universal-output-hub 0.2.0

Collect model outputs, tables, diagnostics, and graphs into reproducible research-output bundles.

Universal Output Hub

universal-output-hub is a lightweight Python reporting layer for collecting model results, regression tables, statistical tables, diagnostics, notes, and graphs into one reproducible output bundle.

It is designed as a broader Python alternative to Stata-style reporting tools such as outreg2, esttab, and asdoc, but with a wider scope: not only regression tables, but also ordinary tables, figures, metadata, diagnostics, table notes, and complete report exports.

The core idea is simple:

Run your models anywhere. Collect your outputs once. Export them everywhere.

---

Purpose

Statistical and applied research workflows often produce outputs from many sources:

• Python models from statsmodels, linearmodels, pyfixest, or custom estimators;
• external statistical software such as Stata, R, MATLAB, SPSS, EViews, SAS, Excel, or CSV exports;
• regression tables, descriptive statistics, robustness checks, diagnostics, and plots;
• final outputs needed in Excel, LaTeX, HTML, Word, PDF, Markdown, CSV, or JSON.

universal-output-hub provides one consistent interface for collecting those outputs and exporting them into publication-ready and review-friendly formats.

---

Core capabilities

Capability	Status
Add Python model results	Yes
Add custom model dictionaries	Yes
Add coefficient tables from external software	Yes
Add ordinary pandas tables	Yes
Add CSV, Excel, Stata .dta, TSV, and Parquet tables	Yes
Add graph/image files	Yes
Save matplotlib-style figures	Yes
Build publication-style regression tables	Yes
Put models side-by-side in columns	Yes
Show coefficient estimates	Yes
Show standard errors below coefficients	Yes
Add configurable significance stars	Yes
Add table notes below model tables	Yes
Store diagnostics and model statistics	Yes
Export CSV	Yes
Export Excel .xlsx	Yes
Export HTML	Yes
Export Markdown	Yes
Export LaTeX .tex tables and reports	Yes
Export JSON	Yes
Export Word .docx reports	Yes
Export PDF tables and reports	Yes
Export combined Excel workbook	Yes
Create reproducible output bundles	Yes

---

Outreg-style reporting

universal-output-hub supports outreg2-style model reporting in Python.

That means it can:

Question	Answer
Export regression/model results like outreg2?	Yes, broadly
Put models side-by-side in columns?	Yes
Show coefficient estimates?	Yes
Show standard errors below coefficients?	Yes
Add significance stars?	Yes
Customize star thresholds?	Yes
Include diagnostics/statistics like N, R², Hansen p, Sargan p, AR tests, instruments?	Yes
Add notes under the table?	Yes
Export to Excel?	Yes
Export to LaTeX?	Yes
Export to PDF?	Yes
Export to Word/DOCX?	Yes
Export to HTML, Markdown, CSV, JSON?	Yes
Accept statsmodels results?	Yes
Accept custom model dictionaries?	Yes
Accept Stata/R/MATLAB/SPSS/EViews outputs?	Yes, if exported as structured coefficient tables

This package is not a full clone of Stata’s outreg2. It provides outreg2-style reporting for Python and external coefficient-table workflows.

---

Current limitations and roadmap

Capability	Status
Native MATLAB .mat model-object parser	Planned
Native Stata .ster parser	Planned
Native R .rds model-object parser	Planned
Exact Stata outreg2 command compatibility	Planned
Full esttab-level formatting grammar	Planned
Multi-equation models with separate panels	Planned
Automatic fixed-effect yes/no row detection from native model objects	Planned
Journal templates	Planned

External software compatibility currently works through structured coefficient tables with at least a term column and a coefficient column. Standard errors, p-values, statistics, diagnostics, and notes can also be supplied.

---

Installation

Install from PyPI:

bash python -m pip install universal-output-hub

Install from GitHub:

python -m pip install git+https://github.com/Akanom/universal-output-hub.git

Development installation:

python -m pip install -e ".[dev,examples]"
pytest -q

Windows PowerShell:

cd "<UNIVERSAL_OUTPUT_HUB_REPO>"

py -3.13 -m venv .venv
.\.venv\Scripts\Activate.ps1

python -m pip install --upgrade pip setuptools wheel
python -m pip install -e ".[dev,examples]"
pytest -q

