logiSense

logiSense is an R package designed to simplify and make sense of logistic regression results. It provides concise and human-readable outputs for both continuous and categorical variables, making it easier to derive insights from your logistic regression models.

Motivation

Interpreting logistic regression results is challenging due to coefficients being presented on the log-odds scale, which can confuse non-statisticians when translating the results into probabilities or understanding predictor effects. From reviewing existing literature, a paper by Rajeev Kumar Malhotra highlights critical errors in interpreting odds ratios, including treating odds ratios as direct probabilities or relative risks, leading to exaggerated conclusions about associations.

Installation

You can install the most recent version of logiSense from Github using the following commands:

install.packages("devtools")
library(devtools)
devtools::install_github("BelinaJang/logiSense")
library(logiSense)

Functions

The package contains two functions:

1. logis

• Provides logistic regression interpretation for models without interaction terms.

2. logint

• Provides logistic regression interpretation for models with one interaction term.

Arguments

The logis and logint functions in logiSense require specific arguments, including the model formula, dataset, and variables of interest, to provide clear and interpretable outputs for both continuous and categorical predictors.

Description of Parameters for logis

Parameter	Definition	Type
formula	Description of the model to be fitted	formula
data	Name of the data	data frame
variable_interest	Variable name interested for interpretation	character
variable_type	Type of variable of interest	character
sigfig	Number indicating the number of significant digits to be used	numeric

Description of Parameters for logint

Parameter	Definition	Type
formula	Description of the model to be fitted	formula
data	Name of the data	data frame
variable1	Name of the first variable in the interaction of interest	character
vairiable2	Name of the second variable in the interaction of interest	character
vairiable1_type	Type of the first variable in the interaction of interest	character
vairiable2_type	Type of the second variable in the interaction of interest	character
sigfig	Number indicating the number of significant digits to be used	numeric

Example

The examples below demonstrate the use of logis and logint functions with a stroke prediction dataset from Kaggle, showcasing both categorical and continuous variables.

logis (no interaction term):

Continuous variable:

library(logiSense)
library(here)

test_data <- read.csv(here("data/test_data.csv"))

logis(formula = stroke ~ gender + age + hypertension + heart_disease + avg_glucose_level + smoking_status,
                    data = test_data,
                    variable_interest = "age",
                    variable_type = "continuous",
                    sigfig=4)

Categorical variable:

logis(formula = stroke ~ gender + age + hypertension + heart_disease + avg_glucose_level + smoking_status,
                    data = test_data,
                    variable_interest = "smoking_status",
                    variable_type = "categorical",
                    sigfig = 4)

logint (two-way interaction):

Continuous × Categorical:

logint(formula = stroke ~ work_type * age,
                  variable1 = "age",
                  variable2 = "work_type",
                  variable1_type = "continuous",
                  variable2_type = "categorical",
                  data = test_data,
                  sigfig = 4)

Continuous × Continuous:

logint(formula = stroke ~ age * avg_glucose_level,
                  variable1 = "age",
                  variable2 = "avg_glucose_level",
                  variable1_type = "continuous",
                  variable2_type = "continuous",
                  data = test_data,
                  sigfig = 4)

Categorical × Categorical:

logint(formula = stroke ~ work_type * Residence_type,
                  variable1 = "work_type",
                  variable2 = "Residence_type",
                  variable1_type = "categorical",
                  variable2_type = "categorical",
                  data = test_data,
                  sigfig = 4)

Comparison with Existing Packages

While existing R packages focus on model performance and visualization, logiSense uniquely addresses the gap in direct and comprehensive interpretation of logistic regression results, as summarized in the comparison below.

Feature	modelsummary	sjPlot	gtsummary	performance	logiSense
Human-readable coefficients	✅	✅	✅	❌	✅
Odds ratio interpretation	✅	✅	✅	❌	✅
User-friendly outputs	❌	✅	✅	❌	✅
Model performance diagnostics	❌	❌	❌	✅	❌
Direct interpretation tools	❌	❌	❌	❌	✅

Presentation

You can find the detailed presentation on how to use logiSense here.

Credits

logiSense was developed by Anna (Jingxuan) He, Belina Jang, Vanessa Liao, Tina (Zhaoyu) Tan, & Victoria Truong, as part of the course CHL 8010: Statistical Programming and Computation for Health Data, taught by Aya Mitani at the University of Toronto in 2024. Special thanks to Aya for her guidance and support!