New tab created; syncing the analysis framework and task status.
History cannot be saved in guest mode.
Upload Data
0%
No file selected
Descriptive Statistics
Select variables for descriptive statistics
Export Results
Descriptive Statistics
Select a categorical variable for grouping
Select numeric variables for descriptive statistics
Mean and standard deviation are output by default; other statistics are optional
Export Results
Frequency Statistics Configuration
Select variables for frequency statistics (numeric and categorical supported)
Export Results
Correlation Analysis Configuration
Select variables for correlation coefficients (at least 2)
The correlation matrix shows Pearson correlation coefficients with significance levels: *** p<0.01, ** p<0.05, * p<0.1
Export Results
VIF Multicollinearity Test
Select independent variables to test for multicollinearity (at least 2)
VIF criteria:
• VIF < 5: no collinearity problem
• 5 ≤ VIF < 10: moderate collinearity
• VIF ≥ 10: severe collinearity; consider dropping the variable
Export Results
Normality Test Configuration
Each variable is tested on its own non-missing sample
Notes:
• Null hypothesis H0: the variable follows a normal distribution
• Output includes skewness, kurtosis, and joint chi-square tests
• Each variable requires at least 8 valid observations
Export Results
t-Test Configuration
Choose the t-test type that fits your data
Select variables for the t-test
Enter the population mean to compare against
Select the significance level (used for interpretation)
Select one-tailed or two-tailed
Select the grouping variable (should be binary)
Select the significance level (used for interpretation)
Test whether the two groups have equal variances
The paired-samples t-test requires exactly two variables
Select the significance level (used for interpretation)
Select one-tailed or two-tailed
Export Results
Chi-Square Test Configuration
Choose the chi-square test type that fits your data
Select variables for the chi-square test
The independence test requires exactly two categorical variables
Select the significance level (used for interpretation)
Show the contingency table and expected frequencies in the results
The goodness-of-fit test requires one categorical variable
Select the significance level (used for interpretation)
Enter the expected frequency distribution, comma-separated. Leave empty for a uniform distribution
Export Results
ANOVA Configuration
Choose the ANOVA type that fits your data
Select a categorical variable for grouping (e.g., teaching method, gender)
Select numeric variables to analyze (multi-select)
Select the significance level (used for interpretation)
Test whether group variances are equal
When the result is significant, run multiple comparisons to identify which groups differ
Select the post-hoc test method
Select a numeric variable to analyze
Select the first categorical variable (e.g., teaching method)
Select the second categorical variable (e.g., gender)
Select the significance level (used for interpretation)
Test whether group variances are equal
Export Results
Regression Models
Panel Structure Settings
Select the dimensions to fix
Performance note:Fixed-effects Logit/Probit generates many dummy variables and can be slow. With a large number of individuals (e.g., 1000+), consider a linear probability model (OLS with FE).
Threshold model settings
Coefficients on X vary across threshold regimes, while control coefficients remain constant. The threshold variable may also be a regressor. Entity fixed effects are included automatically.
Use one value for all stages or separate stage-specific values with spaces. Enter 0 to skip a test; up to 2,000 replications per stage.
For example, 0.05, or 0.05 0.05 for two thresholds.
Advanced options
Mediation Settings
The page shows path regression results and indirect effect decomposition; with Bootstrap selected, it also shows resampling intervals for indirect, direct, and total effects.
Moderation Settings
The interaction term is generated automatically and its regression results are shown; without main effects, only the interaction term and controls remain in the main regression.
Heterogeneity Analysis
With too many categories, consider splitting at the median or a custom threshold to avoid an overly wide table or tiny groups.
Chain Mediation Settings
Chain mediation estimates the path X → Mediator 1 → Mediator 2 → Y by default and decomposes effects via single mediators and the serial chain.
Use “lower|upper” for ranges and commas for sets. This version supports flat AND/OR conditions, not nested groups.
If none are selected, numeric variables among Y, X, and the controls are used automatically.
The system creates collision-free categories from dimension combinations. If RE/Pooled has too many categories, it will suggest FE instead.
Marginal Effects Settings
AME: average marginal effects over all observations (recommended) | MEM: marginal effects computed at variable means | Median: marginal effects computed at variable medians
Leave blank to use the automatic note; entered text fully replaces it.
