Varying Intercepts Models

Hierarchical Models
Regression
Modeling grouped or hierarchical data by allowing the intercept to vary for each group.

General Principles

To model the relationship between a dependent variable and an independent variable while allowing for different intercepts across groups or clusters, we can use a Varying Intercepts model. This approach is particularly useful when data are grouped (e.g., by subject, location, or time period) and we expect the baseline level of the outcome to vary across these groups.

Considerations

Note

  • We have the same considerations as for Regression for a continuous variable.

  • The main idea of varying intercepts is to generate an intercept for each group, allowing each group to start at different levels. Thus, the intercept \alpha_k is defined uniquely for each of the K declared groups.

  • In the code below, the intercept alpha for each of the k declared groups shares two priors, a_bar and sigma, which are respectively modeled by a Normal and an Exponential distribution.

Example

Below is an example code snippet demonstrating Bayesian regression with varying intercepts using the Bayesian Inference (BI) package. The data consists of a dependent variable representing individuals’ survival (surv) and an independent categorical variable (tank), which indicates the tank where the individual was born, with a total of 48 tanks. This example is based on McElreath (2018).

Code 
from BI import bi, jnp

# Setup device------------------------------------------------
m = bi(platform='cpu')

# Import Data & Data Manipulation ------------------------------------------------
# Import
from importlib.resources import files
data_path = m.load.reedfrogs(only_path=True)
m.data(data_path, sep=';') 
# Manipulate
m.df["tank"] = jnp.arange(m.df.shape[0]) 

# Define model ------------------------------------------------
def model(tank, surv, density):
    sigma = m.dist.exponential( 1,  name = 'sigma')
    a_bar = m.dist.normal( 0., 1.5,  name = 'a_bar')
    alpha = m.dist.normal( a_bar, sigma, shape= tank.shape, name = 'alpha')
    p = alpha[tank]
    m.dist.binomial(total_count = density, logits = p, obs=surv)

