Research Intelligence & Statistical Analysis Advanced

1

Research Setup

Basic information used to personalise all generated prompts.

Research title / topic

Data type

Statistical software

AI assistant

Research field

Study context

2

Define Variables

Add the variables used in your study. Codes should match the names used in your statistical software.

Dependent Variables

Name	Code	Measurement / formula

Independent Variables

Name	Code	Measurement / formula

Add any independent variables used in the study. The generic model filters below can estimate them one at a time or jointly without requiring special variable labels.

Mediators / Moderators

Role	Name	Code	Measurement / formula

Control Variables

Name	Code	Measurement / formula

3

Hypotheses

Enter your existing hypotheses or let the form structure them from the variables.

Auto-update hypotheses when variables changeNew IV–DV relationships are added automatically. Turn this off before making extensive manual edits.

No.	Hypothesis statement	Expected direction

4

Data Structure

These details help the generated prompt describe the dataset correctly.

Panel / entity identifier

Time variable

Observation level

Start year

End year

Panel balance

5

Data File & Working Folder

Tell the generated AI prompt how Stata should open the dataset, create logs, and save output files.

Stata data workflow

Working folder / project path

Existing Stata dataset

Exact .dta file name

Use this file automatically in each analysis block?

Stata log & output management

Log policy

Log file prefix

Where should exported results be saved?

Recommended Stata workflow: set the working directory with cd, safely close any open log, open the required log, load the required .dta file, run the stage commands, save/export requested results to the selected folder, then close the log.

File and log automation in Version 1.1 is currently designed for Stata. If EViews is selected, Stata-specific cd, use, and log instructions are omitted.

6

Analysis Plan

Select what the AI prompt should guide the researcher to perform in the statistical software.

A. Preliminary Analysis

Declare data structurePanel/entity and time settings.

Descriptive statisticsMean, median, SD, min, max, percentiles and observations.

Skewness & kurtosisDistribution shape and potential data issues.

Correlation matrixPairwise correlations with significance and observations.

VIFAssess multicollinearity.

Data treatment guidanceTransform, winsorise or trim only when justified.

B. Parametric Tests

One-sample t-testCompare a sample mean with a benchmark.

Independent t-testCompare means across two independent groups.

Paired t-testCompare two related measurements.

One-way ANOVACompare means across three or more groups.

Two-way ANOVATwo categorical factors and interaction.

Welch alternativeUse when equal-variance assumptions are inappropriate.

Parametric test setup

Grouping variable (optional)

Test / benchmark value (optional)

C. Panel Regression & Model Selection

Pooled OLSBaseline pooled specification.

Fixed EffectsEntity fixed effects specification.

Random EffectsRandom effects specification.

FE vs RE selectionHausman-type comparison when technically appropriate.

PCSEOptimized PCSE sensitivity variants selected below.

Difference GMMDynamic panel Difference GMM estimator variants selected below.

System GMMDynamic panel System GMM estimator variants selected below.

D. Generic Model Specification Filter New

Individual DirectEach IV separately + controls only. Exclude all variables designated as Moderator.

Individual ModerationFor each IV: IV + moderator + IV×moderator + all selected controls.

Joint DirectAll IVs together + controls only. Exclude all variables designated as Moderator.

Joint ModerationAll IVs + moderator + each IV×moderator + all selected controls.

Generic by design: the independent variables are defined only in the Variables section above. These filters control how those variables are combined in models and do not assume ESG or any other subject-specific variable names.

E. Optimized PCSE Options

Pairwise + IndependentLightweight independent-error PCSE.

Pairwise + AR(1)Common AR(1) correction.

Independent equation typeDefault/casewise handling with independent serial structure.

Panel-specific AR(1)Optional; avoid automatically in very large N/small T panels.

F. GMM Method Filters New

Difference GMM — 1-Step Non-RobustClassical one-step Difference GMM.

Difference GMM — 1-Step RobustOne-step Difference GMM with robust inference.

Difference GMM — 2-Step Non-RobustReport warnings on uncorrected two-step SEs.

Difference GMM — 2-Step RobustTwo-step Difference GMM with corrected robust SEs.

