Model Diagnostics

Use this section when you need to check whether a model or figure choice matches the structure of your data, especially for regression, bounded outcomes, compositional data, or survival data.

When to use this category

Use this category when your question is whether the model or figure you are using matches the structure of your data.

This section covers model-to-data fit for regression, bounded outcomes, compositional data, and survival data; it is not the place for missingness, multiplicity, independence, or broader study-design questions.

If your main question is about missing values, multiple comparisons, or repeated analytic choices, go to Missingness, Multiplicity, and Degrees of Freedom instead. If your main question is about pairing, pseudoreplication, repeated measures, or batch structure, go to Design and Independence.

Quick routing

If your main question is...Go to...Why
Is a multi-predictor regression unstable because of too many predictors, collinearity, or overlapping predictor structure?Regression Diagnostics GuardCovers predictor count, collinearity, and regression structure before claim-bearing output
Am I trying to use OLS on a binary, proportion, percentage, fraction, or other bounded outcome?Proportion Data OLS PreventionCovers when ordinary least squares is a poor fit for 0-1, binary, or percentage outcomes
Do several columns represent parts of a whole that sum to 1 or 100%?Compositional Data WarningCovers compositional structure, constrained totals, and why standard regression can mislead
Is the outcome really time-to-event data with censoring, such as overall survival, progression-free survival, or relapse timing?Survival Data Detection and Display GuardCovers survival-data detection, censoring, and whether survival-specific safeguards are in place

Child pages

Start here

  1. Start with Regression Diagnostics Guard if you are fitting a regression with multiple predictors and are worried about overfitting, collinearity, or unstable model structure.
  2. Start with Proportion Data OLS Prevention if your outcome is binary, proportional, percentage-based, or otherwise bounded between 0 and 1.
  3. Start with Compositional Data Warning if several columns are parts of a whole and sum to a fixed total.
  4. Start with Survival Data Detection and Display Guard if your outcome is time-to-event and censoring may be present.
  • Related links are registered in frontmatter and rendered below this stub.