# Run sampler ------------------------------------------------
m.fit(model) 

# Diagnostic ------------------------------------------------
m.summary()
jax.local_device_count 32
  0%|          | 0/1000 [00:00<?, ?it/s]warmup:   0%|          | 1/1000 [00:00<08:02,  2.07it/s, 1 steps of size 2.34e+00. acc. prob=0.00]warmup:  22%|β–ˆβ–ˆβ–       | 222/1000 [00:00<00:01, 505.55it/s, 15 steps of size 4.65e-01. acc. prob=0.78]warmup:  50%|β–ˆβ–ˆβ–ˆβ–ˆβ–‰     | 499/1000 [00:00<00:00, 1061.86it/s, 7 steps of size 2.27e-01. acc. prob=0.79]sample:  76%|β–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–‹  | 763/1000 [00:00<00:00, 1470.40it/s, 15 steps of size 4.21e-01. acc. prob=0.90]sample: 100%|β–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆ| 1000/1000 [00:00<00:00, 1147.33it/s, 7 steps of size 4.21e-01. acc. prob=0.89]
arviz - WARNING - Shape validation failed: input_shape: (1, 500), minimum_shape: (chains=2, draws=4)
mean sd hdi_5.5% hdi_94.5% mcse_mean mcse_sd ess_bulk ess_tail r_hat
a_bar 1.35 0.27 0.95 1.79 0.01 0.01 467.13 297.56 NaN
alpha[0] 2.13 0.85 0.69 3.34 0.04 0.04 514.75 388.20 NaN
alpha[1] 3.09 1.09 1.30 4.47 0.05 0.05 565.53 436.09 NaN
alpha[2] 0.99 0.68 0.06 2.14 0.03 0.04 676.80 327.00 NaN
alpha[3] 3.12 1.10 1.27 4.66 0.05 0.06 594.03 428.01 NaN
alpha[4] 2.15 0.89 0.82 3.57 0.03 0.04 726.38 463.33 NaN
alpha[5] 2.19 0.91 0.87 3.66 0.04 0.05 771.96 371.60 NaN
alpha[6] 3.12 1.07 1.57 4.72 0.05 0.04 520.47 347.62 NaN
alpha[7] 2.14 0.87 0.73 3.38 0.03 0.05 1221.40 424.56 NaN
alpha[8] -0.16 0.62 -1.02 0.94 0.02 0.04 811.97 219.25 NaN
alpha[9] 2.10 0.86 0.67 3.33 0.04 0.04 513.47 370.85 NaN
alpha[10] 0.99 0.69 -0.13 2.03 0.02 0.03 927.72 411.04 NaN
alpha[11] 0.62 0.65 -0.49 1.61 0.03 0.03 561.63 290.42 NaN
alpha[12] 1.05 0.72 -0.12 2.05 0.03 0.04 708.09 317.39 NaN
alpha[13] 0.18 0.62 -0.83 1.18 0.02 0.03 680.04 406.36 NaN
alpha[14] 2.16 0.84 0.69 3.32 0.03 0.04 843.95 375.45 NaN
alpha[15] 2.15 0.91 0.80 3.65 0.04 0.04 499.23 283.02 NaN
alpha[16] 2.91 0.79 1.42 3.86 0.03 0.04 716.71 353.12 NaN
alpha[17] 2.40 0.67 1.30 3.38 0.03 0.04 727.24 247.66 NaN
alpha[18] 2.03 0.60 1.10 2.91 0.03 0.03 642.62 326.17 NaN
alpha[19] 3.75 0.95 2.17 5.10 0.04 0.04 582.43 361.94 NaN
alpha[20] 2.35 0.61 1.41 3.24 0.02 0.03 824.43 261.58 NaN
alpha[21] 2.36 0.66 1.34 3.37 0.02 0.03 761.56 320.99 NaN
alpha[22] 2.40 0.62 1.37 3.31 0.03 0.03 573.35 352.38 NaN
alpha[23] 1.70 0.53 0.93 2.56 0.02 0.02 839.61 423.04 NaN
alpha[24] -1.03 0.45 -1.69 -0.33 0.02 0.02 731.40 365.09 NaN
alpha[25] 0.16 0.42 -0.48 0.83 0.02 0.02 651.44 369.19 NaN
alpha[26] -1.41 0.45 -2.15 -0.71 0.02 0.02 597.70 314.86 NaN
alpha[27] -0.48 0.42 -1.08 0.19 0.02 0.02 537.11 388.83 NaN
alpha[28] 0.16 0.40 -0.61 0.67 0.02 0.02 431.36 392.36 NaN
alpha[29] 1.46 0.48 0.75 2.27 0.02 0.02 755.76 404.80 NaN
alpha[30] -0.62 0.43 -1.18 0.20 0.02 0.02 706.26 394.05 NaN
alpha[31] -0.30 0.43 -0.90 0.41 0.02 0.02 692.89 408.34 NaN
alpha[32] 3.20 0.76 2.04 4.48 0.03 0.03 563.22 379.37 NaN
alpha[33] 2.70 0.64 1.83 3.88 0.02 0.03 758.59 428.71 NaN
alpha[34] 2.74 0.65 1.65 3.71 0.03 0.03 545.98 387.72 NaN
alpha[35] 2.06 0.49 1.36 2.94 0.02 0.02 721.44 392.52 NaN
alpha[36] 2.09 0.57 1.25 2.95 0.02 0.03 732.34 371.66 NaN
alpha[37] 3.89 0.95 2.29 5.11 0.04 0.05 508.53 294.22 NaN
alpha[38] 2.70 0.70 1.66 3.76 0.03 0.05 696.43 275.91 NaN
alpha[39] 2.38 0.54 1.46 3.19 0.02 0.03 588.18 352.03 NaN
alpha[40] -1.77 0.47 -2.50 -1.03 0.02 0.02 965.46 269.98 NaN
alpha[41] -0.59 0.31 -1.09 -0.13 0.01 0.01 729.57 401.86 NaN
alpha[42] -0.44 0.37 -1.04 0.16 0.01 0.02 744.12 373.34 NaN
alpha[43] -0.33 0.36 -0.90 0.23 0.01 0.02 818.64 313.40 NaN
alpha[44] 0.57 0.34 0.07 1.09 0.02 0.01 487.25 426.63 NaN
alpha[45] -0.58 0.38 -1.25 -0.04 0.01 0.02 1084.06 421.32 NaN
alpha[46] 2.06 0.46 1.34 2.75 0.02 0.02 569.86 389.45 NaN
alpha[47] 0.02 0.35 -0.50 0.61 0.01 0.02 710.85 353.56 NaN
sigma 1.62 0.21 1.30 1.92 0.01 0.01 339.24 267.98 NaN
Code 
from BI import bi, jnp


