cNORM - Continuous Norming with cNORM in jamovi
jamovi is an intuitive, free, and open-source statistical platform built on R. We have contributed the module cNORMj to the jamovi library, providing you with the comprehensive functionality of the cNORM package (v3.7.0). It covers data preparation (norm group construction and post-stratification raking), norm score generation for single groups, and unified continuous norming using both distribution-free and parametric modeling approaches.
1. Data Preparation (Norm Groups & Raking Weights)
Representative norming often requires preprocessing before models are fitted. The Data Preparation analysis provides two essential tools to optimize your standardization sample:
- Norm Group Formation:
Splits a continuous explanatory variable (e.g., chronological age in decimal years) into discrete norm groups and calculates each group's empirical midpoint. Grouping is helpful for diagnostic inspection and for distribution-free continuous norming.
- Number of norm groups: Set to
0 for automatic determination (targeting approximately n ≥ 100 observations per group, with a minimum of n ≥ 50). Alternatively, enter a specific number of groups.
- Equidistant intervals: When checked, splits the variable into intervals of equal width. When unchecked (default), forms quantile-based groups containing roughly equal numbers of observations.
- Output & Plot: Provides a summary table with group midpoints, bounds, sample sizes, and proportions, accompanied by an inspection plot displaying group sizes against the recommended thresholds (n = 50 and n = 100).
- Save: Save the calculated group midpoints as a new variable directly back to your jamovi spreadsheet.
- Post-Stratification / Raking Weights:
Compensates for sample bias by adjusting marginal distributions (e.g., sex, migration background, parental education, or region) to known census or population benchmarks using iterative proportional fitting (raking). Raking preserves more sample cases than classical cell-based stratification by aplying weights to each case.
- Stratification variables: Select one or more categorical variables from your dataset. The module immediately previews all observed levels and sample proportions in a summary table and provides a preformatted template string in the table footnote.
- Population proportions: Enter target marginals using the syntax displayed in the template (e.g.,
sex: 1=0.51, 2=0.49; migration: 0=0.68, 1=0.32). Within each variabe, the values should sum up roughly to 1.0.
- Standardize weights: Divides all weights by the minimum weight so that all case weights are ≥ 1.0 (ideal for subsequent weighted regression and quantile calculations).
- Weight Diagnostics: Reports minimum, maximum, and mean weights alongside Kish's Design Effect (Deff) and the effective sample size (Neff).
- Save: Save the calculated case weights as a new column in your dataset for use across jamovi analyses.
2. Single Group Norming (Without Age Dependency)
This analysis calculates norm scores and percentile ranks within a single population (where performance does not depend on age, grade, or other continuous covariates). Norm scores are derived using the Rankit procedure and inverse normal transformation. To smooth sampling error and close gaps caused by unobserved raw scores, the relationship can be smoothed via polynomial regression up to the 6th power.
Select your Raw Score Variable and configure the following options:
- Data Preparation & Method Parameters:
- Invert ranking order: Tick this if higher raw scores reflect lower performance (e.g., reaction times, error counts).
- Norm scale: Choose between T-scores (M = 50, SD = 10), IQ-scores (M = 100, SD = 15), z-scores (M = 0, SD = 1), Wechsler subtest scales (M = 10, SD = 3), or PISA scales (M = 500, SD = 100).
- Smooth with polynomial regression: Enabled by default. When deactivated, the module compiles the norm table directly from manifest Rankit percentiles without smoothing.
- Degree of location (k) & Term selection: Select the maximum polynomial degree (1 to 6) and choose between automatic best-subset selection or a manual number of terms. Inrease values for closer fit, reduce them for increasing smoothing.
- Weights (optional): Assign case weights (e.g., raking weights). If not set, each case is treated with an equal weight.
- Output & Visualization:
- Result plot: Displays observed data points against the fitted regression curve for either norm scores or percentiles. If confidence intervals are enabled, a semi-transparent confidence band is plotted along the curve.
- Model summary: Displays separate, clean tables for overall model fit (N, Terms, RMSE, Adj. R², BIC) and individual polynomial coefficients.
- Norm table range: Restricts the tabulated norm score range in standard deviations (default: ±3.0 SD).
