2/2/23, 1:41 PM dynamite • dynamite

dynamite
Source: vignettes/dynamite.Rmd

Introduction
Panel data is common in various fields such as social sciences. These data consists of multiple
subjects, followed over several time points, and there are often many observations per individual at
each time, for example family status and income of each individual at each time point of interest.
Such data can be analyzed in various ways, depending on the research questions and the
characteristics of the data such as the number of individuals and time points, and the assumed
distribution of the response variables. Popular, somewhat overlapping modelling approaches
include dynamic panel models, fixed effect models, cross-lagged panel models (CLPM), CLPM with
fixed or random effects Mulder and Hamaker (2021). Parameter estimation is typically based on
ordinary least squares, generalized method of moments or maximum likelihood using structural
equation models.

There are several R (R Core Team 2021) packages in CRAN focusing on panel data analysis. plm
(Croissant and Millo 2008) provides various estimation methods and tests for linear panel data
model, while fixest (Bergé 2018) supports multiple fixed effects and different distributions of
response variables (based on stats::family ). panelr (Long 2020) contains tools for panel data
manipulation and estimation methods for “within-between” models which combine fixed effect and
random effect models. This is done by using lme4 (Bates et al. 2015), geepack (Halekoh, Højsgaard,
and Yan 2006), and brms (Bürkner 2018) packages as a backend. Finally, the lavaan (Rosseel 2012)
package provides methods for general structural equation modelling, and thus can be used to
estimate various panel data models such as cross-lagged panel models with fixed or random
intercepts.

In traditional panel data models such as the ones provided by the aforementioned packages, the
number of time points considered is often assumed to be relatively small, say less than 10, while the
number of individuals can be for example hundreds or thousands. Due to the small number of time
points, the effects of predictors are typically assumed to be time-invariant (although some
extensions to time-varying effects have emerged (Hayakawa and Hou 2019)). On the other hand,
when the number of time points is moderate or large, say hundreds or thousands, it can be
reasonable to assume that the dynamics of the system change over time. There are various
methods to model time-varying effects in (generalized) linear models based on state space models
(Harvey and Phillips 1982; Durbin and Koopman 2012), with various R implementations (J. Helske
2022; Knaus et al. 2021; Brodersen et al. 2014). The state space modelling approaches are highly
flexible but often computationally demanding due to assumed latent processes for the regression
coefficients and an analytically intractable marginal likelihood, so they are most useful in settings
dynamite
with only few subjects. Methods
1.1.0 based on the varying coefficients models (Hastie and Tibshirani

https://docs.ropensci.org/dynamite/articles/dynamite.html 1/16
2/2/23, 1:41 PM dynamite • dynamite

1993; Eubank et al. 2004) can be computationally more convenient, especially when the varying
coefficients are based on splines. Some of the R implementations of time-varying coefficient models
include (Casas and Fernandez-Casal 2019; Dziak et al. 2021).

The dynamite package provides an alternative approach to panel data inference which avoids
some of the limitations and drawbacks of the aforementioned methods. First, dynamite supports
estimating effects which vary smoothly over time according to splines, with penalization based on
random walk priors. This allows modelling for example effects of interventions which increase or
decrease over time. Second, dynamite supports wide variety of distributions for the response
variables such as categorical distribution commonly encountered in life-course research (S. Helske,
Helske, and Eerola 2018). Third, estimation is done in a fully Bayesian fashion using Stan (Stan
Development Team 2022b, 2022a), which leads to transparent and interpretable quantification of
parameter and predictive uncertainty. Finally, with dynamite we can model arbitrary number of
measurements per individual (i.e. multiple channels following (S. Helske and Helske 2019)). Jointly
modelling all endogenous variables allows us to consider long-term effects of interventions which
take into account the interdependency of several variables of the model, and allows realistic
simulation of complex counterfactual scenarios.

