Bayesian Multilevel Models for Repeated Measures dаta A Conceptual and Practical Introduction in R
One book, placed in its field — what Santiago Barreda, Noah Silbert's book gets right, where it goes further than the rest, and what the rest of the field adds.
Mike West, Editor · Leading With Data · peopleanalyst.com/guides/profile-bayesian_multilevel_models_repeated_measures
Arrow keys or swipe to move · N for notes · F for full screen
Bayesian Multilevel Models for Repeated Measures dаta A Conceptual and Practical Introduction in R
One book, placed in its field — what Santiago Barreda, Noah Silbert's book gets right, where it goes further than the rest, and what the rest of the field adds.
Mike West, Editor · Leading With Data · peopleanalyst.com/guides/profile-bayesian_multilevel_models_repeated_measures
The problem
Apparent Height erodes when it is left to instinct instead of method.
You were taught the moves piecemeal, never the whole model.
Bayesian Multilevel Models for Repeated Measures dаta A Conceptual and Practical Introduction in R · Mike West, Editor · Leading With Data
The model
- 01Fundamental Frequency (f0)
- 02Acoustic Vocal-Tract Length (VTL)
- 03Apparent Age
- 04Apparent Gender
- 05Apparent Height
- 06Listener Variation
- 07Speaker Variation
Bayesian Multilevel Models for Repeated Measures dаta A Conceptual and Practical Introduction in R · Mike West, Editor · Leading With Data
Step 1 of 5 · Fundamental Frequency (f0)
Model f0 on a log or semitone scale to match how pitch is actually perceived.
Misconception
That f0 and vocal-tract length carry the same information about speaker size, so including one makes the other redundant.
Reality
f0 reflects vocal-fold mass and tension while VTL reflects cavity resonance; the two dissociate (e.g. a small woman can have low f0), and models need both because listeners use them for partly different inferences.
Bayesian Multilevel Models for Repeated Measures dаta A Conceptual and Practical Introduction in R · Mike West, Editor · Leading With Data
Step 2 of 5 · Acoustic Vocal-Tract Length (VTL)
Treat VTL as an estimated formant-derived quantity, not a direct observation.
Misconception
That VTL is a raw measured quantity you can read directly off the spectrogram like f0.
Reality
VTL is inferred from the pattern of formant frequencies via a resonance model, so it is an estimate carrying its own uncertainty and depends on the estimation method you choose.
Bayesian Multilevel Models for Repeated Measures dаta A Conceptual and Practical Introduction in R · Mike West, Editor · Leading With Data
Step 3 of 5 · Apparent Age
Code apparent age as a moderator via interactions, not as a standalone additive offset.
Misconception
That apparent age just shifts predicted height up or down by a constant amount.
Reality
Apparent age changes the slopes, not only the intercept: the same VTL implies a different height depending on whether a voice is heard as a child's or an adult's.
Bayesian Multilevel Models for Repeated Measures dаta A Conceptual and Practical Introduction in R · Mike West, Editor · Leading With Data
Step 4 of 5 · Apparent Gender
Apparent gender adds categorical size expectations beyond what f0 and VTL supply.
Misconception
That once f0 and VTL are in the model, apparent gender adds nothing because gender is already encoded in those acoustics.
Reality
Apparent gender carries residual, categorical expectations about typical body size that operate beyond the graded acoustic cues, so it contributes over and above f0 and VTL.
Bayesian Multilevel Models for Repeated Measures dаta A Conceptual and Practical Introduction in R · Mike West, Editor · Leading With Data
Step 5 of 5 · Apparent Height
Apparent height is the response the whole model is built to explain.
Misconception
That because each listener judges many speakers, you can pool all height judgments as independent observations.
Reality
Repeated judgments by the same listener and about the same speaker are correlated; ignoring that non-independence shrinks standard errors and inflates false confidence.
Bayesian Multilevel Models for Repeated Measures dаta A Conceptual and Practical Introduction in R · Mike West, Editor · Leading With Data
A framework to use
Progressive Model Building Framework
This framework is part of the full guide. See the guide →
Bayesian Multilevel Models for Repeated Measures dаta A Conceptual and Practical Introduction in R · Mike West, Editor · Leading With Data
The cheat sheet
- Fundamental Frequency (f0)Model f0 on a log or semitone scale to match how pitch is actually perceived.
- Acoustic Vocal-Tract Length (VTL)Treat VTL as an estimated formant-derived quantity, not a direct observation.
- Apparent AgeCode apparent age as a moderator via interactions, not as a standalone additive offset.
- Apparent GenderApparent gender adds categorical size expectations beyond what f0 and VTL supply.
- Apparent HeightApparent height is the response the whole model is built to explain.
- Listener VariationInclude random slopes, not just intercepts, for listener-level cue weighting.
- Speaker VariationSpeaker enters as a crossed random effect, not nested under listener.
Bayesian Multilevel Models for Repeated Measures dаta A Conceptual and Practical Introduction in R · Mike West, Editor · Leading With Data
Sources
- Bayesian Multilevel Models for Repeated Measures dаta A Conceptual and Practical Introduction in R — Santiago Barreda, Noah Silbert
Bayesian Multilevel Models for Repeated Measures dаta A Conceptual and Practical Introduction in R · Mike West, Editor · Leading With Data