Skip to contents

Overview

The stable trait, autoregressive trait, and state (STARTS) model divides a variable measured repeatedly on the same individuals into three parts (Kenny & Zautra, 1995, 2001). Take life satisfaction measured once a year. Part of it is a stable trait that does not change over the study. Part of it is an autoregressive trait that changes slowly: a good year tends to be followed by a fairly good year. The rest is a state that belongs to one occasion only, such as the mood on the day of the survey or measurement error. The model estimates how much of the variance belongs to each part and how strongly the autoregressive part carries over.

The STARTS model is usually fitted as a structural equation model (SEM). It can also be written as a state space model: the stable trait is a latent state that never changes, the autoregressive trait is a latent state with first-order dynamics, and the state part is the measurement error.

This tutorial does three things:

  1. It writes the STARTS model as a state space model.
  2. It simulates data from the model with SimSSMFixed().
  3. It fits the model to the simulated data as a state space model with dynr and OpenMx and as a structural equation model (SEM) with lavaan and OpenMx, and compares every set of estimates with the values used to simulate the data.

No prior knowledge of state space models is needed.

Model

The measurement model is given by Y_{i, t} = \left( \begin{array}{cc} 1 & 1 \\ \end{array} \right) \left( \begin{array}{c} \eta_{1_{i, t}} \\ \eta_{2_{i, t}} \\ \end{array} \right) + \varepsilon_{i, t}, \quad \mathrm{with} \quad \varepsilon_{i, t} \sim \mathcal{N} \left( 0, \theta \right) where Y_{i, t} is the observed variable at time t and individual i, \eta_{1_{i, t}} is the stable trait, \eta_{2_{i, t}} is the autoregressive trait, and \varepsilon_{i, t} is the state. \theta is the variance of the state.

In words, the observed score is the sum of the three parts. The state \varepsilon_{i, t} is new on every occasion and does not carry over, so it plays the role of measurement error in a state space model.

The dynamic structure is given by \left( \begin{array}{c} \eta_{1_{i, t}} \\ \eta_{2_{i, t}} \\ \end{array} \right) = \left( \begin{array}{cc} 1 & 0 \\ 0 & \beta_{22} \\ \end{array} \right) \left( \begin{array}{c} \eta_{1_{i, t - 1}} \\ \eta_{2_{i, t - 1}} \\ \end{array} \right) + \left( \begin{array}{c} 0 \\ \zeta_{2_{i, t}} \\ \end{array} \right), \quad \mathrm{with} \quad \zeta_{2_{i, t}} \sim \mathcal{N} \left( 0, \psi_{22} \right) .

In words, the stable trait stays the same over time. The autoregressive trait keeps a proportion \beta_{22} of its previous value and adds a new part with variance \psi_{22}. The closer \beta_{22} is to 1, the more slowly the autoregressive trait changes.

The initial condition is given by \boldsymbol{\eta}_{i, 0} \sim \mathcal{N} \left( \left( \begin{array}{c} \mu_{1} \\ 0 \\ \end{array} \right), \left( \begin{array}{cc} \sigma_{11} & 0 \\ 0 & \sigma_{22} \\ \end{array} \right) \right) .

Here, \mu_{1} is the average stable trait and \sigma_{11} is how much individuals differ in it. The autoregressive trait has mean zero, because the stable trait carries the mean, and \sigma_{22} is its variance at the first occasion. The two traits are uncorrelated.

Diagrams

Path diagram of the STARTS model.
Path diagram of the STARTS model.

The STARTS model as a path diagram, for the first three occasions (t = 0, 1, 2). The stable trait \eta_{1}, with variance \sigma_{11}, loads on every occasion with a loading of 1. The autoregressive trait \eta_{2}, with variance \sigma_{22} at t = 0, carries over from one occasion to the next with the autoregressive effect \beta_{22} and gets a new part \zeta_{2}, with variance \psi_{22}, on each occasion. The state \varepsilon, with variance \theta, is new on every occasion. A loop on a variable is its variance and a curved two-headed arrow a covariance.

State space form of the STARTS model.
State space form of the STARTS model.

The same model in state space form, for the first three occasions, with every element of the state drawn. The stable trait is a state with a transition of 1 and no process noise, so \eta_{1, t} = \eta_{1} for all t; the autoregressive trait \eta_{2, t} receives the process noise \zeta_{2, t} from t = 1 on; the state \varepsilon_{t} is the measurement error of the state space model. Each Y_{t} is the sum of the three, all with a loading of 1. At t = 0, the initial condition \boldsymbol{\eta}_{0} \sim \mathcal{N} \left( \boldsymbol{\mu}_{0}, \boldsymbol{\Sigma}_{0} \right) gives the variances \sigma_{11} and \sigma_{22} at t = 0; the two traits do not covary. Below the diagram, \boldsymbol{\beta} carries the state from one occasion to the next, \boldsymbol{\Lambda} maps the state to the observed variable, \boldsymbol{\Psi} is the covariance matrix of the process noise \boldsymbol{\zeta}_{t}, and \boldsymbol{\Theta} is the variance of the state. SimSSMFixed() takes beta, lambda, mu0, and the Cholesky factors of \boldsymbol{\Psi}, \boldsymbol{\Theta}, and \boldsymbol{\Sigma}_{0} (psi_l, theta_l, and sigma0_l).

Data Generation

Notation

Let t = 10 be the number of time points and n = 5000 be the number of individuals. The STARTS model needs many occasions to tell the three parts apart, so this example has more occasions than the other vignettes.

Let the stable trait have mean \mu_{1} = 3 and variance \sigma_{11} = 0.5. Let the autoregressive effect be \beta_{22} = 0.6, the process noise variance \psi_{22} = 0.32, and the state variance \theta = 0.3.