The Model
Consider an individual i with observations y i,t = (y
1
i,t
,…,y
c
i,t
,…,y
C
i,t
,
) t = 1, … , T ,
i = 1, … , N i.e. at each time point t we have C observations from N individuals. Assume that
each element of y i,t can depend on the past values y i,t−1 , as well as additional exogenous
covariates x i,t . Then, assuming that the elements of y i,t are independent given y i,t−1 and x i,t , we
have

C
c c 1 C
y i,t ∼ p t (y i,t |y i,t−1 , x i,t ) = ∏ p (y |y ,…,y , x i,t ).
t i,t i,t−1 i,t−1

c=1

Here the parameters of the conditional distributions p c depend on time, which allows us to take into
account the fact that the dynamics of our system can evolve over time.

For p ct , given a suitable link function depending on our distributional assumptions, we define a linear
predictor η ci,t for the channel c with a following general form:

c c c c c c
η = νi + α + Z β + X δ
i,t t i,t i,t t

νi is the individual-specific random intercept, α t (possibly time-varying) is the common intercept

term, Z i,t
c
defines the predictors corresponding to the vector of time-invariant coefficients β c , and
similarly for the time-varying part X i,t
c
δ . The random intercept terms ν i are assumed to follow zero-
c
t

mean Gaussian and these can be correlated across channels. Note that dynamite does not center
the response nor predictor variables.

dynamite 1.1.0
https://docs.ropensci.org/dynamite/articles/dynamite.html 2/16
2/2/23, 1:41 PM dynamite • dynamite

For time-varying coefficients δ (and similarly for α), we use Bayesian P-splines (penalized B-splines)
(Lang and Brezger 2004) where

c c
δ = Bt ω , k = 1, … , K,
k,t k

where K is the number of covariates, B t is a vector of B-spline values at time t and ω ck is vector of
corresponding spline coefficients. In general the number of B-splines D used for constructing the
splines for the study period 1, … , T can be chosen freely, but too large D can result in overfitting.
To mitigate this, we define a random walk prior for ω ck as

c c c c c 2
ω ∼ p(ω ), ω ∼ N (ω , (τ ) ), d = 2, … , D.
k,1 k,1 k,d k,d−1 k

with user defined prior p(ω ck,1 ) on the first coefficient, which due to the structure of B 1 corresponds
to the prior on the δ ck,1 . Here, the parameter τ kc controls the smoothness of the spline curves.

Defining The Model

The models in the dynamite package are defined by combining the channel-specific formulas
defined via the function dynamiteformula for which a shorthand alias obs is available. The function
obs takes two arguments: formula and family , which define how the response variable of the
channel depends on the predictor variables in the standard R formula syntax and the family of the
response variable, respectively. These channel specific definitions are then combined into a single
model with + . For example, the following formula

obs(y ~ lag(x), family = "gaussian") +

obs(x ~ z, family = "poisson")

defines a model with two channels; first we declare that y is a gaussian variable depending on the
previous value of x ( lag(x) ) and then we add a second channel declaring x as Poisson
distributed depending on some exogenous variable z (for which we do not define any distribution).
Note that the model formula can be defined without any reference to some external data, just as an
R formula can. Currently the dynamite package supports the following distributions for the
observations:

Categorical: categorical (with a softmax link using the first category as reference). See the
documentation of the categorical_logit_glm in the Stan function reference manual
(https://mc-stan.org/users/documentation/).
Gaussian: gaussian (identity link, parameterized using mean and standard deviation).
Poisson: poisson (log-link, with an optional known offset variable).
Negative-binomial: negbin (log-link, using mean and dispersion parameterization, with an
optional known offset variable). See the documentation on NegBinomial2 in the Stan function
reference manual.
Bernoulli: bernoulli (logit-link).
Binomial: binomial (logit-link).
dynamite
Exponential: exponential
1.1.0 (log-link).

https://docs.ropensci.org/dynamite/articles/dynamite.html 3/16
2/2/23, 1:41 PM dynamite • dynamite

Gamma: gamma (log-link, using mean and shape parameterization).

Beta: beta (logit-link, using mean and precision parameterization).

Lagged responses and predictors

The dynamite models do not support contemporaneous dependencies in order to avoid complex
cyclic dependencies which would make handling missing data, subsequent predictions, and causal
inference challenging or impossible. In other words, response variables can only depend on the
previous values of other response variables. This is why we used lag(x) in the previous example, a
shorthand for lag(x, k = 1) , which defines a one-step lag of the variable x . Longer lags can also
be defined by adjusting the argument k .

