orgstud-data-analysis

Data Analysis & Evidence (orgstud-data-analysis)

When to trigger

• You have data but the path from raw material to theory is opaque
• Qualitative: your quotes are decorative, not evidentiary; the coding is undocumented
• Process: you have events but no visible analytic structure turning them into a model
• Quantitative: main results exist but robustness and alternative explanations are thin
• A reviewer asks "how did you get from your data to these constructs?"

OS expects readers to see how data became theory

OS's interpretive, European tradition makes analytic transparency a first-class criterion — qualitative rigor is judged on its own terms, not against a quantitative yardstick. The reader must be able to audit the inference from raw data to theoretical claim. Make the analytic ladder visible.

Branch A — Qualitative analysis (the data-to-theory ladder)

• Transparent coding. Show first-order codes (informant terms), second-order themes (researcher constructs), and aggregate dimensions — the Gioia data structure — or an equivalent (Eisenhardt cross-case tables, Langley process bracketing). State who coded, how disagreements were resolved, and how iteration with theory proceeded.
• Data-to-theory table. A table linking representative raw evidence → codes → constructs, so the inference is auditable (build it with orgstud-tables-figures).
• Power quotes vs. proof quotes. A few vivid "power quotes" in the body carry the argument; corroborating "proof quotes" sit in tables/appendix. Quotes must carry the claim, not illustrate a conclusion reached elsewhere.
• Evidence for each construct. Every construct backed by patterned evidence across informants/cases, with prevalence where appropriate.
• Negative cases. Report disconfirming instances and how they refined the theory — central to trustworthiness at OS.
• Process display. For process theory, show the temporal/event structure (timeline, phase model, visual mapping); make the transitions between phases analytically explicit, not just narrated.

Branch B — Process analysis (when the contribution is a process model)

• Choose a process strategy explicitly: narrative, temporal bracketing, visual mapping, grounded theory, or alternate templates (Langley). Say why it fits.
• Identify events, sequences, and turning points; show what triggers each transition and what each phase accomplishes that the prior could not.
• Distinguish real-time from retrospective data and address the recall/hindsight risks of each.
• The output is a process model figure plus the analytic account that earns it.

Branch C — Quantitative analysis

• Main models match the design (FE/RE, event-history, multilevel, network); standard errors clustered at the right level.
• Robustness that targets the theory's threats — alternative measures, samples, specifications, endogeneity checks, modern staggered-DiD diagnostics if relevant — not a wall of tables that never address the real threat.
• Mechanism evidence. Don't stop at the reduced-form relationship; probe why (mediation/moderation or supplementary tests).
• Effect interpretation in organizational terms — magnitudes, not just significance.

Either branch — the "so what" of the evidence

• Tie every analytic result back to the mechanism and the theoretical puzzle.
• Distinguish what the data can and cannot establish — overclaiming is a fast OS rejection.
• Prepare exhibits jointly with orgstud-tables-figures.

Execution bridge (StatsPAI / Stata MCP)

Run the battery, don't just enumerate it. Full map:

[execution-with-mcp](../../../shared-resources/empirical-methods/execution-with-mcp.md). Organization Studies is largely qualitative/theoretical; use the chain below only for its quantitative-empirical papers, and say so when a study is interpretive.

• 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](../../../Journal-of-Finance-Skills/resources/worked-examples/02-execution-walkthrough.md).

Checklist

• [ ] Qual: data structure (first-order → second-order → dimensions) documented
• [ ] Qual: a data-to-theory / evidence table built; quotes carry (not decorate) claims
• [ ] Qual: negative cases reported and used to refine the theory
• [ ] Process: process strategy named; turning points and transitions made explicit
• [ ] Quant: SEs clustered appropriately; robustness targets the theory's threats; magnitudes interpreted
• [ ] Mechanism is probed, not just the headline relationship
• [ ] Claims are matched to what the evidence can actually support

Anti-patterns

• "Anecdotal" qualitative work: cherry-picked quotes with no coding transparency
• Quotes that illustrate a pre-set conclusion rather than generating/supporting it
• A process "model" that is really a narrative with no analytic structure or transition logic
• Robustness theater: many tables that never address the real identification threat
• Reporting significance with no interpretation of organizational magnitude
• Overclaiming causality or generalizability beyond what the design supports

Output format

【Branch】qualitative / process / quantitative
【Data-to-theory link】data structure / process strategy / mechanism tests done
【Key evidence】power quotes, the process model, or main estimates
【Trustworthiness/robustness】checks completed + gaps (negative cases, clustering, alt explanations)
【What evidence cannot show】explicit limits
【Next skill】orgstud-contribution-framing