Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed...
Transcript of Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed...
![Page 1: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/1.jpg)
Mixed Models for Longitudinal Binary Outcomes
Don HedekerDepartment of Public Health Sciences
Biological Sciences DivisionUniversity of Chicago
Hedeker, D. (2005). Generalized linear mixed models. In B. Everitt & D. Howell (Eds.), Encyclopedia
of Statistics in Behavioral Science. Wiley.
Hedeker, D. & Gibbons, R.D. (2006). Longitudinal Data Analysis, chapter 9. Wiley.
This work was supported by National Institute of Mental Health Contract N44MH32056.
1
![Page 2: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/2.jpg)
Mixed-effects models for categorical outcomes
• dichotomous outcomes
– mixed-effects logistic regression
• ordinal outcomes
– mixed-effects ordinal logistic regression
∗ proportional odds model
∗ partial or non-proportional odds model
• nominal outcomes
– mixed-effects nominal logistic regression
• discrete or grouped time-to-event data
– mixed-effects dichotomous or ordinal regression
∗ complementary log-log link for proportional (andnon-proportional) hazards models
2
![Page 3: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/3.jpg)
Logistic Regression Model
log
P (Yi = 1)
1− P (Yi = 1)
= x′iβ
• Dichotomous outcome (Y = 0 absence, Y = 1 presence).
• Function that links probabilities to regressors is the logit (or logodds) function log [P/(1− P ]. Logit is called the link function.
The model can be written in terms of probabilities:
P (Yi = 1) =1
1 + exp(−x′iβ)
•Model is a linear model for the logits, not for the probabilities.Logits can take on any values between negative and positiveinfinity, probabilities can only take on values between 0 and 1
3
![Page 4: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/4.jpg)
Dichotomous Response and Threshold ConceptContinuous yi - an unobservable latent variable - related todichotomous response Yi via “threshold concept”
Response occurs (Yi = 1) if γ < yiotherwise, a response does not occur (Yi = 0)
4
![Page 5: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/5.jpg)
The Threshold Concept in Practice“How was your day?” (what is your satisfaction level today?)
• Satisfaction may be continuous, but we usually emit adichotomous response:
5
![Page 6: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/6.jpg)
Model for Latent Continuous ResponsesConsider the model with p covariates for the latent responsestrength yi (i = 1, 2, . . . , N):
yi = x′iβ + εi
• probit: εi ∼ standard normal (mean=0, variance=1)
• logistic: εi ∼ standard logistic (mean=0, variance=π2/3)
⇒ β estimates from logistic regression are larger (in abs. value)than from probit regression by approximately
√π2/3 = 1.8
Underlying latent variable
• useful way of thinking of the problem
• not an essential assumption of the model
6
![Page 7: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/7.jpg)
Random-intercept Logistic Regression Model
Consider the model with p covariates for the response Yij for subjecti (i = 1, 2, . . . , N) at time j (j = 1, 2, . . . , ni):
log
P (Yij = 1 | υ0i)
1− P (Yij = 1 | υ0i)
= x′ijβ + υ0i
where
Yij = dichotomous response for subject i at time j
xij = (p + 1)× 1 covariate vector (includes 1 for intercept)
β = (p + 1)× 1 vector of unknown parameters
υ0i = subject effects distributed NID(0, σ2υ)
7
![Page 8: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/8.jpg)
Model for Latent Continuous Responses
Consider the model with p covariates for the ni × 1 latent responsestrength yij:
yij = x′ijβ + υ0i + εij
where assuming
• εij ∼ standard normal (mean 0 and σ2 = 1) leads tomixed-effects probit regression
• εij ∼ standard logistic (mean 0 and σ2 = π2/3) leads tomixed-effects logistic regression
8
![Page 9: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/9.jpg)
Underlying latent variable
• not an essential assumption of the model
• useful for obtaining intra-class correlation (r)
r =σ2υ
σ2υ + σ2
and for design effect (d)
d =σ2υ + σ2
σ2 = 1/(1− r)
ratio of actual variance to the variance that would be obtained bysimple random sampling (holding sample size constant)
9
![Page 10: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/10.jpg)
Scaling of regression coefficients
β estimates from mixed-effects model are larger (in abs. value) thanfrom fixed-effects model by approximately
√d =
√√√√√√√√σ2υ + σ2
σ2
because
• V (y) = σ2υ + σ2 in mixed-effects model
• V (y) = σ2 in fixed-effects model
difference depends on size of random-effects variance σ2υ
10
![Page 11: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/11.jpg)
Treatment-Related Change Across Time
Data from the NIMH Schizophrenia collaborative study on treatment relatedchanges in overall severity. IMPS item 79, Severity of Illness, was scored as:
1 = normal2 = borderline mentally ill3 = mildly ill4 = moderately ill5 = markedly ill6 = severely ill7 = among the most extremely ill
The experimental design and corresponding sample sizes:
Sample size at WeekGroup 0 1 2 3 4 5 6 completersPLC (n=108) 107 105 5 87 2 2 70 65%DRUG (n=329) 327 321 9 287 9 7 265 81%Drug = Chlorpromazine, Fluphenazine, or Thioridazine
Main question of interest:
•Was there differential improvement for the drug groups relative to the controlgroup?