A special model component lags can also be used to quickly add lagged responses as predictors.
This component adds a lagged value of each response in the model as a predictor to every channel.
For example, calling

obs(y ~ z, family = "gaussian") +

obs(x ~ z, family = "poisson") +
lags(k = 1)

would add lag(y, k = 1) as a predictor of x and conversely, lag(x, k = 1) as a predictor of y .

Therefore, the previous code would produce the exact same model as writing

obs(y ~ z + lag(x, k = 1), family = "gaussian") +

obs(x ~ z + lag(y, k = 1), family = "poisson")

Just as with the function lag() , the argument k in lags can be adjusted to add longer lags of
each response to each channel, but for lags it can also be a vector. The inclusion of lagged
response variables in the model implies that some time points have to be considered fixed in the
estimation. The number of fixed time points in the model is equal to the largest shift value of any
observed response variable in the model (defined either via lag or the global lags ).

Time-varying and time-invariant predictors

The formula within obs can also contain an additional special function varying , which defines the
time-varying part of the model equation, in which case we could write for example obs(x ~ z +
varying(~ -1 + w), family = "poisson") , which defines a model equation with a constant
intercept and time-invariant effect of z , and a time-varying effect of w . We also remove the
duplicate intercept with -1 in order to avoid identifiability issues in the model estimation (we could
also define a time varying intercept, in which case we would write obs(x ~ -1 + z + varying(~ w),
family = "poisson) ). The part of the formula not wrapped with varying is assumed to correspond
to the fixed part of the model, so obs(x ~ z + varying(~ -1 + w), family = "poisson") is actually
identical to obs(x ~ -1 + fixed(~ z) + varying(~ -1 + w), family = "poisson") and obs(x ~
dynamite + w), family = "poisson") . The use of fixed is therefore optional.
fixed(~ z) + varying(~ -1 1.1.0

https://docs.ropensci.org/dynamite/articles/dynamite.html 4/16
2/2/23, 1:41 PM dynamite • dynamite

When defining varying effects, we also need to define how the these time-varying regression
coefficient behave. For this, a splines component should be added to the model, e.g.  obs(x ~
varying(~ -1 + w), family = "poisson) + splines(df = 10) defines a cubic B-spline with 10
degrees of freedom for the time-varying coefficient corresponding to the w . If the model contains
multiple time-varying coefficients, same spline basis is used for all coefficients, with unique spline
coefficients and their standard deviation, as defined in the previous Section.
Random intercepts
Defining the random intercept term for each group ca be done using component random where the
first argument defines for which channels the intercept should be added, and second argument
defines whether or not these intercepts should be correlated between channels. This leads to a
model where the in addition to the common intercept each individual/group has their own intercept
with zero-mean normal prior and unknown standard deviation (or multivariate gaussian in case
argument correlated = TRUE ), analogously with the typical mixed models. Note however, that
currently random intercepts are not supported for categorical distribution due to technical reasons.

Auxiliary channels
In addition to declaring response variables via obs , we can also use the function aux to define
auxiliary channels which are deterministic functions of other variables. Defining these auxiliary
variables explicitly instead of defining them implicitly on the right-hand side of the formulas (i.e., by
using the “as is” function I() ) makes the subsequent prediction steps more clear and allows easier
checks on the model validity. In fact, we do not allow the use of I() within dynamiteformula . The
values of auxiliary variables are computed dynamically during prediction, making the use of lagged
values and other transformations possible. An example of a formula using an auxiliary channel could
be

obs(y ~ lag(log1x), family = "gaussian") +

obs(x ~ z, family = "poisson") +
aux(numeric(log1x) ~ log(1 + x) | init(0))

For auxiliary channels, the formula declaration ~ should be understood as a mathematical equality,
where the right hand side provides the defining expression. Thus the example above defines a new
auxiliary channel whose response is log1x defined as the logarithm of 1 + x and assigns it to be
of type numeric . The type declaration is required, because it might not be possible to
unambiguously determine the type of the response variable based on its expression alone from the
data, especially for factor responses.

