Model Fitting

The Model Fitting module provides a graphical interface for fitting and diagnosing regression models. It supports multiple model families, diagnostic plotting with bootstrap overlays, and model comparison.

Access via Advanced > Model Fitting.

Getting Started

When you open the module, you'll see three main sections:

1. Response variable — select the outcome variable and model type
2. Explanatory variables — choose predictors from numeric and categorical variable lists
3. Model/Residual Plots — diagnostic plots and model comparisons

Model Types

Response Types

Select the response type based on the nature of your outcome variable:

Response Type	Model	Use When
Numeric	Linear regression (gaussian)	Continuous outcome (e.g., height, income)
Binary	Logistic regression (binomial)	Two-level outcome (e.g., yes/no, pass/fail)
Count	Poisson regression	Count data (e.g., number of events)
Count (overdispersed)	Negative binomial	Count data with more variance than Poisson predicts

For binary (logistic) models, you can also choose between logit and probit link functions.

Building a Model

1. Select your response variable from the dropdown
2. Choose the response type (numeric, binary, count)
3. Add explanatory variables by selecting from the numeric and categorical variable lists
4. Optionally add confounding variables that you want to control for
5. Toggle the intercept on or off

The model formula is displayed and updated as you make changes. Each time you modify the model, iNZight automatically fits it and updates the summary and diagnostic plots.

Multiple Models

You can fit and name multiple models, then switch between them using the model list. This makes it easy to compare different model specifications. Use the Compare button to generate a comparison table of model fit statistics.

Diagnostic Plots

The module provides a range of diagnostic plots to assess model fit. Select from the Residual plots dropdown:

Plot	Description
Residual	Residuals vs. fitted values — check for non-linearity and heteroscedasticity
Scale-Location	Square root of standardised residuals vs. fitted — check for constant variance
Leverage	Residuals vs. leverage — identify influential observations
Cook's Distance	Influence of each observation on the model
Normal Q-Q	Check if residuals follow a normal distribution
Histogram	Histogram of residuals
Summary Matrix	All six diagnostic plots displayed together
Partial Residual	Partial residual plots for each predictor
Factor Comparisons	Visual comparison of factor level effects
Marginal Model Plots	Compare observed and fitted marginal relationships
Forest Plot	Forest plot of coefficient estimates and confidence intervals

Bootstrap Diagnostics

When the Show bootstraps checkbox is enabled, diagnostic plots include bootstrap resampling overlays. This helps assess the stability of model diagnostics — if the bootstrapped lines vary widely, the diagnostic pattern may not be reliable.

Bootstrap is enabled by default for datasets with 30–4000 observations.

Survey Data

The Model Fitting module supports survey data. When a survey design has been specified, iNZight automatically uses svyglm() from the survey package to fit the model, correctly accounting for the survey design in parameter estimates and standard errors.

Model Summary

The model summary output includes:

• Coefficient estimates with standard errors, t/z values, and p-values
• R-squared and adjusted R-squared (for linear models)
• AIC and other fit statistics
• ANOVA table for overall model significance

The summary is displayed in a text window alongside the diagnostic plots.

Code History

The Model Fitting module keeps a history of the R code used to fit models and generate plots. Use the toolbar buttons to:

• Save code (R icon) — save the R code for the current model or plot to the history
• View code history (history icon) — open the code history window showing all saved code

This is useful for reproducing your analysis outside of iNZight or for learning the underlying R commands.