Item Response Theory
This is the course about how a modern test actually measures you. Every adaptive exam — the GRE, the NCLEX, the TExES practice engine inside QuantegyAI — runs on Item Response Theory (IRT): a model that puts people and questions on the same scale, so a hard question answered correctly counts for more than an easy one, and the test can choose the most informative next question for you specifically.
Every module is hands-on and runs entirely in your browser. You will drag the item characteristic curve until it fits, slide difficulty and discrimination and watch the curve move, build a 3-parameter model with guessing, estimate a student’s ability from a pattern of right and wrong answers, and step a computer-adaptive test as it homes in. Each module also shows the matching formula in clean notation, and ends with a short mastery check; pass it to mark the module complete.
The problem & the curve
Module 1Why Percent-Correct Isn’t Enough
Classical test theory and its fatal flaw: a raw score depends on which questions you happened to get. The IRT idea — put ability \( \theta \) and item difficulty on one shared scale. Activity: see the same student “change ability” on an easy vs. hard form.
Module 2The Item Characteristic Curve
The heart of IRT: a curve giving the probability of a correct answer at every ability level. The logistic S-shape, why it rises, and how to read it. Activity: drag an ability marker along a live ICC and watch the probability update.
The three models
Module 3The Rasch / 1PL Model — Difficulty
One parameter: difficulty \( b \), the ability where the chance of success is 50/50. Why the Rasch model is special. Activity: slide \( b \) and watch the whole curve glide left (easier) and right (harder).
Module 4The 2PL Model — Discrimination
A second parameter: discrimination \( a \), the slope. Steep items separate students sharply; flat items barely tell anyone apart. Activity: compare two items at once and find which one better splits the borderline students.
Module 5The 3PL Model — Guessing
A third parameter: the lower asymptote \( c \), the floor a low-ability student reaches by guessing on multiple choice. Activity: raise \( c \) and watch the curve lift off the floor — the model QuantegyAI actually uses.
Using the model
Module 6Estimating Ability
Run the model backwards: given a pattern of right and wrong answers, which ability \( \theta \) makes that pattern most likely? Maximum likelihood, in pictures. Activity: flip answers right/wrong and watch the likelihood peak slide.
Module 7Information & Adaptive Testing
How a test gets short and accurate: each item carries the most information near its own difficulty, so the test asks the question you’re most uncertain about. Activity: run a computer-adaptive test that picks each next item for a simulated student.
Doing it right · Capstone
Module 8 · CapstoneCalibration, Fit & Fairness
Where the parameters come from: estimating \( a, b, c \) from thousands of real responses, checking model fit, and screening for differential item functioning so an item is fair across groups. Ties every earlier module together — and to exactly how QuantegyAI calibrates its bank.