---

Basic example

from universal_output_hub import OutputHub

hub = OutputHub(
    "Model Output Report",
    metadata={
        "project": "Panel-data model comparison",
        "software": "Python / Stata / R compatible",
    },
)

hub.add_model(
    {
        "name": "System GMM",
        "depvar": "growth_rate",
        "params": {
            "L.growth_rate": 0.31,
            "lPA": 0.08,
            "s_techshare": 0.16,
            "s_tech_frag_polity": -0.022,
        },
        "std_errors": {
            "L.growth_rate": 0.12,
            "lPA": 0.03,
            "s_techshare": 0.07,
            "s_tech_frag_polity": 0.012,
        },
        "pvalues": {
            "L.growth_rate": 0.011,
            "lPA": 0.007,
            "s_techshare": 0.041,
            "s_tech_frag_polity": 0.078,
        },
        "statistics": {
            "N": 946,
            "Instruments": 42,
        },
        "diagnostics": {
            "AR(1) p": 0.085,
            "AR(2) p": 0.316,
            "Hansen p": 0.190,
            "Sargan p": 0.098,
            "Diff-Hansen p": 0.089,
        },
    }
)

hub.add_table_note("Standard errors are reported in parentheses.")
hub.add_table_note("Hansen, Sargan, and AR tests are reported as p-values.")

hub.export_bundle("outputs/model_output")

---

Output bundle structure

outputs/model_output/
├── index.html
├── manifest.json
├── regression_tables/
│   ├── regression_table.csv
│   ├── regression_table.xlsx
│   ├── regression_table.html
│   ├── regression_table.md
│   ├── regression_table.tex
│   ├── regression_table.pdf
│   └── regression_table.json
├── tables/
├── figures/
├── workbooks/
│   └── output_workbook.xlsx
└── reports/
    ├── output_report.docx
    ├── output_report.pdf
    └── output_report.tex

The default output filenames are generic:

output_report.docx
output_report.pdf
output_report.tex
output_workbook.xlsx

The package does not assume that the project is a thesis, dissertation, academic paper, business report, or internal analysis.

---

Regression tables

Export regression/model tables to multiple formats:

hub.export_regression_table("outputs/regression_table.xlsx")
hub.export_regression_table("outputs/regression_table.tex")
hub.export_regression_table("outputs/regression_table.pdf")
hub.export_regression_table("outputs/regression_table.html")
hub.export_regression_table("outputs/regression_table.md")
hub.export_regression_table("outputs/regression_table.csv")
hub.export_regression_table("outputs/regression_table.json")

A typical table includes:

                         System GMM
L.growth_rate             0.310**
                          (0.120)

lPA                       0.080***
                          (0.030)

N                         946
Instruments               42
AR(1) p                   0.085
AR(2) p                   0.316
Hansen p                  0.190
Sargan p                  0.098
Diff-Hansen p             0.089

Significance              *** p≤0.01, ** p≤0.05, * p≤0.10
Notes                     Standard errors are reported in parentheses.
                          Hansen, Sargan, and AR tests are reported as p-values.

---

Significance stars

Stars are enabled by default:

*** p≤0.01
**  p≤0.05
*   p≤0.10

Disable stars:

hub.export_regression_table(
    "outputs/regression_table.tex",
    stars=False,
)

Use custom thresholds:

hub.export_regression_table(
    "outputs/regression_table.tex",
    star_levels={"***": 0.001, "**": 0.01, "*": 0.05},
)

---

Table notes

Use table notes for outreg2-style notes under regression/model tables:

hub.add_table_note("Standard errors in parentheses.")
hub.add_table_note("Country and year fixed effects included.")
hub.add_table_note("Hansen, Sargan, and AR tests are reported as p-values.")

Table notes are included in:

• regression/model tables;
• Excel exports;
• LaTeX exports;
• PDF exports;
• DOCX reports;
• PDF reports;
• LaTeX reports;
• manifest files.

You can also pass notes directly when building a regression table:

table = hub.regression_table(
    table_notes=[
        "Standard errors in parentheses.",
        "Country and year fixed effects included.",
    ]
)

---

Diagnostics and model statistics

Diagnostics and statistics can be supplied directly:

hub.add_model(
    {
        "name": "System GMM",
        "params": {"L.y": 0.42, "x": 0.11},
        "std_errors": {"L.y": 0.10, "x": 0.04},
        "pvalues": {"L.y": 0.001, "x": 0.030},
        "statistics": {
            "N": 946,
            "Instruments": 42,
        },
        "diagnostics": {
            "AR(1) p": 0.085,
            "AR(2) p": 0.316,
            "Hansen p": 0.190,
            "Sargan p": 0.098,
            "Diff-Hansen p": 0.089,
        },
    }
)

Supported diagnostic/statistic rows include, among others:

N
R2
Adj. R2
Within R2
Overall R2
AIC
BIC
Log Likelihood
F statistic
F p-value
AR(1) p
AR(2) p
Hansen p
Sargan p
Diff-Hansen p
Instruments
Entity FE
Time FE
Fixed effects
Clustered SE

The package exports diagnostics. It does not compute Hansen, Sargan, AR tests, or other estimator-specific tests by itself unless a backend or model object supplies them.

---

External model outputs

External statistical software can be used by exporting coefficient tables into a structured format.

