to-event** data instead of survival models.104- **Survivor-only inference** read as antecedents of venture creation.105- **Ignoring founding self-selection** in observational venture data.106- **p-values with no effect size** or practical entrepreneurial meaning.107108## Output format109110```111【Estimator】survival / selection / panel-FE / DiD / experiment / qual ...112【Survivorship & selection】how handled ...113【Attrition / missing data】...114【Endogeneity】strategy + identifying assumption ...115【Robustness】alt specs, bounds, alternative explanation ...116【Effect sizes】magnitude for the phenomenon ...117【Open issues for reviewers】...118【Next step】jbv-contribution-framing119```
brycewang-stanford