mlogitviz 0.1.3

A library to compute and visualize marginal effects for multinomial logistic regression models

mlogitviz

mlogitviz is a Python package designed to visualize changes in predicted probabilities from multinomial logistic regression models. It simplifies interpreting complex models, especially those involving Likert-scale predictors or other categorical data.

Installation

To install mlogitviz, run:

pip install mlogitviz

Usage

import pandas as pd
from mlogitviz import visualize_mlogit

# Load your dataset
df = pd.read_csv("your_data.csv")

# Define key parameters
dependent_var = "outcome"  # Your categorical dependent variable
independent_vars = ["predictor1", "predictor2", "predictor3"]  # Independent variables
baseline_category = "Baseline_Outcome"

# Visualize with custom parameters
visualize_mlogit(df, dependent_var, independent_vars,
                  baseline=baseline_category,
                  significance=0.05,
                  margeff_at='overall',
                  margeff_count=False,
                  output_file="heatmap.png")

Parameters

• df (pd.DataFrame): Data containing the dependent and independent variables.
• dependent (str): The name of the dependent variable (categorical).
• independent (list of str): List of predictor variables (continuous or categorical).
• baseline (str): Reference category for interpreting marginal effects.
• significance (float, default=0.05): Threshold for displaying statistically significant effects.
• margeff_at (str, default='overall'): Controls how marginal effects are computed.
  • 'overall': Recommended for Likert scales or diverse data types.
  • 'mean': Uses the mean values of predictors for marginal effects.
• margeff_count (bool, default=False): Treats count variables as continuous (False) or calculates changes when increased by one unit (True).
• output_file (str or None): Path to save the visualization as an image file. If None, it will be displayed directly.

Example Output

• The heatmap visualizes changes in probabilities, showing:
  • Positive values (red) = increased likelihood.
  • Negative values (blue) = decreased likelihood.
  • Baseline effects appear as a separate column with no significance indicators.

Credits

This package was co-authored by Payam Saeedi and Eric Williams.

License

This project is licensed under the MIT License.