Concept map / Notation

Notation

The symbols used across the lessons, with the page where each one is introduced. A few letters do double duty in the literature (α, β and Φ each have two meanings); those rows say which is which.

Populations and samples

Greek letters name population quantities; Latin letters name the matching sample estimates.

SymbolMeaningIntroduced in
μ, x̄Population mean and sample mean.Variance
σ², s²Population variance (divide by N) and sample variance (divide by n − 1). σ and s are the standard deviations.Variance
ρ, rPopulation and sample correlation.Correlation
N, nPopulation size or number of people; sample size.
p, qA probability or proportion, and q = 1 − p.Bernoulli
E[X], Var(X)Expected value and variance of a random variable X.

Inference

SymbolMeaningIntroduced in
α (significance)Significance level: the Type I error rate a test allows, often 0.05.Hypothesis testing
β (error rate)Type II error rate; power is 1 − β. Not the same β as a regression coefficient.Hypothesis testing
ν, dfDegrees of freedom of a t, chi-square or F distribution.Inference distributions
χ²Chi-square distribution or statistic.Inference distributions, CFA model fit
Φ(z)Standard normal cumulative probability: the share of the normal curve below z.Distributions
L(θ), log LLikelihood of a parameter value given the data, and its natural logarithm. On this site log always means the natural log.Likelihood

Regression and multilevel models

SymbolMeaningIntroduced in
β₀, β₁Intercept and slope in a regression model.Regression
exp(β₁)In logistic regression, the odds ratio: how many times the odds of a yes multiply when the predictor rises by one unit.Logistic regression
a, b, c′ (paths)In a mediation model, the X → M path, the M → Y path and the direct X → Y path; a × b is the indirect effect. Not the IRT a and b.Path analysis and mediation
ŷPredicted value; y − ŷ is the residual.Regression diagnostics
τ², σ²Between-group (random intercept) variance and within-group residual variance in a multilevel model.HLM
ICCIntraclass correlation. In a multilevel model, the share of variance between groups, τ² / (τ² + σ²). For two raters, a measure of agreement.Nested and crossed effects, Rater agreement

Measurement and reliability

SymbolMeaningIntroduced in
X = T + EClassical test theory: observed score equals true score plus error.Reliability
ρXX′Reliability: the share of observed-score variance that is true-score variance.Reliability
α (coefficient)Coefficient alpha, a reliability estimate from item covariances. Unrelated to the significance level α.Alpha and omega
ωCoefficient omega, a reliability estimate from factor loadings.Alpha and omega
λFactor loading: how strongly an item depends on a factor.Factor analysis
φCorrelation between two factors.Factor analysis
κCohen’s kappa: agreement between raters corrected for chance.Rater agreement
G, Φ (G theory)Generalizability and dependability coefficients for relative and absolute decisions. This Φ is not the normal Φ(z).Generalizability theory

Item response theory

Every item curve on this site uses the logistic function with no scaling constant (D = 1), the same as the R package mirt. Some textbooks multiply by D = 1.702 so the curve matches a normal ogive; their discriminations are then smaller by that factor.

SymbolMeaningIntroduced in
θAbility (the latent trait), on a scale where the population usually has mean 0 and standard deviation 1. CFA texts often use Θ for error variances; this site does not.Item curves
θ̂An estimate of ability from a response pattern (MLE, MAP or EAP).Ability estimation
aDiscrimination: how steeply the item curve rises.Item curves
bDifficulty: the ability at which the curve is halfway between its floor and 1.Item curves
cLower asymptote (pseudo-guessing): the chance of a correct answer at very low ability in the 3PL.Item curves
P(θ)Probability of a correct response at ability θ: P(θ) = c + (1 − c) / (1 + e^(−a(θ − b))).Item curves
I(θ)Item or test information at θ. The standard error of an ability estimate is about 1 / √I(θ).Item information
A, B (linking)Linking constants that put one calibration on another scale: θ on the base scale = A × θ on the new scale + B.IRT scale linking
α (DCM)An attribute mastery profile in a diagnostic classification model, such as (1, 0, 1).DCM