Bootstrap Progress
0 / 0
Waiting for the task to start...
Export Results
Raw Regression Results
Endogeneity & Causality Configuration
2SLS suits cases with clear exclusion-restriction instruments; GMM suits settings that call for more general moment conditions and robust weighting matrices; the Heckman two-step suits selection-bias scenarios where the dependent variable is observed only for part of the sample.
Entropy Balancing Settings
In automatic balancing mode, select the 1st/2nd/3rd moment for each covariate; in single-group mode, enter the target mean for each covariate. Custom covariate expressions are not supported here—generate interaction terms on the Data Processing page first if needed.
Instruments & Identification
e.g., firm ID, year, region. Fixed effects are implemented as dummy variables, so computation slows down with very high cardinality.
By default, IV-GMM uses a weighting matrix consistent with your standard error choice: `IID -> unadjusted`, `Robust -> robust`, `Cluster -> clustered`.
Dynamic Panel GMM Settings
Complete the two common steps in order to start the analysis.
Recommended Preset
By default, the system uses system GMM, two-step estimation, robust (Windmeijer) correction, and instrument collapse, and automatically reports AR(1), AR(2), Sargan, and Hansen tests. Most analyses only require selecting the panel individual and time variables and confirming the endogeneity of the regressors.
1 Set Panel Structure
2 Set Dynamics & Instruments
The dependent variable automatically uses its own lags as internal instruments; X variables not selected here are treated as exogenous by default.
Reduces the risk of an inflated Hansen test from too many instruments.
Automatically applies the finite-sample (Windmeijer) SE correction in two-step estimation.
Advanced OptionsTime structure, transformation, clustering, custom IV groups, and compatibility options
0 for contemporaneous, positive for lags, negative for leads.
Multi-select; if none selected, the robust SE setting above is used.
Enable only when different variables need different lag windows or equation ranges; once enabled, it replaces the recommended IV scheme above.
Internal IV Groups
Exogenous IV Groups
Other Estimation & Output Options
Selection Equation
1 means the dependent variable is observed; 0 means unobserved.
Select at least one variable that does not directly enter the outcome equation but affects the probability of observation.
Leave blank to use the automatic note; entered text fully replaces it.
Export Results
Raw Estimation Output
DID Diagnostics Configuration
The DID baseline regression can be done in the Fixed Effects module; this page provides parallel trends, placebo, and PSM matching diagnostics.
Single-Period Group Trends Settings
For policies occurring at a single point in time. Only plots outcome-level trends for the treatment and control groups; no event-study coefficients are estimated.
Policy Timing Settings
Pass the actual policy year or policy-period variable directly to avoid mistaking the first year post == 1 for the reform period.
Enter directly when all treated units share the same policy time.
When no policy_year is selected, the system infers the policy time from the first period this variable equals 1 for each individual.
Only used for DID-signal auto-detection of policy_year; not involved when policy_year is entered or selected manually.
Event Study Settings
Select the dimensions to absorb per your design, e.g., industry, year, province, firm.
Chart Settings
Placebo Settings
The placebo test no longer defaults to entity and time fixed effects; select them per your design, e.g., firm, year, industry, region.
PSM Matching Settings
Chart Output Selection
All three are selected by default; unchecked plots are neither generated nor exported.
Post-Matching Regression Settings
Select the key explanatory variables that actually enter the post-matching regression, e.g., the DID interaction, policy variable, or treatment variable.
Regression controls may be the same as or different from the matching covariates above.
For a `reghdfe`-style regression, select the fixed effects dimensions to absorb here.
Entropy Balancing Settings
Post-Entropy-Balancing Regression Settings
Select the key explanatory variables that actually enter the weighted regression, e.g., the DID interaction, policy variable, or treatment variable.
Regression controls may be the same as or different from the matching covariates above.
For a reghdfe weighted regression, select the fixed effects dimensions to absorb here.
Leave blank to use the automatic note; entered text fully replaces it.
Placebo Progress
0 / 0
Waiting for the task to start...
Export Results
DID Raw Output
Model Diagnostics Configuration
Panel Structure Settings
Hausman Test Options
This option resolves the non-positive-definite covariance matrix issue and produces results consistent with Stata.
When disabled, the raw covariance matrix is used and you may see the "V_b-V_B is not positive definite" warning.