The autoregressive trait starts from its stationary distribution, so its variance at the first occasion is the variance it keeps at every later occasion,

\sigma_{22} = \frac{\psi_{22}}{1 - \beta_{22}^{2}} = 0.5 .

With these values, the stable trait, the autoregressive trait, and the state contribute 0.5, 0.5, and 0.3 to the variance of the observed score on every occasion.

R Function Arguments

The simulation functions take covariance matrices as lower triangular Cholesky factors (the arguments ending in _l). Because the stable trait has no process noise, psi has a row and a column of zeros, and its Cholesky factor is written directly.

n
#> [1] 5000
time
#> [1] 10
mu0
#> [1] 3 0
sigma0
#>      [,1] [,2]
#> [1,]  0.5  0.0
#> [2,]  0.0  0.5
sigma0_l # sigma0_l <- t(chol(sigma0))
#>           [,1]      [,2]
#> [1,] 0.7071068 0.0000000
#> [2,] 0.0000000 0.7071068
alpha
#> [1] 0 0
beta
#>      [,1] [,2]
#> [1,]    1  0.0
#> [2,]    0  0.6
psi
#>      [,1] [,2]
#> [1,]    0 0.00
#> [2,]    0 0.32
psi_l
#>      [,1]      [,2]
#> [1,]    0 0.0000000
#> [2,]    0 0.5656854
nu
#> [1] 0
lambda
#>      [,1] [,2]
#> [1,]    1    1
theta
#>      [,1]
#> [1,]  0.3
theta_l # theta_l <- t(chol(theta))
#>           [,1]
#> [1,] 0.5477226

Visualizing the Dynamics Without Process Noise and Measurement Error (n = 5 with Different Initial Condition)

Without process noise and the state, each individual’s autoregressive trait decays toward zero, so each trajectory settles at the individual’s stable trait.

Using the SimSSMFixed Function from the simStateSpace Package to Simulate Data

Now we add the process noise and the state and simulate the full data set. as.data.frame() returns the data in long format, one row per individual and time point, with the columns id, time, and y1.

library(simStateSpace)
sim <- SimSSMFixed(
  n = n,
  time = time,
  mu0 = mu0,
  sigma0_l = sigma0_l,
  alpha = alpha,
  beta = beta,
  psi_l = psi_l,
  nu = nu,
  lambda = lambda,
  theta_l = theta_l,
  type = 0
)
data <- as.data.frame(sim)
head(data)
#>   id time       y1
#> 1  1    0 2.724102
#> 2  1    1 2.554146
#> 3  1    2 2.834498
#> 4  1    3 1.952812
#> 5  1    4 1.715812
#> 6  1    5 1.798422
summary(data)
#>        id            time           y1        
#>  Min.   :   1   Min.   :0.0   Min.   :-1.844  
#>  1st Qu.:1251   1st Qu.:2.0   1st Qu.: 2.215  
#>  Median :2500   Median :4.5   Median : 2.993  
#>  Mean   :2500   Mean   :4.5   Mean   : 2.996  
#>  3rd Qu.:3750   3rd Qu.:7.0   3rd Qu.: 3.764  
#>  Max.   :5000   Max.   :9.0   Max.   : 7.797
plot(sim)

Model Fitting

The same simulated data are fitted as a state space model with dynr and OpenMx and as a structural equation model (SEM) with lavaan and OpenMx. All of them estimate the parameters by maximum likelihood. The section Population Values and Estimates compares their estimates with the population values.

dynr

We fit the model with dynr, which estimates state space models by maximum likelihood using the Kalman filter.

A dynr model is built from pieces that match the equations above: the data, the initial condition, the measurement model, the dynamic model, and the noise. Each parameter gets a name (for example, beta_2_2 for \beta_{22}), and the same name used in two places means one shared parameter. The starting values are set to the population values for convenience; with real data, they would be rough guesses.

Prepare Data

Tell dynr which columns hold the individual, the time, and the observed variables.

dynr_data <- dynr::dynr.data(
  dataframe = data,
  id = "id",
  time = "time",
  observed = "y1"
)

Prepare Initial Condition

The mean of the stable trait and the variances of both traits at the first occasion are estimated. The mean of the autoregressive trait and the covariance of the two traits are fixed at zero.

dynr_initial <- dynr::prep.initial(
  values.inistate = mu0,
  params.inistate = c("mu0_1_1", "fixed"),
  values.inicov = sigma0,
  params.inicov = matrix(
    data = c(
      "sigma0_1_1", "fixed",
      "fixed", "sigma0_2_2"
    ),
    nrow = 2
  )
)

Prepare Measurement Model

The loadings \left( 1, 1 \right) are fixed.

dynr_measurement <- dynr::prep.measurement(
  values.load = lambda,
  params.load = matrix(data = "fixed", nrow = 1, ncol = 2),
  state.names = c("eta_1", "eta_2"),
  obs.names = "y1"
)

Prepare Dynamic Process

The stable trait carries over unchanged; the autoregressive trait has the autoregressive effect \beta_{22}.

dynr_dynamics <- dynr::prep.formulaDynamics(
  formula = list(
    eta_1 ~ eta_1,
    eta_2 ~ beta_2_2 * eta_2
  ),
  startval = c(beta_2_2 = beta[2, 2]),
  isContinuousTime = FALSE
)

Prepare Process Noise

Only the autoregressive trait has process noise. The state variance \theta is estimated.

dynr_noise <- dynr::prep.noise(
  values.latent = psi,
  params.latent = matrix(
    data = c(
      "fixed", "fixed",
      "fixed", "psi_2_2"
    ),
    nrow = 2
  ),
  values.observed = theta,
  params.observed = "theta"
)

Prepare the Model

The pieces are put together into one model. dynr writes the model as C code to outfile and compiles it.

model <- dynr::dynr.model(
  data = dynr_data,
  initial = dynr_initial,
  measurement = dynr_measurement,
  dynamics = dynr_dynamics,
  noise = dynr_noise,
  outfile = "starts.c"
)

The plot below shows the model equations with the parameter names.