System GMM — 1-Step Non-RobustOne-step System GMM.

System GMM — 1-Step RobustOne-step System GMM with robust inference.

System GMM — 2-Step Non-RobustTwo-step System GMM; flag uncorrected SE warnings.

System GMM — 2-Step RobustTwo-step System GMM with corrected robust SEs.

Difference GMM transformation and instrument-grid options

Forward Orthogonal DeviationsUse noleveleq orthogonal.

First DifferencesUse noleveleq without orthogonal.

CollapseParsimonious instrument matrix.

No CollapseEstimate but screen carefully for proliferation.

Macro in GMM-StyleTest selected macro controls inside GMM blocks.

Macro in IV-StyleRemove macro controls from GMM blocks and place in ivstyle.

Focal IV with Lagged DVTest each IV one at a time beside L.DV.

Separate Interaction InstrumentsUse dedicated lags for moderation interactions.

GMM classification, lag grid and qualification rules

Endogenous variable codes

Predetermined variable codes

Exogenous / standard-IV variable codes

Macro control codes

Lagged-DV lag windows

Main IV lag windows

Interaction lag windows

Other-control lag windows

Maximum instruments

Minimum significant relationships (%)

Coefficient significance cut-off

Best models per relationship

Significance-percentage denominator

Default GMM qualification: AR(2) p > 0.05, Hansen p > 0.05 wherever Hansen is reported, and at least the selected percentage of substantive relationships significant at the selected cut-off. Also report AR(1), Sargan, Difference-in-Hansen, Wald p, instrument count and warnings. The full selected grid must be completed before model ranking.

G. Heterogeneity Analysis

Subgroup regressionsEstimate models separately by selected groups.

Interaction-based heterogeneityFormal interaction test of group differences.

Period-based analysisCompare effects across time periods.

Industry heterogeneityCompare effects by industry group.

Country / region heterogeneityCompare effects by country or region categories.

Size-based heterogeneityCompare smaller vs larger entities.

Heterogeneity setup

Grouping variable

Group labels / split rule

H. Standard Error Treatment

Conventional

Use standard model-based errors where appropriate.

Robust

Request heteroskedasticity-robust inference.

Cluster-robust

Cluster by the entity/panel identifier.

I. Results & Table Output Filters

Complete ScreeningKeep accepted and rejected specifications.

Qualifying ModelsApply the selected diagnostic and significance rules.

Best ModelsRank the requested number of defensible models.

Cross-Estimator TableCompare OLS, FE, RE, PCSE and GMM.

Excel-Ready TablesUse coefficient*** [p-value] in one cell.

Diagnostics FooterShow estimator-appropriate diagnostics below each model.

Significance levels shown in tables

Results table preference

7

Research Framework

Create one of three common conceptual-framework diagrams using the variables already entered.

Framework type

Diagram labels

Include control variables

8

Generated Research Workflow

Choose the analysis options above, then copy a clear software-ready prompt into ChatGPT, Gemini, Claude, Copilot or another capable AI assistant. The site itself does not fabricate results or analyse your dataset.

Complete the form and select “Generate Research Workflow”.

Methodological safeguard: generated prompts tell the AI to verify assumptions, use the selected software, avoid inventing values, and ask for actual output before interpreting empirical results.

i

About RISA Advanced — Direct/Moderation Rule Updated

A general secondary-data research workflow prompt builder with configurable OLS, FE, RE, PCSE and expanded GMM methods, diagnostic screening, generic model filters, and framework diagrams.

Google login setup: the page includes a ready hook for Firebase Authentication, but you must paste your own Firebase web-app configuration into the script before Google sign-in can work on your deployed domain. Until then, the button clearly shows that authentication is not configured.

Build advanced statistical analysis prompts for secondary-data research.

Research Setup

Define Variables

Hypotheses

Data Structure

Data File & Working Folder

Existing Stata dataset

Excel → Stata conversion

Analysis Plan

Parametric test setup

Difference GMM transformation and instrument-grid options

GMM classification, lag grid and qualification rules

Heterogeneity setup

Research Framework

Generated Research Workflow

About RISA Advanced — Direct/Moderation Rule Updated