# Setup device------------------------------------------------
m = bi(platform='cpu')

# Import Data & Data Manipulation ------------------------------------------------
# Import
from importlib.resources import files
data_path =  m.load.reedfrogs(only_path=True)
m.data(data_path, sep=';') 
# Manipulate
m.df["tank"] = jnp.arange(m.df.shape[0]) 

# Define model ------------------------------------------------
def model(tank, surv, density):
    alpha = m.effects.varying_intercept(N_groups=48,group_id=tank, group_name = 'tank')
    m.dist.binomial(total_count = density, logits = alpha, obs=surv)

# Run sampler ------------------------------------------------
m.fit(model) 

# Diagnostic ------------------------------------------------
m.summary()
jax.local_device_count 32
  0%|          | 0/1000 [00:00<?, ?it/s]warmup:   0%|          | 1/1000 [00:00<07:27,  2.23it/s, 1 steps of size 2.34e+00. acc. prob=0.00]warmup:  20%|β–ˆβ–‰        | 196/1000 [00:00<00:01, 473.55it/s, 15 steps of size 5.67e-01. acc. prob=0.78]warmup:  47%|β–ˆβ–ˆβ–ˆβ–ˆβ–‹     | 466/1000 [00:00<00:00, 1044.43it/s, 15 steps of size 5.59e-01. acc. prob=0.79]sample:  75%|β–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–  | 746/1000 [00:00<00:00, 1515.14it/s, 7 steps of size 4.53e-01. acc. prob=0.88] sample: 100%|β–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆ| 1000/1000 [00:00<00:00, 1193.10it/s, 7 steps of size 4.53e-01. acc. prob=0.88]
arviz - WARNING - Shape validation failed: input_shape: (1, 500), minimum_shape: (chains=2, draws=4)
mean sd hdi_5.5% hdi_94.5% mcse_mean mcse_sd ess_bulk ess_tail r_hat
global_intercept_tank 1.36 0.26 0.94 1.77 0.01 0.01 563.88 182.64 NaN
intercept_tank[0] 2.10 0.86 0.72 3.38 0.03 0.04 811.23 369.49 NaN
intercept_tank[1] 3.04 1.11 1.31 4.69 0.05 0.06 592.21 290.26 NaN
intercept_tank[2] 1.00 0.65 -0.06 2.06 0.03 0.04 647.60 343.02 NaN
intercept_tank[3] 3.06 1.10 1.16 4.67 0.05 0.07 547.75 178.68 NaN
intercept_tank[4] 2.14 0.89 0.80 3.44 0.03 0.05 953.17 307.45 NaN
intercept_tank[5] 2.15 0.89 0.91 3.72 0.03 0.04 850.84 421.61 NaN
intercept_tank[6] 3.13 1.11 1.40 4.93 0.06 0.06 455.95 282.69 NaN
intercept_tank[7] 2.18 0.89 1.00 3.73 0.04 0.04 504.63 338.18 NaN
intercept_tank[8] -0.16 0.62 -1.02 0.89 0.02 0.03 602.28 387.08 NaN
intercept_tank[9] 2.15 0.92 0.81 3.47 0.04 0.06 588.68 335.17 NaN
intercept_tank[10] 1.00 0.70 0.03 2.17 0.02 0.04 921.27 305.21 NaN
intercept_tank[11] 0.58 0.61 -0.31 1.64 0.02 0.03 718.11 320.40 NaN