Auxiliary variables can be used directly in the formulas of other channels, just like any other variable.
Note that an auxiliary channel can also depend on other response variables without lags, unlike
observed channels. The function aux also does not use the family argument, which is
automatically set to deterministic and is a special channel type of obs . Note that lagged values of
deterministic aux channels do not imply fixed time points. Instead they must be given starting

dynamite
values using one of the two 1.1.0
the | symbol.
special functions, init or past after the main formula separated by

https://docs.ropensci.org/dynamite/articles/dynamite.html 5/16
2/2/23, 1:41 PM dynamite • dynamite

In the example above, because the channel for y contains a lagged value of log1x as a predictor,
we also need to supply log1x with a single initial value that determines the value of the lag at the
first time point. Here, init(0) defines the initial value of lag(log1x) to be zero. In general, if the
model contains higher order lags of an auxiliary channel, then init can be supplied with a vector
initializing each lag.

Note that init defines the same starting value to be used for all individuals. As an alternative, the
function past can be used, which takes an R expression as its argument, and computes the starting
value for each individual based on that expression. The expression is evaluated in the context of the
data supplied to dynamite . For example, instead of init(0) in the example above, we could write:

obs(y ~ lag(log1x), family = "gaussian") +

obs(x ~ z, family = "poisson") +
aux(numeric(log1x) ~ log(1 + x) | past(log(z)))

which defines that the value of lag(log1x) at the first time point is log(z) for each individual,
using the value of z in the data to be supplied to compute the actual value of the expression. The
function past can also be used if the model contains higher order lags of auxiliary responses. In
this case, more observations from the variables bound by the expression given as the argument will
simply be used.
Fitting the Model
The declared model formula is supplied to dynamite , which has the following arguments and
default values:

dynamite(dformula, data, time, group = NULL,priors = NULL, verbose = TRUE,

debug = NULL, ...)

The first argument dformula is the dynamiteformula object that defines the model and the second
argument data is a data.frame that contains the variables used in the model formula. The
argument group is a column name of data that specified the unique groups (individuals) and
similarly, time is a column name of data that specifies the unique time points. Note that group is
optional: when group = NULL we assume that there is only a single group (or individual). The
argument priors supplies optional user-defined priors for the model parameters, and ... supplies
additional arguments to rstan::sampling which handles the model estimation. Finally, verbose
controls the verbosity of the output and debug can be used for various debugging options (see ?
dynamite for further information on these parameters). For example,

dynamite(
dformula = obs(x ~ varying(~ -1 + w), family = "poisson") + splines(df = 10),
data = d,
time = "year",

dynamite 1.1.0
group = "id"
chains = 2,

https://docs.ropensci.org/dynamite/articles/dynamite.html 6/16
2/2/23, 1:41 PM dynamite • dynamite

cores = 2
)

estimates the model using the data in the data frame d , which contains variables year and id
(defining the time-index and group-index variables of the data). Arguments chains and cores are
passed to rstan::sampling which then uses to parallel Markov chains in the the Markov chain
Monte Carlo (MCMC) sampling of the model parameters.
Example of Multichannel Model
As an example, consider the following simulated multichannel data:

library(dynamite)
head(multichannel_example)
#> id time g p b
#> 1 1 1 -0.6264538 5 1
#> 2 1 2 -0.2660091 12 0
#> 3 1 3 0.4634939 9 1
#> 4 1 4 1.0451444 15 1
#> 5 1 5 1.7131026 10 1
#> 6 1 6 2.1382398 8 1

The data contains 50 unique groups (variable id ), over 20 time points ( time ), a continuous variable
g , a variable with non-negative integer values p , and a binary variable b . We define the following
model (which actually matches the data generating process used in simulating the data):

set.seed(1)
f <- obs(g ~ lag(g) + lag(logp), family = "gaussian") +
obs(p ~ lag(g) + lag(logp) + lag(b), family = "poisson") +
obs(b ~ lag(b) * lag(logp) + lag(b) * lag(g), family = "bernoulli") +
aux(numeric(logp) ~ log(p + 1))