Fit the Model

results <- dynr::dynr.cook(
  model,
  debug_flag = TRUE,
  verbose = FALSE
)
#> [1] "Get ready!!!!"
#> using C compiler: ‘gcc (Ubuntu 15.2.0-16ubuntu1) 15.2.0’
#> Optimization function called.
#> Starting Hessian calculation ...
#> Finished Hessian calculation.
#> Original exit flag:  3 
#> Modified exit flag:  3 
#> Optimization terminated successfully: ftol_rel or ftol_abs was reached. 
#> Original fitted parameters:  0.6027093 -1.105952 -1.224083 2.996103 -0.6966334 
#> -0.713488 
#> 
#> Transformed fitted parameters:  0.6027093 0.3308957 0.2940272 2.996103 
#> 0.4982599 0.4899323 
#> 
#> Doing end processing
#> Successful trial
#> Total Time: 25.94517 
#> Backend Time: 25.38211

Summary

The summary lists each estimate with its standard error, t value, confidence interval, and p value.

summary(results)
#> Coefficients:
#>            Estimate Std. Error t value ci.lower ci.upper            Pr(>|t|)
#> beta_2_2    0.60271    0.01815   33.21  0.56714  0.63828 <0.0000000000000002
#> psi_2_2     0.33090    0.01398   23.68  0.30350  0.35829 <0.0000000000000002
#> theta       0.29403    0.01168   25.18  0.27114  0.31692 <0.0000000000000002
#> mu0_1_1     2.99610    0.01174  255.15  2.97309  3.01912 <0.0000000000000002
#> sigma0_1_1  0.49826    0.01563   31.89  0.46763  0.52889 <0.0000000000000002
#> sigma0_2_2  0.48993    0.02206   22.21  0.44670  0.53316 <0.0000000000000002
#>               
#> beta_2_2   ***
#> psi_2_2    ***
#> theta      ***
#> mu0_1_1    ***
#> sigma0_1_1 ***
#> sigma0_2_2 ***
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#> 
#> -2 log-likelihood value at convergence = 130172.19
#> AIC = 130184.19
#> BIC = 130237.11

OpenMx: State Space Model

OpenMx (Neale et al., 2016) fits the same state space model with its own Kalman filter (OpenMx::mxExpectationStateSpace(); Hunter (2018)). Its matrices have the names used in engineering: A is the transition matrix \boldsymbol{\beta}, B holds the intercepts of the dynamic model, C is the loading matrix \boldsymbol{\Lambda}, D holds the intercepts of the measurement model, Q is the process noise covariance matrix, and R is the measurement error covariance matrix. The parameter labels are the dynr names; as in dynr, a label used in two places is one shared parameter, and the starting values are the population values.

Two details differ from dynr. First, OpenMx reads the data as a single time series. To start the filter again at the first occasion of every individual, each individual gets a group, and all groups use the matrices of one parent model, so every individual has the same parameters. Second, OpenMx starts the filter one occasion before the first row, so the first row has the state mean \mathbf{A} \mathbf{x}_{0} + \mathbf{B} \mathbf{u} and the state covariance \mathbf{A} \mathbf{P}_{0} \mathbf{A}^{\prime} + \mathbf{Q}, with the constant input \mathbf{u} = 1. Setting \mathbf{x}_{0} = \mathbf{A}^{-1} \left( \boldsymbol{\mu}_{0} - \mathbf{B} \mathbf{u} \right) and \mathbf{P}_{0} = \mathbf{A}^{-1} \left( \boldsymbol{\Sigma}_{0} - \mathbf{Q} \right) \left( \mathbf{A}^{-1} \right)^{\prime} gives the first occasion the initial condition of the model, so the likelihood is the same as in dynr. This needs an invertible \mathbf{A}, which holds here.

The two helper functions below are the same in every vignette of the package. MxMat() makes an OpenMx matrix from values and dynr-style labels ("fixed" marks a fixed element), and FitMxSSM() builds the groups and fits the model.

MxMat <- function(values, labels, name) {
  values <- as.matrix(values)
  labels <- matrix(
    data = labels,
    nrow = nrow(values),
    ncol = ncol(values)
  )
  free <- labels != "fixed"
  labels[!free] <- NA
  OpenMx::mxMatrix(
    type = "Full",
    values = values,
    free = free,
    labels = labels,
    name = name
  )
}
FitMxSSM <- function(data, observed, matrices, name) {
  # the groups refer to the matrices of the parent model
  shared <- function(x) {
    paste0(name, ".", x)
  }
  expectation <- OpenMx::mxExpectationStateSpace(
    A = shared("A"),
    B = shared("B"),
    C = shared("C"),
    D = shared("D"),
    Q = shared("Q"),
    R = shared("R"),
    x0 = shared("x0"),
    P0 = shared("P0"),
    u = shared("u")
  )
  # OpenMx needs the names of the observed variables and of the states on C
  for (i in seq_along(matrices)) {
    if (matrices[[i]]$name == "C") {
      dimnames(matrices[[i]]) <- list(
        observed,
        paste0("eta_", seq_len(ncol(matrices[[i]]$values)))
      )
    }
  }
  persons <- split(
    x = data[, observed, drop = FALSE],
    f = data[, "id"]
  )
  groups <- paste0("id_", names(persons))
  submodels <- lapply(
    X = seq_along(persons),
    FUN = function(i) {
      OpenMx::mxModel(
        name = groups[i],
        expectation,
        OpenMx::mxFitFunctionML(),
        OpenMx::mxData(
          observed = persons[[i]],
          type = "raw"
        )
      )
    }
  )
  model <- OpenMx::mxModel(
    name = name,
    matrices,
    OpenMx::mxMatrix(
      type = "Unit",
      nrow = 1,
      ncol = 1,
      name = "u"
    ),
    # the first occasion gets the initial condition of the model
    OpenMx::mxAlgebra(
      expression = solve(A) %*% (mu0 - B %*% u),
      name = "x0"
    ),
    OpenMx::mxAlgebra(
      expression = solve(A) %*% (sigma0 - Q) %*% t(solve(A)),
      name = "P0"
    ),
    submodels,
    OpenMx::mxFitFunctionMultigroup(groups = groups)
  )
  # retries from perturbed starting values only if the first run fails
  OpenMx::mxTryHard(
    model = model,
    extraTries = 5,
    silent = TRUE
  )
}