Export Results
Stationarity Test
Variable that identifies the time series order
Auto-difference when non-stationary
Runs the ADF test (and optionally the PP test) on each variable under the selected regression type. Non-stationary series are automatically differenced and retested.
Used for information criterion selection
Used to determine the optimal lag order
Non-stationary series must be differenced first
VAR requires at least 2 analysis variables. The optimal lag order is selected automatically by information criteria, and coefficient tables for each equation are output.
When to difference? VAR requires stationary series. If variables have a unit root (non-stationary), difference them first.
Use the Stationarity Test first to check whether variables are stationary; if not, choose the corresponding difference order.
Used for information criterion selection
Used to determine the optimal lag order
Non-stationary series must be differenced first
The Granger causality test checks whether one variable's lags help predict another. At least 2 variables are required; all pairs are tested in both directions.
When to difference? The Granger causality test requires stationary series. If variables have a unit root (non-stationary), difference them first.
Use the Stationarity Test first to check whether variables are stationary; if not, choose the corresponding difference order.
Select the cointegration test method
Used by the Johansen test
VAR model lag order
The cointegration test checks whether multiple non-stationary variables share a long-run equilibrium. At least 2 variables are required.
Method selection advice:
The Engle-Granger two-step suits cointegration between two variables—simple and intuitive;
the Johansen test suits multivariate systems and can detect multiple cointegrating relationships.
With more than 2 variables, use the Johansen test or run both methods.
Consistent with Stata's lags()
Determined by the cointegration test
Keep consistent with the cointegration test
VECM adds cointegration constraints on top of VAR, suitable for non-stationary variables with cointegration. At least 2 variables are required.
Usage advice:
First determine the cointegration rank (r) via the Cointegration Test, then enter it here. Lag order and trend settings should match those used in the cointegration test.
Used for information criterion selection
Used to determine the optimal lag order
Non-stationary series must be differenced first
Checks whether all eigenvalues of the VAR companion matrix lie inside the unit circle (modulus < 1). If all are inside, the model is stable; otherwise it is unstable and you need to adjust the lag order or the data.
Usage advice:
Run this test after fitting a VAR model. Parameter settings (lag order, trend term, difference order) should match the VAR model.
The variable whose forecast error sources are analyzed
Select analysis variables above and choose a response variable first
Cholesky ordering: the most exogenous variable comes first, and the response variable is placed last automatically. Impulse variables appear after selecting a response variable; drag to reorder.
Used for information criterion selection
Non-stationary series must be differenced first
Number of forecast steps for variance decomposition
Variance decomposition is based on Cholesky orthogonalization, and the ordering of impulse variables affects the results. Order them by theoretically expected causality (most exogenous first).
Usage advice:
Parameter settings should match the VAR model. Output is in percent: each impulse variable's share of the response variable's forecast error variance.
Actual years used: pending validation
GDP reference coverage: 2000–2024
Cities and years are aligned together. For a 2015–2024 sample, economic matrices use only those 10 years of per-capita GDP; 2000–2024 indicates reference-data coverage only.
Matrix alignment has not been validated yet.
Complete Entity Mapping Manually
Configure Variables
When using SDM, spatially lagged WX terms are generated automatically for all explanatory and control variables; no additional selection is needed.
Model and Inference Settings
Advanced OptionsEffect standard errors, simulation draws, and random seed
Each run adds one set of test rows. Use “Append to Current Table” below to combine multiple diagnostics.
Enter one period, comma-separated periods, or a continuous year range; enter "all" to analyze every period in the sample.
By default, only period summaries are shown. Select this option to output each entity's Local Moran statistic, p-value, and quadrant.
Leave blank to use the automatic note; entered text fully replaces it.
Export Results
Raw Spatial Econometric Output
Statistical analysis setup
Dimensions and items
Sequential variable blocks
Paired variables
Structural paths
Analysis settings
Result output
Analysis scope
Factor relationships
Result output
Follow-up tests
Follow-up tests
Result output
Result output
Variable processing
Result output
Result output
Analyzing
Export Results
Raw statistical output
Data Preview (first 10 rows)
Batch Export Analysis Results
Select Analysis Tabs
0
Adjust Export Order
0
File Format
Export Mode
Include written interpretations
STATAU
Cloud-Based Econometrics Platform
Settings
The page will reload to apply the new language after you confirm