As a directed acyclic graph (DAG), the model for the first three time points is

dynamite
(The deterministic channel logp
1.1.0 is omitted for clarity).

https://docs.ropensci.org/dynamite/articles/dynamite.html 7/16
2/2/23, 1:41 PM dynamite • dynamite

We fit the model using the dynamite function. In order to satisfy the package size requirements of
CRAN, we have to use small number of iterations and additional thinning:

multichannel_example_fit <- dynamite(

f, data = multichannel_example,
time = "time", group = "id"
chains = 1, cores = 1, iter = 2000, warmup = 1000, init = 0, refresh = 0,
thin = 5, save_warmup = FALSE)

You can update the estimated model which is included in the package as

multichannel_example_fit <- update(multichannel_example_fit,

iter = 2000,
warmup = 1000,
chains = 4,
cores = 4,
refresh = 0,
save_warmup = FALSE
)

Note that the dynamite call above produces a warning message related to the deterministic
channel logp :

#> Warning: Deterministic channel `logp` has a maximum lag of 1 but you've supplied no
#> initial values:
#> ℹ This may result in NA values for `logp`.

This happens because the formulas of the other channels have lag(logp) as a predictor, whose
value is undefined at the first time point without the initial value. However, in this case we can safely
ignore the warning because the model contains lags of b and g as well meaning that the first time
point in the model is fixed and does not enter into the model fitting process.

We can obtain posterior samples or summary statistics of the model using the as.data.frame ,
coef , and summary functions, but here we opt for visualizing the results using the plot_betas
function:

plot_betas(multichannel_example_fit)

dynamite 1.1.0
https://docs.ropensci.org/dynamite/articles/dynamite.html 8/16
2/2/23, 1:41 PM dynamite • dynamite

Note the naming of the model parameters; for example beta_b_g_lag1 corresponds to a fixed
coefficient beta of lagged variable g of the channel b .

Assume now that we are interested in studying the causal effect of variable b 5 on variable g t at
t = 6, … , 20 . There is no direct effect from b 5 to g 6 , but as g t at affects b t+1 (and p t+1 ), which in
turn has an effect on all the variables at t + 2, we should see an indirect effect of b 5 to g t from time
t = 7 onward.

For this task, we first create a new dataset where the values of our response variables after time 5
are set to NA:

d <- multichannel_example
d <- d |>
dplyr::mutate(
g = ifelse(time > 5, NA, g),
p = ifelse(time > 5, NA, p),
b = ifelse(time > 5, NA, b))

We then predict time points 6 to 20 when b at time 5 is 0 or 1 for everyone:

newdata0 <- d |>

dplyr::mutate(
b = ifelse(time == 5, 0, b))
pred0 <- predict(multichannel_example_fit, newdata = newdata0, type = "mean")

dynamite 1.1.0
newdata1 <- d |>
dplyr::mutate(

https://docs.ropensci.org/dynamite/articles/dynamite.html 9/16
2/2/23, 1:41 PM dynamite • dynamite

b = ifelse(time == 5, 1, b))
pred1 <- predict(multichannel_example_fit, newdata = newdata1, type = "mean")

By default, the output from predict is a single data frame containing the original new data and the
samples from the posterior predictive distribution new observations. By defining type = "mean" we
specify that we are interested in the posterior distribution of the expected values instead. In this
case the the predicted values in a data frame are in the columns g_mean , p_mean , and b_mean
where the NA values of newdata are replaced with the posterior samples from the model (the
output also contains additional column corresponding to the auxiliary channel logp and index
variable .draw ). For more memory efficient format, by using argument expand = FALSE the output
is returned as a list with two components, simulated and observed , with new samples and
original newdata respectively.