The matrices of the model, with the dynr labels.

mx_matrices <- list(
  MxMat(
    values = beta,
    labels = c("fixed", "fixed", "fixed", "beta_2_2"),
    name = "A"
  ),
  MxMat(values = alpha, labels = "fixed", name = "B"),
  MxMat(values = lambda, labels = "fixed", name = "C"),
  MxMat(values = nu, labels = "fixed", name = "D"),
  MxMat(
    values = psi,
    labels = c("fixed", "fixed", "fixed", "psi_2_2"),
    name = "Q"
  ),
  MxMat(values = theta, labels = "theta", name = "R"),
  MxMat(values = mu0, labels = c("mu0_1_1", "fixed"), name = "mu0"),
  MxMat(
    values = sigma0,
    labels = c("sigma0_1_1", "fixed", "fixed", "sigma0_2_2"),
    name = "sigma0"
  )
)
fit_mx_ssm <- FitMxSSM(
  data = data,
  observed = "y1",
  matrices = mx_matrices,
  name = "starts"
)
#> 
Beginning initial fit attempt
Fit attempt 0, fit=130172.19038846, new current best! (was 130175.551473008)
                                                                            
#> 
#>  Solution found!  Final fit=130172.19 (started at 130175.55)  (1 attempt(s): 1 valid, 0 errors)
summary(fit_mx_ssm)
#> Summary of starts 
#>  
#> free parameters:
#>         name matrix row col  Estimate  Std.Error A
#> 1   beta_2_2      A   2   2 0.6029308 0.01842698  
#> 2    psi_2_2      Q   2   2 0.3307593 0.01423413  
#> 3      theta      R   1   1 0.2941579 0.01189362  
#> 4    mu0_1_1    mu0   1   1 2.9961528 0.01174301  
#> 5 sigma0_1_1 sigma0   1   1 0.4982005 0.01568720  
#> 6 sigma0_2_2 sigma0   2   2 0.4898306 0.02226156  
#> 
#> Model Statistics: 
#>                |  Parameters  |  Degrees of Freedom  |  Fit (-2lnL units)
#>        Model:              6                  49994              130172.2
#>    Saturated:             NA                     NA                    NA
#> Independence:             NA                     NA                    NA
#> Number of observations/statistics: 50000/50000
#> 
#> Information Criteria: 
#>       |  df Penalty  |  Parameters Penalty  |  Sample-Size Adjusted
#> AIC:       30184.19               130184.2                 130184.2
#> BIC:     -410751.81               130237.1                 130218.0
#> To get additional fit indices, see help(mxRefModels)
#> timestamp: 2026-11-05 07:19:46 
#> Wall clock time: 91.82578 secs 
#> optimizer:  SLSQP 
#> OpenMx version number: 2.22.11 
#> Need help?  See help(mxSummary)

lavaan: Structural Equation Model

The same model in SEM form, fitted with lavaan, should give nearly the same estimates. In SEM, each occasion is a separate observed variable in one row per individual (wide format). The stable trait is a factor with loadings of 1 on all occasions, and the autoregressive trait is a factor on each occasion with a loading of 1 on that occasion.

Wide Data Set

as.data.frame() with long = FALSE returns the data in wide format.

data_wide <- as.data.frame(sim, long = FALSE)
head(data_wide)
#>   id     y1_0      y1_1     y1_2     y1_3     y1_4         y1_5      y1_6
#> 1  1 2.724102 2.5541457 2.834498 1.952812 1.715812  1.798422003 2.6524161
#> 2  2 1.398728 1.5898005 2.078358 3.136913 3.420155  1.656597551 1.7696126
#> 3  3 5.130244 3.5302297 3.728487 3.017891 5.113005  4.727711315 3.2796103
#> 4  4 1.891987 1.9632526 3.094126 1.995233 2.613028  2.185718347 3.3743940
#> 5  5 1.725919 0.5120299 0.753101 2.145500 2.795792  2.154440779 1.9059497
#> 6  6 2.957243 1.4160208 1.596537 1.762764 1.159233 -0.005812339 0.8095458
#>        y1_7      y1_8      y1_9
#> 1 2.0029863 2.4650079 2.7663724
#> 2 0.6304025 1.7566680 0.4364570
#> 3 2.7692017 3.2305090 3.4921984
#> 4 2.7263657 0.8028354 1.1643963
#> 5 1.7443716 1.6288711 0.7896516
#> 6 1.5117643 1.3294805 2.4095639

Model

The model has the same structure for every occasion, so we write it with paste0(). The parameter labels match the dynr names. Giving the autoregressive effect, the process noise variance, and the state variance the same label on every occasion constrains them to be equal over time, as in the state space model.