11
![Page 12: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/12.jpg)
• Under SSI, Inc > “SuperMix (English)” or “SuperMix (English) Student”
• Under “File” click on “Open Spreadsheet”
• Open C:\SuperMixEn Examples\Workshop\Binary\SCHIZX1.ss3(or C:\SuperMixEn Student Examples\Workshop\Binary\SCHIZX1.ss3)
12
![Page 13: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/13.jpg)
C:\SuperMixEn Examples\Workshop\Binary\SCHIZX1.ss3
13
![Page 14: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/14.jpg)
Select Imps79D column, then “Edit” > “Set Missing Value”
14
![Page 15: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/15.jpg)
Select “File” > “Data-based Graphs” > “Bivariate”
15
![Page 16: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/16.jpg)
16
![Page 17: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/17.jpg)
Select “File” > “Data-based Graphs” > “Bivariate”
17
![Page 18: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/18.jpg)
18
![Page 19: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/19.jpg)
Observed proportions ≥ “moderately ill”week 0 week 1 week 3 week 6
placebo .98 .91 .89 .71drug .99 .82 .66 .42
Observed odds ≥ “moderately ill”week 0 week 1 week 3 week 6
placebo 52.5 9.50 7.70 2.50drug 80.8 4.63 1.93 .73
ratio .65 2.05 3.99 3.42
Observed log odds ≥ “moderately ill”week 0 week 1 week 3 week 6
placebo 3.96 2.25 2.04 .92drug 4.39 1.53 .66 -.31
difference -.43 .72 1.38 1.23exp (odds ratio) .65 2.05 3.99 3.42
19
![Page 20: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/20.jpg)
Observed Proportions across Time by Condition
• model is not linear in terms of probabilities
P (Yij = 1 | υ0i) =1
1 + exp−
x′ijβ + υ0i
20
![Page 21: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/21.jpg)
Observed Logits across Time by Condition
model is linear in terms of logits: log
P (Yij = 1 | υ0i)1− P (Yij = 1 | υ0i)
= x′ijβ + υ0i
21
![Page 22: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/22.jpg)
Within-Subjects / Between-Subjects components
Within-subjects model - level 1 (j = 1, . . . , ni obs)
logitij = b0i + b1i√Weekj
Between-subjects model - level 2 (i = 1, . . . , N subjects)
b0i = β0 + β2Grpi + υ0i
b1i = β1 + β3Grpi
υ0i ∼ NID(0, σ2υ)
22
![Page 23: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/23.jpg)
Under “File” click on “Open Existing Model Setup”
Open C:\SuperMixEn Examples\Workshop\Binary\schizb1.mum(or C:\SuperMixEn Student Examples\Workshop\Binary\schizb1.mum)
23
![Page 24: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/24.jpg)
Note “Dependent Variable Type” should be “binary”
24
![Page 25: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/25.jpg)
25
![Page 26: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/26.jpg)
Note “Optimization Method” should be “adaptive quadrature”
26
![Page 27: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/27.jpg)
27
![Page 28: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/28.jpg)
28
![Page 29: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/29.jpg)
SuperMix is FAST for full-likelihood estimation of non-normalmodels, and up to three level models
29
![Page 30: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/30.jpg)
Empirical Bayes Estimates of Random EffectsSelect “Analysis” > “View Level-2 Bayes Results”
ID, random effect number, estimate, variance, name
30
![Page 31: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/31.jpg)
Close output, select “File” > “Model-based Graphs” > “Equations”
31
![Page 32: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/32.jpg)
32
![Page 33: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/33.jpg)
Estimated (subject-specific) Logits across Time byCondition: random-intercepts model
log
P (Yij = 1 | υ0i)1− P (Yij = 1 | υ0i)
= 5.39− .025Di − 1.50Tj − 1.01 (Di × Tj) + υ0i
υ0i ∼ NID(0, σ̂2υ = 4.48)
β̂ change in (conditional) logit due to x for subjects with the samevalue of υ0i (the above plot is for υ0i = 0)
33
![Page 34: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/34.jpg)
Random-intercepts Logistic Regression
logitij = x′ijβ + υ0i
• every subject has their own propensity for response (υ0i)
• the influence of covariates x is determined controlling (oradjusting) for the subject effect
• the covariance structure, or dependency, of the repeatedobservations is explicitly modeled
34
![Page 35: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/35.jpg)
β0 = log odds of response for a typical subject with x = 0 andυ0i = 0
β = log odds ratio for response associated with unit changes inx for the same subject value υ0i∗ referred to as “subject-specific”∗ how a subject’s response probability depends on x
σ2υ = degree of heterogeneity across subjects in the probability
of response not attributable to x
• most useful when the objective is to make inference aboutsubjects rather than the population average
• interest is in the heterogeneity of subjects
35
![Page 36: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/36.jpg)
Estimated Subject-Specific Probabilites
P (Yij = 1 | υ0i) =1
1 + exp [− (5.39− .03Di − 1.50Tj − 1.01DiTj + υ0i)]
where υ0i =
−1συ
1συand σ̂υ = 2.12
36
![Page 37: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/37.jpg)
Instead of the mixed model, consider the following marginal model
log
P (Yij = 1)
1− P (Yij = 1)
= x′ijβpa
• βpa have marginal or “population-average” interpretation
• Not conditional on subject random effects
• Estimates from a GEE model are of this type
• βss 6= βpa unless random effect variance(s) equal 0 (or β = 0)
⇒ Can one obtain βpa from βss?