- Raw score bounds & Step size: Customize the minimum raw score, maximum raw score, and stepping granularity of the table.
- Confidence intervals: Enable Kelley's true-score confidence intervals by specifying the test's reliability coefficient and desired confidence level (e.g., 90% or 95%). Intervals are calculated for both norm scores and percentile ranks.
- Save:
- Save manifest norm scores, manifest percentile ranks, fitted norm scores, and fitted percentiles directly back into your dataset.
3. Continuous Norming (Unified Distribution-Free & Parametric)
Instead of splitting samples into arbitrary age brackets and norming each group separately, continuous norming models test performance across the entire age spectrum simultaneously. This reduces the required sample size, eliminates "jumps" between consecutive age tables, and substantially reduces sampling variance (W. Lenhard & Lenhard, 2021). The analysis supports four distribution families:
- Modeling Methods:
- Distribution-Free (Taylor Polynomials): Models the raw score as a bivariate polynomial function of norm score (location, degree k) and age (degree t). It makes no distributional assumptions. If continuous age is entered (> 20 distinct values), cNORMj automatically discretizes age into well-sized norm groups. You can also specify the number of groups manually.
- Beta-Binomial: Tailored for accuracy and power tests with a fixed number of items n (scores 0 to n). Parameters α and β are modeled as smooth polynomial functions of age. It is ideal for Rasch/1PL scaled assessments.
- Sinh-Arcsinh (SHASH): A continuous 4-parameter distribution (location μ, scale σ, skewness ε, tail weight δ) modeled over age. It accommodates unbounded scores, response times, and asymmetric physical or cognitive growth data. The tail parameter δ can optionally be fixed to a constant (e.g., 1.0 for normal tails).
- Conway-Maxwell-Poisson (CMP): Designed for speeded tests, counts, and rate data (non-negative integers). It models location μ and dispersion ν across age, capturing over-dispersion (ν < 1), equi-dispersion (ν = 1, Poisson), or under-dispersion (ν > 1). It also supports truncation for tests with an upper ceiling (maximum attainable score).
- Model Evaluation & Tables:
- Percentile Plot: Displays model-implied percentile trajectories (smooth curves or discrete steps in case of betabinomial models) alongside manifest group percentiles (diamonds) and individual raw observations.
- Model Fit & Coefficients Tables: Reports N, parameter count, Log-likelihood, AIC, BIC, R², RMSE, and bias, together with parameter estimates, standard errors, z-values, and p-values.
- Norm Score Table: Enter a target value for the explanatory variable (e.g.,
normAge = 9.5) to generate an exact norm table. You can adjust stepping, raw score limits, and add reliability-based confidence intervals.
- Save:
- Store fitted norm scores and fitted percentile ranks in the spreadsheet for every participant as new variables. Saved columns automatically clear and re-populate whenever model settings change.
4. Built-in Example Datasets
cNORMj includes four standard datasets from the cNORM package to help you explore each methodology. To load a dataset, click on the jamovi burger menu (three horizontal bars at the top left), select Open, and navigate to the Data Library → cNORMj folder:
- Sentence comprehension (ELFE 1-6): Reading comprehension test for primary school students (grades 1-4, raw scores 0-28). Ideal for Taylor polynomials and Beta-Binomial models.
- Vocabulary development (PPVT-IV): Peabody Picture Vocabulary Test standardization data covering ages 2.5 to 17.5 years (scores 0-228). Includes demographic variables (
sex, migration) to demonstrate post-stratification raking.
- Growth data (CDC BMI): Pediatric growth and BMI measurements across childhood and adolescence (ages 2-18). Ideal for the Sinh-Arcsinh (SHASH) distribution or distribution-free modeling.
- Reading fluency (ELFE-II Speed): Speeded word-picture matching task measuring decoding speed across ages 7 to 11 (scores 0-75 with a test ceiling). Since it only depends on speed, it is tailored for Conway-Maxwell-Poisson (CMP) continuous norming.
Installation
The 'cNORMj' module is available via the Jamovi library for all platforms. Please click on the 'Modules' option on the top right. You will find a library for the installation and management of the modules.
You can as well download the module for Win64 and install it manually via 'Sideload'.