waves <- 0:(time - 1)
now <- waves[-1]
before <- waves[-time]
model <- paste(
  c(
    "# stable trait: loadings fixed at 1",
    paste0("eta_1 =~ ", paste0("1 * y1_", waves, collapse = " + ")),
    "# autoregressive trait on each occasion",
    paste0("ar_", waves, " =~ 1 * y1_", waves),
    "# autoregressive effect",
    paste0("ar_", now, " ~ beta_2_2 * ar_", before),
    "# autoregressive trait at the first occasion",
    "ar_0 ~~ sigma0_2_2 * ar_0",
    "# process noise at later occasions",
    paste0("ar_", now, " ~~ psi_2_2 * ar_", now),
    "# stable trait: mean and variance",
    "eta_1 ~ mu0_1_1 * 1",
    "eta_1 ~~ sigma0_1_1 * eta_1",
    "# the two traits are uncorrelated",
    "eta_1 ~~ 0 * ar_0",
    "# state variance, equal over time; no observed intercepts",
    paste0("y1_", waves, " ~~ theta * y1_", waves),
    paste0("y1_", waves, " ~ 0 * 1")
  ),
  collapse = "\n"
)
cat(model)
#> # stable trait: loadings fixed at 1
#> eta_1 =~ 1 * y1_0 + 1 * y1_1 + 1 * y1_2 + 1 * y1_3 + 1 * y1_4 + 1 * y1_5 + 1 * y1_6 + 1 * y1_7 + 1 * y1_8 + 1 * y1_9
#> # autoregressive trait on each occasion
#> ar_0 =~ 1 * y1_0
#> ar_1 =~ 1 * y1_1
#> ar_2 =~ 1 * y1_2
#> ar_3 =~ 1 * y1_3
#> ar_4 =~ 1 * y1_4
#> ar_5 =~ 1 * y1_5
#> ar_6 =~ 1 * y1_6
#> ar_7 =~ 1 * y1_7
#> ar_8 =~ 1 * y1_8
#> ar_9 =~ 1 * y1_9
#> # autoregressive effect
#> ar_1 ~ beta_2_2 * ar_0
#> ar_2 ~ beta_2_2 * ar_1
#> ar_3 ~ beta_2_2 * ar_2
#> ar_4 ~ beta_2_2 * ar_3
#> ar_5 ~ beta_2_2 * ar_4
#> ar_6 ~ beta_2_2 * ar_5
#> ar_7 ~ beta_2_2 * ar_6
#> ar_8 ~ beta_2_2 * ar_7
#> ar_9 ~ beta_2_2 * ar_8
#> # autoregressive trait at the first occasion
#> ar_0 ~~ sigma0_2_2 * ar_0
#> # process noise at later occasions
#> ar_1 ~~ psi_2_2 * ar_1
#> ar_2 ~~ psi_2_2 * ar_2
#> ar_3 ~~ psi_2_2 * ar_3
#> ar_4 ~~ psi_2_2 * ar_4
#> ar_5 ~~ psi_2_2 * ar_5
#> ar_6 ~~ psi_2_2 * ar_6
#> ar_7 ~~ psi_2_2 * ar_7
#> ar_8 ~~ psi_2_2 * ar_8
#> ar_9 ~~ psi_2_2 * ar_9
#> # stable trait: mean and variance
#> eta_1 ~ mu0_1_1 * 1
#> eta_1 ~~ sigma0_1_1 * eta_1
#> # the two traits are uncorrelated
#> eta_1 ~~ 0 * ar_0
#> # state variance, equal over time; no observed intercepts
#> y1_0 ~~ theta * y1_0
#> y1_1 ~~ theta * y1_1
#> y1_2 ~~ theta * y1_2
#> y1_3 ~~ theta * y1_3
#> y1_4 ~~ theta * y1_4
#> y1_5 ~~ theta * y1_5
#> y1_6 ~~ theta * y1_6
#> y1_7 ~~ theta * y1_7
#> y1_8 ~~ theta * y1_8
#> y1_9 ~~ theta * y1_9
#> y1_0 ~ 0 * 1
#> y1_1 ~ 0 * 1
#> y1_2 ~ 0 * 1
#> y1_3 ~ 0 * 1
#> y1_4 ~ 0 * 1
#> y1_5 ~ 0 * 1
#> y1_6 ~ 0 * 1
#> y1_7 ~ 0 * 1
#> y1_8 ~ 0 * 1
#> y1_9 ~ 0 * 1

lavaan::lavaan() adds nothing to the model by itself, so every parameter above is exactly the one written.