Minimum structure:

term	coef	se	pvalue
lPA	0.080	0.022	0.001
s_techshare	0.161	0.071	0.024

Then import:

hub.add_model_file(
    "stata_fe_results.csv",
    name="Stata FE",
    term_col="term",
    coef_col="coef",
    se_col="se",
    pvalue_col="pvalue",
    statistics={"N": 885, "Within R2": 0.21},
    diagnostics={"Clustered SE": "country"},
    source="Stata export",
)

Supported model/table file formats:

.csv
.tsv
.txt
.xlsx
.xls
.dta
.parquet
.pq

---

Embedded diagnostics from imported regression tables

Some tools export diagnostics as rows below the coefficients. For example:

term,coef,se,pvalue
L.y,0.42,0.10,0.001
x,0.11,0.04,0.030
N,946,,
Instruments,42,,
AR(1) p,0.085,,
AR(2) p,0.316,,
Hansen p,0.190,,
Sargan p,0.098,,
Diff-Hansen p,0.089,,
Entity FE,Yes,,
Time FE,Yes,,

The package can treat these as table statistics/diagnostics rather than coefficients when the embedded-diagnostics extractor is enabled in the adapter layer.

This is useful for Stata/R/MATLAB/SPSS/EViews-style result exports.

---

Python models

import statsmodels.api as sm
from universal_output_hub import OutputHub

hub = OutputHub("Python Model Output")

fit = sm.OLS(y, X).fit()
hub.add_model(fit, name="OLS", adapter="statsmodels")

hub.export_bundle("outputs/python_models")

The package extracts common model information such as coefficients, standard errors, p-values, N, R², AIC, BIC, log likelihood, and related statistics when available.

---

Custom model dictionaries

Custom estimators can be passed directly:

hub.add_model(
    {
        "name": "Custom Estimator",
        "depvar": "y",
        "params": {
            "x1": 1.25,
            "x2": -0.40,
        },
        "std_errors": {
            "x1": 0.20,
            "x2": 0.15,
        },
        "pvalues": {
            "x1": 0.004,
            "x2": 0.080,
        },
        "statistics": {
            "N": 100,
            "R2": 0.45,
        },
        "diagnostics": {
            "Hansen p": 0.31,
            "AR(2) p": 0.42,
        },
    }
)

This is the recommended interface for custom econometric packages, including dynamic-panel GMM workflows.

---

Ordinary tables

Add descriptive statistics, robustness checks, correlations, or any pandas-compatible table:

hub.add_table("Descriptive statistics", df.describe())
hub.add_table("Correlation matrix", df.corr())
hub.add_table("Robustness summary", robustness_df)

hub.export_tables(
    "outputs/tables",
    formats=("csv", "xlsx", "html", "md", "tex", "pdf", "json"),
)

---

Figures and graphs

Add an existing graph file:

hub.add_figure(
    "Marginal effects plot",
    path="figures/marginal_effects.png",
    caption="Estimated marginal effects across institutional quality.",
)

Save a matplotlib-style figure directly:

hub.add_figure(
    "Model comparison",
    fig=fig,
    output_dir="outputs/figures",
    filename="model_comparison.png",
)

Figures are copied into reproducible output bundles and included in full reports where supported.

---

Complete reports

Export full reports:

hub.export_report("outputs/output_report.docx")
hub.export_report("outputs/output_report.pdf")
hub.export_report("outputs/output_report.tex")
hub.export_excel_workbook("outputs/output_workbook.xlsx")

With a bundle:

hub.export_bundle(
    "outputs/model_output",
    report_filename="output_report",
    workbook_filename="output_workbook",
)

Use custom generic names:

hub.export_bundle(
    "outputs/project_run_001",
    report_filename="model_output_report",
    workbook_filename="model_output_workbook",
)

---

Recommended workflow

from universal_output_hub import OutputHub

hub = OutputHub("Model Output Report")

hub.add_model(model_1, name="OLS")
hub.add_model(model_2, name="Fixed Effects")
hub.add_model(model_3, name="System GMM")

hub.add_table_note("Standard errors in parentheses.")
hub.add_table_note("Country and year fixed effects included.")

hub.add_table("Descriptive statistics", descriptives)
hub.add_table("Robustness checks", robustness)

hub.add_figure("Coefficient plot", path="figures/coefplot.png")

hub.export_bundle("outputs/run_001")

---

Design philosophy

1. Model-agnostic: accept Python models, custom dictionaries, and structured external coefficient tables.
2. Research-friendly: support regression tables, diagnostics, notes, tables, figures, and reproducible bundles.
3. Format-flexible: export to Excel, LaTeX, PDF, Word, HTML, Markdown, CSV, and JSON.
4. Transparent: avoid pretending to be a native parser for every statistical software object.
5. Simple API: collect outputs once and export them everywhere.

---

Project status

Current release: 0.2.0

This is an early-stage package. The public API is usable, but some internals may still change as the package matures.

---

License

MIT License.