head(pred0, n = 10) # first rows are NA as those were treated as fixed

#> g_mean p_mean b_mean logp id time .draw g p b
#> 1 NA NA NA 1.7917595 1 1 1 -0.6264538 5 1
#> 2 NA NA NA 2.5649494 1 2 1 -0.2660091 12 0
#> 3 NA NA NA 2.3025851 1 3 1 0.4634939 9 1
#> 4 NA NA NA 2.7725887 1 4 1 1.0451444 15 1
#> 5 NA NA NA 2.3978953 1 5 1 1.7131026 10 0
#> 6 1.836878 3.658028 0.7359881 1.0986123 1 6 1 NA NA NA
#> 7 1.583641 4.057826 0.7063930 0.6931472 1 7 1 NA NA NA
#> 8 1.151348 3.476842 0.6524427 1.6094379 1 8 1 NA NA NA
#> 9 1.247886 2.469857 0.6889052 1.3862944 1 9 1 NA NA NA
#> 10 1.612244 5.014211 0.6963598 1.9459101 1 10 1 NA NA NA

We can now compute summary statistics over the individuals and then over the posterior samples to
obtain posterior distribution of the causal effects as

sumr <- dplyr::bind_rows(

list(b0 = pred0, b1 = pred1), .id = "case") |>
dplyr::group_by(case, .draw, time) |>
dplyr::summarise(mean_t = mean(g_mean)) |>
dplyr::group_by(case, time) |>
dplyr::summarise(
mean = mean(mean_t),
q5 = quantile(mean_t, 0.05, na.rm = TRUE),
q95 = quantile(mean_t, 0.95, na.rm = TRUE))
#> `summarise()` has grouped output by 'case', '.draw'. You can override using the
#> `.groups` argument.
#> `summarise()` has grouped output by 'case'. You can override using the
#> `.groups` argument.

It is also possible to do the marginalization over groups inside predict by using the argument funs

dynamite
which can be used to provide a named list of lists which contains functions to be applied for the
1.1.0
corresponding channel. This approach can save considerably amount of memory in case of large
https://docs.ropensci.org/dynamite/articles/dynamite.html 10/16
2/2/23, 1:41 PM dynamite • dynamite

number of groups. The names of the outermost list should be either channel names with or without
suffix _mean depending on whether the functions should be applied to response or mean. In our
case we can write

pred0b <- predict(multichannel_example_fit, newdata = newdata0, type = "mean",

funs = list(g_mean = list(mean_t = mean)))$simulated
pred1b <- predict(multichannel_example_fit, newdata = newdata1, type = "mean",
funs = list(g_mean = list(mean_t = mean)))$simulated
sumrb <- dplyr::bind_rows(
list(b0 = pred0b, b1 = pred1b), .id = "case") |>
dplyr::group_by(case, time) |>
dplyr::summarise(
mean = mean(mean_t_g_mean),
q5 = quantile(mean_t_g_mean, 0.05, na.rm = TRUE),
q95 = quantile(mean_t_g_mean, 0.95, na.rm = TRUE))

The resulting tibble sumrb is equal to the previous sumr (apart from randomness due to simulation
of new trajectories).

We can the visualize our predictions with the ggplot2 (Wickham 2016) as

library(ggplot2)
ggplot(sumr, aes(time, mean)) +
geom_ribbon(aes(ymin = q5, ymax = q95), alpha = 0.5) +
geom_line() +
scale_x_continuous(n.breaks = 10) +
facet_wrap(~ case) +
theme_bw()
#> Warning: Removed 5 rows containing missing values (`geom_line()`).

dynamite 1.1.0
https://docs.ropensci.org/dynamite/articles/dynamite.html 11/16
2/2/23, 1:41 PM dynamite • dynamite

Note that these estimates do indeed coincide with the causal effects (assuming of course that our
model is correct), as we can find the backdoor adjustment formula (Pearl 1995)

E(g t |do(b 5 = x)) = ∑ E(g t |b 5 = x, g 5 , p 5 )P (g 5 , p 5 ),

g 5 ,p 5

which is equal to mean_t in above codes.