intercept_tank[12] 1.07 0.69 -0.06 2.08 0.03 0.03 511.78 369.19 NaN
intercept_tank[13] 0.20 0.60 -0.69 1.20 0.02 0.02 646.21 339.77 NaN
intercept_tank[14] 2.12 0.83 0.90 3.52 0.04 0.04 563.92 353.53 NaN
intercept_tank[15] 2.15 0.91 0.69 3.50 0.04 0.05 565.23 355.31 NaN
intercept_tank[16] 2.87 0.76 1.74 4.10 0.03 0.03 918.26 411.31 NaN
intercept_tank[17] 2.42 0.68 1.36 3.40 0.03 0.05 642.69 209.59 NaN
intercept_tank[18] 2.03 0.60 0.95 2.83 0.03 0.02 460.37 346.04 NaN
intercept_tank[19] 3.73 1.04 2.02 5.18 0.05 0.05 587.35 381.98 NaN
intercept_tank[20] 2.38 0.64 1.47 3.46 0.03 0.03 592.59 386.77 NaN
intercept_tank[21] 2.39 0.69 1.30 3.37 0.03 0.03 622.27 322.91 NaN
intercept_tank[22] 2.42 0.65 1.26 3.35 0.03 0.04 435.39 247.89 NaN
intercept_tank[23] 1.69 0.53 0.88 2.48 0.02 0.03 676.05 393.95 NaN
intercept_tank[24] -1.00 0.48 -1.69 -0.26 0.02 0.02 792.69 344.82 NaN
intercept_tank[25] 0.14 0.40 -0.44 0.81 0.02 0.02 570.16 385.30 NaN
intercept_tank[26] -1.42 0.48 -2.04 -0.54 0.02 0.03 699.69 349.23 NaN
intercept_tank[27] -0.50 0.42 -1.12 0.17 0.02 0.02 482.27 381.98 NaN
intercept_tank[28] 0.15 0.39 -0.39 0.82 0.02 0.02 655.74 295.17 NaN
intercept_tank[29] 1.44 0.48 0.66 2.19 0.02 0.02 491.85 336.45 NaN
intercept_tank[30] -0.62 0.42 -1.24 0.11 0.02 0.02 601.10 369.15 NaN
intercept_tank[31] -0.28 0.44 -0.96 0.37 0.02 0.02 587.55 285.26 NaN
intercept_tank[32] 3.18 0.72 2.05 4.19 0.03 0.03 481.03 336.93 NaN
intercept_tank[33] 2.67 0.62 1.57 3.49 0.02 0.03 735.44 289.00 NaN
intercept_tank[34] 2.75 0.62 1.76 3.73 0.03 0.03 388.61 347.85 NaN
intercept_tank[35] 2.06 0.47 1.27 2.80 0.02 0.02 840.45 343.02 NaN
intercept_tank[36] 2.09 0.53 1.32 2.91 0.02 0.02 756.53 393.66 NaN
intercept_tank[37] 3.90 0.95 2.47 5.27 0.05 0.04 336.15 304.76 NaN
intercept_tank[38] 2.70 0.67 1.72 3.72 0.03 0.04 665.74 311.83 NaN
intercept_tank[39] 2.34 0.53 1.45 3.15 0.02 0.03 738.32 281.88 NaN
intercept_tank[40] -1.78 0.45 -2.48 -1.06 0.01 0.02 913.76 357.43 NaN
intercept_tank[41] -0.58 0.34 -1.14 -0.07 0.01 0.01 730.83 425.09 NaN
intercept_tank[42] -0.44 0.39 -1.03 0.21 0.01 0.02 693.24 438.47 NaN
intercept_tank[43] -0.33 0.34 -0.86 0.20 0.01 0.02 549.94 352.19 NaN
intercept_tank[44] 0.58 0.33 0.06 1.09 0.01 0.01 653.17 339.77 NaN
intercept_tank[45] -0.59 0.38 -1.24 -0.02 0.02 0.02 620.34 294.19 NaN
intercept_tank[46] 2.03 0.48 1.30 2.77 0.02 0.03 657.69 351.35 NaN
intercept_tank[47] 0.01 0.33 -0.52 0.54 0.01 0.01 735.81 303.89 NaN
sd_tank 1.61 0.20 1.28 1.92 0.01 0.01 223.71 348.72 NaN 
library(BayesianInference)