37
![Page 38: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/38.jpg)
For a random-intercept model with estimates β̂ss
and σ̂2υ
β̂pa ≈ β̂ss/
√√√√√√√√√σ̂2υ + π2/3
π2/3
• π2/3 is the variance of the standard logistic distribution
• square-root term on the right-hand side can be viewed as the“marginalization” factor; transforms subject-specific parametersinto their population-averaged counterparts
• In a random-intercepts model, the variance is equal across time;marginalization factor is equal across time and is a scalar
• For models with multiple random effects, this is not the case andso there is no simple relationship
38
![Page 39: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/39.jpg)
• Hedeker, du Toit, Demirtas, Gibbons (2014) describe a generalmarginalization approach that has been implemented in theupdate of Supermix to yield Population Average Estimates
39
![Page 40: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/40.jpg)
log
P (Yij = 1)
1− P (Yij = 1)
= 3.54− .055Di − 1.05Tj − .60 (Di × Tj)
⇒ these are the Population Average Estimates from Supermix
40
![Page 41: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/41.jpg)
P (Yij = 1) =1
1 + exp[−
(3.54− .055Di − 1.05Tj − .60DiTj
)]
⇒ these are the Population Average Estimates from Supermix
41
![Page 42: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/42.jpg)
Within-Subjects / Between-Subjects components
Within-subjects model - level 1 (j = 1, . . . , ni obs)
logitij = b0i + b1i√Weekj
Between-subjects model - level 2 (i = 1, . . . , N subjects)
b0i = β0 + β2Grpi + υ0i
b1i = β1 + β3Grpi + υ1i
υi ∼ NID(0,Συ)
42
![Page 43: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/43.jpg)
Under “File” click on “Open Existing Model Setup”
Open C:\SuperMixEn Examples\Workshop\Binary\schizb2.mum(or C:\SuperMixEn Student Examples\Workshop\Binary\schizb2.mum)
43
![Page 44: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/44.jpg)
Note “Dependent Variable Type” should be “binary”
44
![Page 45: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/45.jpg)
SqrtWeek is a level-2 (subject) random effect
45
![Page 46: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/46.jpg)
Note “Optimization Method” should be “adaptive quadrature”
46
![Page 47: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/47.jpg)
⇒ Comparing models: H0 : σ2υ1 = συ01 = 0; χ22 = 1249.73− 1227.76 = 21.97, p < .001
47
![Page 48: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/48.jpg)
⇒ Supermix is FAST for a full-likelihood solution using bivariate numerical integration involving
100 quadrature points!
48
![Page 49: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/49.jpg)
Empirical Bayes Estimates of Random EffectsSelect “Analysis” > “View Level-2 Bayes Results”
ID, random effect number, estimate, variance, name
49
![Page 50: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/50.jpg)
Close output, select “File” > “Model-based Graphs” > “Equations”
50
![Page 51: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/51.jpg)
51
![Page 52: Don Hedeker Department of Public Health Sciences ... › images › pdfs › Binary_Long.pdfMixed Models for Longitudinal Binary Outcomes Don Hedeker Department of Public Health Sciences](https://reader034.fdocuments.in/reader034/viewer/2022042323/5f0d55f77e708231d439d798/html5/thumbnails/52.jpg)
Summary - mixed models for binary outcomes
• link functions: logistic, probit, log-log, complementary log-log
• multiple random effects (correlated or independent) for up to3-level models
• fast full-likelihood estimation using adaptive Gauss-Hermitequadrature
• subject-specific and population-average estimates and inference
• discrete/grouped time survival analysis via person-period dataset
• Advanced > Level-2 (Co)variance Patterns > Unidimensional
– varying ICC model for MZ/DZ twin pair data: create dummyvariables MZ and DZ, specify both as random effects, select“Unidimensional”
– Item-response theory (IRT) models: create indicator variablesfor items, specify all as random effects, select “Unidimensional”
52