fit_lavaan <- lavaan::lavaan(
  model = model,
  data = data_wide
)
lavaan::summary(fit_lavaan)
#> lavaan 0.7-3 ended normally after 36 iterations
#> 
#>   Estimator                                         ML
#>   Optimization method                           NLMINB
#>   Number of model parameters                        31
#>   Number of equality constraints                    25
#> 
#>   Number of observations                          5000
#> 
#> Model Test User Model:
#>                                                       
#>   Test statistic                                53.844
#>   Degrees of freedom                                59
#>   P-value (Chi-square)                           0.665
#>                                                       
#>   Browne's residual (NT model-based) test             
#>   Test statistic                                53.517
#>   Degrees of freedom                                59
#>   P-value (Chi-square)                           0.677
#> 
#> Parameter Estimates:
#> 
#>   Standard errors                             Standard
#>   Information                                 Expected
#>   Information saturated (h1) model          Structured
#> 
#> Latent Variables:
#>                    Estimate  Std.Err  z-value  P(>|z|)
#>   eta_1 =~                                            
#>     y1_0              1.000                           
#>     y1_1              1.000                           
#>     y1_2              1.000                           
#>     y1_3              1.000                           
#>     y1_4              1.000                           
#>     y1_5              1.000                           
#>     y1_6              1.000                           
#>     y1_7              1.000                           
#>     y1_8              1.000                           
#>     y1_9              1.000                           
#>   ar_0 =~                                             
#>     y1_0              1.000                           
#>   ar_1 =~                                             
#>     y1_1              1.000                           
#>   ar_2 =~                                             
#>     y1_2              1.000                           
#>   ar_3 =~                                             
#>     y1_3              1.000                           
#>   ar_4 =~                                             
#>     y1_4              1.000                           
#>   ar_5 =~                                             
#>     y1_5              1.000                           
#>   ar_6 =~                                             
#>     y1_6              1.000                           
#>   ar_7 =~                                             
#>     y1_7              1.000                           
#>   ar_8 =~                                             
#>     y1_8              1.000                           
#>   ar_9 =~                                             
#>     y1_9              1.000                           
#> 
#> Regressions:
#>                    Estimate  Std.Err  z-value  P(>|z|)
#>   ar_1 ~                                              
#>     ar_0    (b_2_)    0.603    0.018   34.402    0.000
#>   ar_2 ~                                              
#>     ar_1    (b_2_)    0.603    0.018   34.402    0.000
#>   ar_3 ~                                              
#>     ar_2    (b_2_)    0.603    0.018   34.402    0.000
#>   ar_4 ~                                              
#>     ar_3    (b_2_)    0.603    0.018   34.402    0.000
#>   ar_5 ~                                              
#>     ar_4    (b_2_)    0.603    0.018   34.402    0.000
#>   ar_6 ~                                              
#>     ar_5    (b_2_)    0.603    0.018   34.402    0.000
#>   ar_7 ~                                              
#>     ar_6    (b_2_)    0.603    0.018   34.402    0.000
#>   ar_8 ~                                              
#>     ar_7    (b_2_)    0.603    0.018   34.402    0.000
#>   ar_9 ~                                              
#>     ar_8    (b_2_)    0.603    0.018   34.402    0.000
#> 
#> Covariances:
#>                    Estimate  Std.Err  z-value  P(>|z|)
#>   eta_1 ~~                                            
#>     ar_0              0.000                           
#> 
#> Intercepts:
#>                    Estimate  Std.Err  z-value  P(>|z|)
#>     eta_1   (m0_1)    2.996    0.012  255.138    0.000
#>    .y1_0              0.000                           
#>    .y1_1              0.000                           
#>    .y1_2              0.000                           
#>    .y1_3              0.000                           
#>    .y1_4              0.000                           
#>    .y1_5              0.000                           
#>    .y1_6              0.000                           
#>    .y1_7              0.000                           
#>    .y1_8              0.000                           
#>    .y1_9              0.000                           
#> 
#> Variances:
#>                    Estimate  Std.Err  z-value  P(>|z|)
#>     ar_0    (s0_2)    0.490    0.022   22.617    0.000
#>    .ar_1    (p_2_)    0.331    0.014   23.895    0.000
#>    .ar_2    (p_2_)    0.331    0.014   23.895    0.000
#>    .ar_3    (p_2_)    0.331    0.014   23.895    0.000
#>    .ar_4    (p_2_)    0.331    0.014   23.895    0.000
#>    .ar_5    (p_2_)    0.331    0.014   23.895    0.000
#>    .ar_6    (p_2_)    0.331    0.014   23.895    0.000
#>    .ar_7    (p_2_)    0.331    0.014   23.895    0.000
#>    .ar_8    (p_2_)    0.331    0.014   23.895    0.000
#>    .ar_9    (p_2_)    0.331    0.014   23.895    0.000
#>     eta_1   (s0_1)    0.498    0.016   32.127    0.000
#>    .y1_0    (thet)    0.294    0.011   25.598    0.000
#>    .y1_1    (thet)    0.294    0.011   25.598    0.000
#>    .y1_2    (thet)    0.294    0.011   25.598    0.000
#>    .y1_3    (thet)    0.294    0.011   25.598    0.000
#>    .y1_4    (thet)    0.294    0.011   25.598    0.000
#>    .y1_5    (thet)    0.294    0.011   25.598    0.000
#>    .y1_6    (thet)    0.294    0.011   25.598    0.000
#>    .y1_7    (thet)    0.294    0.011   25.598    0.000
#>    .y1_8    (thet)    0.294    0.011   25.598    0.000
#>    .y1_9    (thet)    0.294    0.011   25.598    0.000

OpenMx: Structural Equation Model

The same SEM in OpenMx, written with OpenMx::mxPath() on the wide data. One-headed arrows are loadings and regressions, two-headed arrows are variances and covariances, and arrows from "one" are means. A path that is not written is fixed at zero, so the zero residual variances, covariances, and intercepts of the lavaan model need no lines here. The labels are the dynr names, and the starting values are the population values.

