Python Tools for Management Research

3b: Code-Generated Tables, Figures, and Results

Jason T. Kiley

From rendered paper to reproducible results

The next step is making the paper generate its own numbers, tables, figures, and model results.

This is an elegant form of reproducible research: the paper is a recipe for the results.

Generated results

Why generate results?

Copying results by hand creates quiet failure points:

• sample sizes get out of sync;
• tables lag behind data cleaning changes;
• figures are regenerated but not replaced;
• text reports one model while a table reports another.

Reviewers notice this, and you really don’t want that at the margin.

What code can write for us

Some useful examples:

Inline values

Sample size, coefficient estimates, p values, and other single numbers in the text.

Tables

Descriptives, correlations, model summaries, and robustness checks.

Figures

Scatter matrices, distributions, diagnostics, and coefficient plots.

In-text values

The sample includes `{python} n` observations.

The n value in the sentence comes from Python. If the data changes and we render again, the sentence updates.

Code can stay out of the way

#| echo: false
#| warning: false
#| message: false

# setup code here

Some code is part of the paper’s machinery rather than part of the visible paper. These settings turn off code, warnings, and messages in the rendered paper, but the code still runs.

Results workflow

Add the analysis tools

For this segment, the project needs a few more packages:

• maketables for descriptive, correlation, and regression tables;
• matplotlib for static figures that work well in Word and PDF;
• statsmodels for regression models and model summaries.

We add them to pyproject.toml because they are now part of the project recipe.

Tables

descriptives
correlations
regression_table

The table objects are built in code, then displayed in the paper.

Figures

axes = scatter_matrix(results_data[analysis_vars])
fig = axes[0, 0].figure
fig

Whenever we use some of the more interesting interactive visualizations, we need to make a static version for the paper.

In-text model results

Attention is associated with performance
(b = `{python} round(attention_coef, 2)`,
p = `{python} f"{attention_p:.3f}"`).

This is the same fitted model used for the regression table.

Hands-on

Open the 3b activity page.