We can also compute the difference P (g t |do(b 5 = 1)) − P (g t |do(b 5 = 0)) as

sumr_diff <- dplyr::bind_rows(

list(b0 = pred0, b1 = pred1), .id = "case") |>
dplyr::group_by(.draw, time) |>
dplyr::summarise(
mean_t = mean(g_mean[case == "b1"] - g_mean[case == "b0"])) |>
dplyr::group_by(time) |>
dplyr::summarise(
mean = mean(mean_t),
q5 = quantile(mean_t, 0.05, na.rm = TRUE),
q95 = quantile(mean_t, 0.95, na.rm = TRUE))
#> `summarise()` has grouped output by '.draw'. You can override using the
#> `.groups` argument.

ggplot(sumr_diff, aes(time, mean)) +

geom_ribbon(aes(ymin = q5, ymax = q95), alpha = 0.5) +
geom_line() +

dynamite
scale_x_continuous(n.breaks
1.1.0 = 10) +

https://docs.ropensci.org/dynamite/articles/dynamite.html 12/16
2/2/23, 1:41 PM dynamite • dynamite

theme_bw()
#> Warning: Removed 5 rows containing missing values (`geom_line()`).

Which shows that there is a short term effect of b on g, although the posterior uncertainty is quite
large. Note that for multistep causal effects these estimates contain also Monte Carlo variation due
to the simulation of the trajectories (e.g., there is also uncertainty in mean_t for fixed parameters
given finite number of individuals we are marginalizing over).

References
Allison, Paul D., Richard Williams, and Enrique Moral-Benito. 2017. “Maximum Likelihood for Cross-
Lagged Panel Models with Fixed Effects.” Socius 3: 2378023117710578.
https://doi.org/10.1177/2378023117710578.
Arellano, Manuel, and Stephen Bond. 1991. “Some Tests of Specification for Panel Data: Monte Carlo
Evidence and an Application to Employment Equations.” The Review of Economic Studies 58
(2): 277–97. https://doi.org/10.2307/2297968.
Bates, Douglas, Martin Mächler, Ben Bolker, and Steve Walker. 2015. “Fitting Linear Mixed-Effects
Models Using lme4.” Journal of Statistical Software 67 (1): 1–48.
https://doi.org/10.18637/jss.v067.i01.
Bergé, Laurent. 2018. “Efficient Estimation of Maximum Likelihood Models with Multiple Fixed-
Effects: The R Package FENmlm.” CREA Discussion Papers, no. 13.
Bollen, Kenneth A., and Jennie E. Brand. 2010. “A General Panel Model with Random and Fixed
Effects: A Structural Equations Approach.” Social Forces 89 (1): 1–34.
https://www.jstor.org/stable/40927552.
Brodersen, Kay H., Fabian Gallusser, Jim Koehler, Nicolas Remy, and Steven L. Scott. 2014. “Inferring

dynamite
Causal Impact Using Bayesian
1.1.0 Structural Time-Series Models.” Annals of Applied Statistics 9:
247–74. https://research.google/pubs/pub41854/.
https://docs.ropensci.org/dynamite/articles/dynamite.html 13/16
2/2/23, 1:41 PM dynamite • dynamite

