Multinomial and Ordinal Outcome in Excel with BESHStatNG

Multinomial and ordinal regression models are useful when the outcome has more than two categories. In this tutorial, we show how to fit both types of models in Excel with BESH Stat NG using the GUI workflow and the UDF workflow.

The tutorial uses two real clinical-style examples from biomedical datasets:

a multinomial logistic regression example with a six-class dermatology diagnosis outcome
an ordinal logistic regression example with an ordered heart-disease severity outcome coded from 0 to 4

This page follows the same structure as the other BESH Stat NG tutorials:

real-world example data
step-by-step GUI workflow
reproducible worksheet/UDF workflow
practical interpretation of the output
downloadable workbook template

Download the example workbook

The workbook contains these sheets:

README — overview of both examples
Data_Multinomial — multiclass dermatology example
MN_Reference — reference multinomial output
MN_UDF_Workflow — live multinomial UDF workflow
Data_Ordinal — ordered heart-disease example
ORD_Reference — reference ordinal output
ORD_UDF_Workflow — live ordinal UDF workflow
Sources — dataset and documentation links

When to use multinomial vs ordinal regression

These two model families are related, but they are used for different kinds of outcomes.

Multinomial logistic regression

Use a multinomial model when the response categories are nominal and have no natural order.

Examples:

disease subtype
histology class
treatment group
diagnostic label

In the workbook, the multinomial example predicts one of six dermatology classes.

Ordinal logistic regression

Use an ordinal model when the response categories are ordered.

Examples:

disease severity grade
toxicity grade
symptom score
progression stage

In the workbook, the ordinal example predicts heart-disease severity coded from 0 to 4.

Multinomial logistic regression in Excel

Example data

The multinomial workflow uses a dermatology dataset with six outcome classes:

psoriasis
seborrheic dermatitis
lichen planus
pityriasis rosea
chronic dermatitis
pityriasis rubra pilaris

The workbook uses these predictors:

age
erythema
scaling
itching
definite borders

This is a good tutorial example because the outcome is clearly multiclass and the predictors are clinically interpretable.

GUI workflow: multinomial logistic regression

Step 1. Open the tool

Go to:

BESH Stat NG → Analyse → Regression → Multinomial Logistic Regression

Excel ribbon screenshot showing the BESH Stat NG Regression menu with Multinomial Logistic Regression and Ordinal Logistic Regression tools.

Step 2. Select the dependent variable and predictors

On the Select Variables tab:

set the dependent variable to class_code
move the predictors to the predictor box:
- age
- erythema
- scaling
- itching
- def_borders

BESH Stat NG Multinomial Logistic Regression input dialog in Excel with class_code as the dependent variable and age, erythema, scaling, itching, and definite borders as predictors. — Multinomial Logistic Regression input dialog in BESH Stat NG with class_code selected as the outcome and five dermatology predictors assigned.

Step 3. Build the model

On the Build Model tab, add the selected effects to the model.

In this tutorial, the model uses a simple additive specification with an intercept:

age
erythema
scaling
itching
def_borders

BESH Stat NG Multinomial Logistic Regression Build Model tab in Excel showing age, erythema, scaling, itching, and definite borders added as selected effects. — Build Model tab for the multinomial workflow showing the additive predictor effects used in the dermatology example.

Step 4. Set the options

On the Options tab:

reference category: First
convergence criterion: 1E-06
max. IRLS iterations: 150
alpha: 0.05

BESH Stat NG Multinomial Logistic Regression Options tab in Excel showing reference category first, alpha 0.05, and convergence settings. — Options tab for the multinomial workflow with reference category set to First, alpha 0.05, and convergence settings for the fit.

Step 5. Fit the model

Click Fit.

The GUI returns:

coefficient tables for each class contrast
odds-ratio output
likelihood-ratio model statistics
a classification table

Multinomial results for this example

In this workbook, the multinomial model was fit with class 1 as the reference category.

Key model results:

Observations: 358
Null log likelihood: -602.6121
Final log likelihood: -362.3298
Likelihood-ratio chi-square: 480.5646
McFadden pseudo R²: 0.3987
AIC: 784.6595
BIC: 901.0755
Training accuracy: 59.78%

This is a meaningful fit for a six-class clinical outcome. The model is clearly better than the null model, and several class-specific coefficients are strongly associated with the diagnostic contrasts.

Examples from the fitted output:

for class 2 vs class 1, def_borders has a strong negative coefficient, with an odds ratio well below 1
for class 3 vs class 1, scaling is strongly negative while itching is strongly positive
the class-specific coefficients differ across contrasts, which is exactly what makes multinomial modeling useful in multiclass diagnosis problems

The classification table also shows that some classes are predicted much more accurately than others, which is typical in multiclass biomedical data. Download the full GUI based excel workbook with results here.

UDF workflow: multinomial logistic regression

The GUI workflow is convenient for interactive analysis, but the UDF workflow is better when you want a reproducible workbook, a template, or a tutorial-ready sheet.

The workbook uses these BESH Stat NG functions:

BESH.REGR.MNLOGIT_FIT(...)
BESH.REGR.MNLOGIT_SUMMARY(...)
BESH.REGR.MNLOGIT_TESTS(...)
BESH.REGR.MNLOGIT_CLASS(...)
BESH.REGR.MNLOGIT_PRED(...)

Fit handle formula

=BESH.REGR.MNLOGIT_FIT(Data_Multinomial!$F$8:$F$365,Data_Multinomial!$A$8:$E$365,$B$5,,,$B$7,TRUE,$B$6,"relative",$B$9,$B$10,$B$8)

Summary spill

=BESH.REGR.MNLOGIT_SUMMARY($B$12,TRUE,$B$8)

Model tests spill

=BESH.REGR.MNLOGIT_TESTS($B$12,TRUE)

Classification spill

=BESH.REGR.MNLOGIT_CLASS($B$12,TRUE)

Scenario prediction spill

=BESH.REGR.MNLOGIT_PRED($B$12,B49:F52,,TRUE)

The multinomial prediction block in the workbook shows how to score new hypothetical dermatology profiles directly in Excel.

Interpreting the multinomial example

The multinomial model answers this question:

How do the predictor patterns differ across multiple diagnosis categories?

In this dermatology example:

the outcome is nominal rather than ordered
the fitted coefficients are interpreted relative to a reference class
each diagnostic class has its own contrast against the baseline

That makes multinomial logistic regression a good fit for multiclass diagnostic workflows where the outcome categories are distinct but not naturally ranked.