waves <- 0:(time - 1)
manifest <- paste0("y1_", waves)
ar <- paste0("ar_", waves)
mx_sem <- OpenMx::mxModel(
  name = "starts_sem",
  type = "RAM",
  manifestVars = manifest,
  latentVars = c("eta_1", ar),
  OpenMx::mxData(observed = data_wide[, manifest], type = "raw"),
  # stable trait: loadings fixed at 1
  OpenMx::mxPath(from = "eta_1", to = manifest, free = FALSE, values = 1),
  # autoregressive trait on each occasion
  OpenMx::mxPath(from = ar, to = manifest, free = FALSE, values = 1),
  # autoregressive effect
  OpenMx::mxPath(
    from = ar[-time],
    to = ar[-1],
    labels = "beta_2_2",
    values = beta[2, 2]
  ),
  # autoregressive trait at the first occasion, process noise later
  OpenMx::mxPath(
    from = "ar_0",
    arrows = 2,
    labels = "sigma0_2_2",
    values = sigma0[2, 2]
  ),
  OpenMx::mxPath(
    from = ar[-1],
    arrows = 2,
    labels = "psi_2_2",
    values = psi[2, 2]
  ),
  # stable trait: mean and variance
  OpenMx::mxPath(
    from = "one",
    to = "eta_1",
    labels = "mu0_1_1",
    values = mu0[1]
  ),
  OpenMx::mxPath(
    from = "eta_1",
    arrows = 2,
    labels = "sigma0_1_1",
    values = sigma0[1, 1]
  ),
  # state variance, equal over time
  OpenMx::mxPath(
    from = manifest,
    arrows = 2,
    labels = "theta",
    values = theta[1, 1]
  )
)
fit_mx_sem <- OpenMx::mxRun(
  model = mx_sem,
  silent = TRUE
)
summary(fit_mx_sem)
#> Summary of starts_sem 
#>  
#> free parameters:
#>         name matrix   row   col  Estimate  Std.Error A
#> 1   beta_2_2      A  ar_1  ar_0 0.6029308 0.01845227  
#> 2      theta      S  y1_0  y1_0 0.2941579 0.01190797  
#> 3 sigma0_1_1      S eta_1 eta_1 0.4982005 0.01569699  
#> 4 sigma0_2_2      S  ar_0  ar_0 0.4898306 0.02226597  
#> 5    psi_2_2      S  ar_1  ar_1 0.3307593 0.01425239  
#> 6    mu0_1_1      M     1 eta_1 2.9961528 0.01174332  
#> 
#> Model Statistics: 
#>                |  Parameters  |  Degrees of Freedom  |  Fit (-2lnL units)
#>        Model:              6                  49994              130172.2
#>    Saturated:             65                  49935                    NA
#> Independence:             20                  49980                    NA
#> Number of observations/statistics: 5000/50000
#> 
#> Information Criteria: 
#>       |  df Penalty  |  Parameters Penalty  |  Sample-Size Adjusted
#> AIC:       30184.19               130184.2                 130184.2
#> BIC:     -295636.37               130223.3                 130204.2
#> To get additional fit indices, see help(mxRefModels)
#> timestamp: 2026-11-05 07:23:00 
#> Wall clock time: 0.03482652 secs 
#> optimizer:  SLSQP 
#> OpenMx version number: 2.22.11 
#> Need help?  See help(mxSummary)

Population Values and Estimates

Each table below puts the population values used to simulate the data next to the estimates of one method. Bias is the estimate minus the population value. Because each model is fitted to a single simulated data set, this is the error of one estimate, not a Monte Carlo bias, which would average over many replications. The 95% confidence interval is the Wald interval, \hat{\theta} \pm 1.96 \times \mathrm{SE}, unless the caption says otherwise, and Covers shows whether it contains the population value. The parameter names are the dynr names in every table.

EstimatesTable <- function(population, estimate, se, ll = NULL, ul = NULL) {
  z <- stats::qnorm(p = 0.975)
  if (is.null(ll)) {
    ll <- estimate - z * se
  }
  if (is.null(ul)) {
    ul <- estimate + z * se
  }
  data.frame(
    Parameter = names(population),
    Population = unname(population),
    Estimate = unname(estimate),
    SE = unname(se),
    Bias = unname(estimate - population),
    `2.5%` = unname(ll),
    `97.5%` = unname(ul),
    Covers = unname(population >= ll & population <= ul),
    check.names = FALSE
  )
}
population <- c(
  mu0_1_1 = mu0[1],
  sigma0_1_1 = sigma0[1, 1],
  sigma0_2_2 = sigma0[2, 2],
  beta_2_2 = beta[2, 2],
  psi_2_2 = psi[2, 2],
  theta = theta[1, 1]
)

dynr

dynr_est <- coef(results)
dynr_se <- sqrt(diag(vcov(results)))
knitr::kable(
  x = EstimatesTable(
    population = population,
    estimate = dynr_est[names(population)],
    se = dynr_se[names(population)]
  ),
  digits = 3,
  caption = "Population values and dynr estimates (one simulated data set)"
)
Population values and dynr estimates (one simulated data set)
Parameter Population Estimate SE Bias 2.5% 97.5% Covers
mu0_1_1 3.00 2.996 0.012 -0.004 2.973 3.019 TRUE
sigma0_1_1 0.50 0.498 0.016 -0.002 0.468 0.529 TRUE
sigma0_2_2 0.50 0.490 0.022 -0.010 0.447 0.533 TRUE
beta_2_2 0.60 0.603 0.018 0.003 0.567 0.638 TRUE
psi_2_2 0.32 0.331 0.014 0.011 0.304 0.358 TRUE
theta 0.30 0.294 0.012 -0.006 0.271 0.317 TRUE

OpenMx: State Space Model

mx_ssm_est <- OpenMx::omxGetParameters(fit_mx_ssm)
mx_ssm_se <- fit_mx_ssm$output$standardErrors[, 1]
knitr::kable(
  x = EstimatesTable(
    population = population,
    estimate = mx_ssm_est[names(population)],
    se = mx_ssm_se[names(population)]
  ),
  digits = 3,
  caption = paste(
    "Population values and OpenMx state space estimates",
    "(one simulated data set)"
  )
)
Population values and OpenMx state space estimates (one simulated data set)
Parameter Population Estimate SE Bias 2.5% 97.5% Covers
mu0_1_1 3.00 2.996 0.012 -0.004 2.973 3.019 TRUE
sigma0_1_1 0.50 0.498 0.016 -0.002 0.467 0.529 TRUE
sigma0_2_2 0.50 0.490 0.022 -0.010 0.446 0.533 TRUE
beta_2_2 0.60 0.603 0.018 0.003 0.567 0.639 TRUE
psi_2_2 0.32 0.331 0.014 0.011 0.303 0.359 TRUE
theta 0.30 0.294 0.012 -0.006 0.271 0.317 TRUE

