R-Squared and Model Fit

College Depth 81 in the knowledge graph I know this Set as goal
Unlocks 72 downstream topics
R-squared goodness-of-fit adjusted-R-squared model-selection

Core Idea

R² measures the fraction of variation in y explained by the regressors: R² = 1 − SSR/SST, where SSR is the sum of squared residuals and SST is total variance. It always lies between 0 and 1, and adding any regressor — even irrelevant — cannot decrease it. The adjusted R² penalizes for additional regressors, making it more appropriate for model comparison: R̄² = 1 − [SSR/(n−k−1)]/[SST/(n−1)]. High R² does not imply unbiased coefficient estimates; low R² does not imply the estimates are wrong or the model is useless for causal inference.

How It's Best Learned

Compare R² and adjusted R² across nested models (same data, different regressors). Note that adding noise variables can raise R² but lower R̄².

Common Misconceptions

Explainer

From bivariate regression, you learned how to fit a line through data by minimizing squared residuals — the vertical distances between data points and the fitted line. Those residuals capture what the model fails to explain. formalizes this intuition into a single summary statistic: the fraction of the total variation in y that your regression accounts for.

The formula makes the decomposition explicit. Total sum of squares (SST) = Σ(yᵢ − ȳ)² measures the total variation in the outcome around its unconditional mean. Residual sum of squares (SSR) = Σ(yᵢ − ŷᵢ)² is the unexplained variation that remains after fitting the model. R² = 1 − SSR/SST. When the model perfectly fits every data point, SSR = 0 and R² = 1. When the model simply predicts the mean for every observation (no regressors at all), SSR = SST and R² = 0. An R² of 0.60 means the regressors collectively account for 60% of the variation in y; the remaining 40% is unexplained.

A crucial mechanical fact: adding any variable to a regression can never decrease R². OLS can always set a new coefficient to zero if the variable adds nothing, so SSR can only stay flat or fall, meaning R² can only stay flat or rise. This is why comparing R² across models with different numbers of predictors is misleading — you could achieve R² = 0.99 by including enough noise variables. Adjusted R² corrects for this by penalizing the loss of degrees of freedom: R̄² = 1 − [SSR/(n−k−1)] / [SST/(n−1)], where k is the number of regressors. The adjustment means adding a truly uninformative variable can lower R̄², making it a better model comparison tool than raw R².

The deepest point — and the most consequential misconception — is that R² has nothing to do with whether your regression is correctly specified for causal inference. The key OLS assumption for unbiased estimation is E[u|x] = 0: the regressors are uncorrelated with the error term. R² measures explained variance regardless of whether this assumption holds. You can have R² = 0.95 with severe omitted variable bias, and R² = 0.04 with a clean randomized experiment delivering perfectly unbiased coefficients. As you move further into econometrics, you will regularly see researchers report very low R² without apology — they are pursuing credible identification of a causal effect, not maximizing explained variance. The two goals are genuinely separate.

Practice Questions 5 questions

Prerequisite Chain

Counting to 10Counting to 20Understanding ZeroThe Number ZeroCounting to FiveOne-to-One CorrespondenceCombining Small Groups Within 5Addition Within 10Addition Within 20Two-Digit Addition Without RegroupingTwo-Digit Addition with RegroupingAddition Within 100Repeated Addition as MultiplicationMultiplication Facts Within 100Division as Equal SharingDivision as Grouping (Measurement Division)Division: Grouping (Repeated Subtraction) ModelDivision: Fair Sharing ModelDivision as Equal SharingDivision as GroupingBasic Division FactsDivision Facts Within 100Two-Digit by One-Digit DivisionDivision with RemaindersRemainders and Quotients in DivisionDivision Word ProblemsIntroduction to Long DivisionFactors and MultiplesPrime and Composite NumbersEquivalent FractionsRelating Fractions and DecimalsDecimal Place ValueReading and Writing DecimalsComparing and Ordering DecimalsAdding and Subtracting DecimalsMultiplying DecimalsDividing DecimalsDividing FractionsMixed Number ArithmeticOrder of OperationsInteger Order of OperationsVariable ExpressionsCombining Like TermsOne-Step EquationsTwo-Step EquationsSolving Multi-Step EquationsEquations with Variables on Both SidesAngle Pairs: Complementary, Supplementary, and VerticalParallel Lines and TransversalsCorresponding AnglesAlternate Interior AnglesTriangle Angle Sum TheoremExterior Angle TheoremTriangle Inequality TheoremSimilar Triangles: AA SimilaritySimilar Triangles: SSS and SAS SimilarityProportions in Similar TrianglesRight Triangle Trigonometry IntroductionTrigonometric Ratios ReviewRadian MeasureConverting Between Degrees and RadiansThe Unit CircleGraphing Sine and CosineGraphing Tangent and Reciprocal Trigonometric FunctionsDerivatives of Trigonometric FunctionsAntiderivativesIndefinite IntegralsBasic Integration RulesRiemann SumsDefinite Integral DefinitionProbability Density Functions and Continuous DistributionsCumulative Distribution FunctionsContinuous Random VariablesNormal DistributionCentral Limit TheoremConfidence Intervals for MeansZ-Tests and T-Tests for MeansOne-Sample Z-Test for MeansOne-Sample and Two-Sample T-TestsOne-Way ANOVAF-Test and Joint SignificanceR-Squared and Model Fit

Longest path: 82 steps · 420 total prerequisite topics

Prerequisites (4)

Leads To (4)