# setup platform------------------------------------------------
m=importBI(platform='cpu')

# Import data ------------------------------------------------
m$data(m$load$reedfrogs(only_path=T), sep=';')
m$df$tank = c(0:(nrow(m$df)-1)) # Manipulate
m$data_to_model(list('tank', 'surv', 'density')) # Manipulate
m$data_on_model$tank = m$data_on_model$tank$astype(jnp$int32) # Manipulate
m$data_on_model$surv = m$data_on_model$surv$astype(jnp$int32) # Manipulate


# Define model ------------------------------------------------
model <- function(tank, surv, density){
  # Parameter prior distributions
  sigma = bi.dist.exponential( 1,  name = 'sigma',shape=c(1))
  a_bar =  bi.dist.normal(0, 1.5, name='a_bar',shape=c(1))
  alpha = bi.dist.normal(a_bar, sigma, name='alpha', shape =c(48))
  p = alpha[tank]
  # Likelihood
  m$dist$binomial(total_count = density, logits = p, obs=surv)
} 

# Run MCMC ------------------------------------------------
m$fit(model) # Optimize model parameters through MCMC sampling

# Summary ------------------------------------------------
m$summary() # Get posterior distribution

using BayesianInference

# Setup device------------------------------------------------
m = importBI(platform="cpu")

# Import Data & Data Manipulation ------------------------------------------------
# Import
data_path = m.load.reedfrogs(only_path = true)
m.data(data_path, sep=';')
m.df["tank"] = jnp.arange(m.df.shape[0]) 

# Define model ------------------------------------------------
@BI function model(tank, surv, density)
    alpha = m.effects.varying_intercept(N_groups=48,group_id=tank, group_name = "tank")
    m.dist.binomial(total_count = density, logits = alpha, obs=surv)
end

# Run mcmc ------------------------------------------------
m.fit(model)  # Optimize model parameters through MCMC sampling

# Summary ------------------------------------------------
m.summary() # Get posterior distributions

Mathematical Details

We model the relationship between the independent variable X and the outcome variable Y while accounting for varying intercepts \alpha for each group where k(i) give us group belonging for observation i, using the following equation:

Y_{i} \sim \text{Normal}(\mu_{i}, \sigma) \mu_{i} = \alpha_{[k(i)]} + \beta X_{i}

\beta \sim \text{Normal}(0, 1)

\sigma \sim \text{Exponential}(1)

\alpha_{[k]} \sim \text{Normal}(\bar{\alpha}, \varsigma)

\bar{\alpha} \sim \text{Normal}(0, 1) \varsigma \sim \text{Exponential}(1)

Where:

  • Y_{i} is the outcome variable for observation i.

  • \alpha_{[k(i)]} is the varying intercept corresponding to the group k of observation i%.

  • \beta is the regression coefficient.

  • \sigma is a standard deviation parameter, which here has an Exponential prior that constrains it to be positive.

  • \bar{\alpha} is the overall mean intercept.

  • \varsigma is the variance of the intercepts across groups.

Notes

Note

  • We can apply multiple variables similarly to Chapter 2.

  • We can apply interaction terms similarly to Chapter 3.

  • We can apply categorical variables similarly to Chapter 4.

  • We can apply varying intercepts with any distribution developed in previous chapters.

Reference(s)

McElreath, Richard. 2018. Statistical Rethinking: A Bayesian course with examples in R and Stan. Chapman; Hall/CRC.