lavaan: Structural Equation Model

pe <- lavaan::parameterEstimates(fit_lavaan)
# a label shared by several paths is one parameter
pe <- pe[pe$label != "" & !duplicated(pe$label), ]
lavaan_est <- stats::setNames(object = pe$est, nm = pe$label)
lavaan_se <- stats::setNames(object = pe$se, nm = pe$label)
knitr::kable(
  x = EstimatesTable(
    population = population,
    estimate = lavaan_est[names(population)],
    se = lavaan_se[names(population)]
  ),
  digits = 3,
  caption = "Population values and lavaan estimates (one simulated data set)"
)
Population values and lavaan estimates (one simulated data set)
Parameter Population Estimate SE Bias 2.5% 97.5% Covers
mu0_1_1 3.00 2.996 0.012 -0.004 2.973 3.019 TRUE
sigma0_1_1 0.50 0.498 0.016 -0.002 0.468 0.529 TRUE
sigma0_2_2 0.50 0.490 0.022 -0.010 0.447 0.532 TRUE
beta_2_2 0.60 0.603 0.018 0.003 0.569 0.637 TRUE
psi_2_2 0.32 0.331 0.014 0.011 0.304 0.358 TRUE
theta 0.30 0.294 0.011 -0.006 0.272 0.317 TRUE

OpenMx: Structural Equation Model

mx_sem_est <- OpenMx::omxGetParameters(fit_mx_sem)
mx_sem_se <- fit_mx_sem$output$standardErrors[, 1]
knitr::kable(
  x = EstimatesTable(
    population = population,
    estimate = mx_sem_est[names(population)],
    se = mx_sem_se[names(population)]
  ),
  digits = 3,
  caption = "Population values and OpenMx SEM estimates (one simulated data set)"
)
Population values and OpenMx SEM estimates (one simulated data set)
Parameter Population Estimate SE Bias 2.5% 97.5% Covers
mu0_1_1 3.00 2.996 0.012 -0.004 2.973 3.019 TRUE
sigma0_1_1 0.50 0.498 0.016 -0.002 0.467 0.529 TRUE
sigma0_2_2 0.50 0.490 0.022 -0.010 0.446 0.533 TRUE
beta_2_2 0.60 0.603 0.018 0.003 0.567 0.639 TRUE
psi_2_2 0.32 0.331 0.014 0.011 0.303 0.359 TRUE
theta 0.30 0.294 0.012 -0.006 0.271 0.317 TRUE

All Methods

The estimates of every method side by side. The methods maximize the same likelihood, so their estimates should agree closely.

knitr::kable(
  x = data.frame(
    Parameter = names(population),
    Population = unname(population),
    `dynr` = unname(dynr_est[names(population)]),
    `OpenMx (SSM)` = unname(mx_ssm_est[names(population)]),
    `lavaan` = unname(lavaan_est[names(population)]),
    `OpenMx (SEM)` = unname(mx_sem_est[names(population)]),
    check.names = FALSE
  ),
  digits = 3,
  caption = "Population values and the estimates of every method"
)
Population values and the estimates of every method
Parameter Population dynr OpenMx (SSM) lavaan OpenMx (SEM)
mu0_1_1 3.00 2.996 2.996 2.996 2.996
sigma0_1_1 0.50 0.498 0.498 0.498 0.498
sigma0_2_2 0.50 0.490 0.490 0.490 0.490
beta_2_2 0.60 0.603 0.603 0.603 0.603
psi_2_2 0.32 0.331 0.331 0.331 0.331
theta 0.30 0.294 0.294 0.294 0.294

References

Chow, S.-M., Ho, M. R., Hamaker, E. L., & Dolan, C. V. (2010). Equivalence and differences between structural equation modeling and state-space modeling techniques. Structural Equation Modeling: A Multidisciplinary Journal, 17(2), 303–332. https://doi.org/10.1080/10705511003661553
Hunter, M. D. (2018). State space modeling in an open source, modular, structural equation modeling environment. Structural Equation Modeling: A Multidisciplinary Journal, 25(2), 307–324. https://doi.org/10.1080/10705511.2017.1369354
Kenny, D. A., & Zautra, A. (1995). The trait-state-error model for multiwave data. Journal of Consulting and Clinical Psychology, 63(1), 52–59. https://doi.org/10.1037/0022-006x.63.1.52
Kenny, D. A., & Zautra, A. (2001). Trait-state models for longitudinal data. In L. M. Collins & A. G. Sayer (Eds.), New methods for the analysis of change (pp. 243–263). American Psychological Association. https://doi.org/10.1037/10409-008
Neale, M. C., Hunter, M. D., Pritikin, J. N., Zahery, M., Brick, T. R., Kirkpatrick, R. M., Estabrook, R., Bates, T. C., Maes, H. H., & Boker, S. M. (2016). OpenMx 2.0: Extended structural equation and statistical modeling. Psychometrika, 81(2), 535–549. https://doi.org/10.1007/s11336-014-9435-8
Ou, L., Hunter, M. D., & Chow, S.-M. (2019). What’s for dynr: A package for linear and nonlinear dynamic modeling in R. The R Journal, 11(1), 91. https://doi.org/10.32614/rj-2019-012
R Core Team. (2026). R: A language and environment for statistical computing. R Foundation for Statistical Computing. https://www.R-project.org/
Rosseel, Y. (2012). lavaan: An R package for structural equation modeling. Journal of Statistical Software, 48(2). https://doi.org/10.18637/jss.v048.i02