Bürkner, Paul-Christian. 2018. “Advanced Bayesian Multilevel Modeling with the R Package brms.”
The R Journal 10 (1): 395–411. https://doi.org/10.32614/RJ-2018-017.
Casas, Isabel, and Ruben Fernandez-Casal. 2019. “tvReg: Time-Varying Coefficient Linear
Regression for Single and Multi-Equations in R.” {SSRN} {Scholarly} {Paper} ID 3363526.
Rochester, NY: Social Science Research Network. https://doi.org/10.2139/ssrn.3363526.
Croissant, Yves, and Giovanni Millo. 2008. “Panel Data Econometrics in R: The plm Package.”
Journal of Statistical Software 27 (2): 1–43. https://doi.org/10.18637/jss.v027.i02.
Durbin, James, and Siem Jan Koopman. 2012. Time Series Analysis by State Space Methods. 2nd
ed. New York: Oxford University Press.
Dziak, John J., Donna L. Coffman, Runze Li, Kaylee Litson, and Yajnaseni Chakraborti. 2021. Tvem:
Time-Varying Effect Models. https://CRAN.R-project.org/package=tvem.
Eubank, R. L., Chunfeng Huang, Y. Muñoz Maldonado, Naisyin Wang, Suojin Wang, and R. J.
Buchanan. 2004. “Smoothing Spline Estimation in Varying-Coefficient Models.” Journal of the
Royal Statistical Society. Series B (Statistical Methodology) 66 (3): 653–67.
https://www.jstor.org/stable/3647499.
Fixed Effects Regression Models. 2009. Thousand Oaks: SAGE Publications, Inc.
https://doi.org/10.4135/9781412993869.
Halekoh, Ulrich, Søren Højsgaard, and Jun Yan. 2006. “The r Package Geepack for Generalized
Estimating Equations.” Journal of Statistical Software 15/2: 1–11.
Hamaker, Ellen L, Rebecca M Kuiper, and Raoul PPP Grasman. 2015. “A Critique of the Cross-Lagged
Panel Model.” Psychological Methods 20 (1): 102.
Harvey, A. C., and G. D. A. Phillips. 1982. “The Estimation of Regression Models with Time-Varying
Parameters.” In Games, Economic Dynamics, and Time Series Analysis, edited by M. Deistler, E.
Fürst, and G. Schwödiauer, 306–21. Heidelberg: Physica-Verlag HD.
Hastie, Trevor, and Robert Tibshirani. 1993. “Varying-Coefficient Models.” Journal of the Royal
Statistical Society. Series B (Methodological) 55 (4): 757–96.
https://www.jstor.org/stable/2345993.
Hayakawa, Kazuhiko, and Jie Hou. 2019. “Estimation of Time-Varying Coefficient Dynamic Panel
Data Models.” Communications in Statistics - Theory and Methods 48 (13): 3311–24.
https://doi.org/10.1080/03610926.2018.1476704.
Helske, Jouni. 2022. “Efficient Bayesian Generalized Linear Models with Time-Varying Coefficients:
The Walker Package in r.” SoftwareX 18: 101016.
https://doi.org/https://doi.org/10.1016/j.softx.2022.101016.
Helske, Satu, and Jouni Helske. 2019. “Mixture Hidden Markov Models for Sequence Data: The
seqHMM Package in R.” Journal of Statistical Software 88 (3): 1–32.
https://doi.org/10.18637/jss.v088.i03.
Helske, Satu, Jouni Helske, and Mervi Eerola. 2018. “Combining Sequence Analysis and Hidden
Markov Models in the Analysis of Complex Life Sequence Data.” In Sequence Analysis and
Related Approaches: Innovative Methods and Applications, edited by Gilbert Ritschard and
Matthias Studer, 185–200. Cham: Springer International Publishing. https://doi.org/10.1007/978-
3-319-95420-2_11.
Knaus, Peter, Angela Bitto-Nemling, Annalisa Cadonna, and Sylvia Frühwirth-Schnatter. 2021.
dynamite
“Shrinkage in the Time-Varying
1.1.0 Parameter Model Framework Using the R Package shrinkTVP.”
Journal of Statistical Software 100 (November): 1–32. https://doi.org/10.18637/jss.v100.i13.
https://docs.ropensci.org/dynamite/articles/dynamite.html 14/16
2/2/23, 1:41 PM dynamite • dynamite

Lang, Stefan, and Andreas Brezger. 2004. “Bayesian P-Splines.” Journal of Computational and
Graphical Statistics 13 (1): 183–212. https://doi.org/10.1198/1061860043010.
Long, Jacob A. 2020. Panelr: Regression Models and Utilities for Repeated Measures and Panel
Data. https://cran.r-project.org/package=panelr.
Mulder, Jeroen D., and Ellen L. Hamaker. 2021. “Three Extensions of the Random Intercept Cross-
Lagged Panel Model.” Structural Equation Modeling: A Multidisciplinary Journal 28 (4): 638–48.
https://doi.org/10.1080/10705511.2020.1784738.
Pearl, Judea. 1995. “Causal Diagrams for Empirical Research.” Biometrika 82 (4): 669–88.
https://doi.org/10.1093/biomet/82.4.669.
R Core Team. 2021. R: A Language and Environment for Statistical Computing. Vienna, Austria: R
Foundation for Statistical Computing. https://www.R-project.org/.
Rosseel, Yves. 2012. “lavaan: An R Package for Structural Equation Modeling.” Journal of Statistical
Software 48 (2): 1–36. https://doi.org/10.18637/jss.v048.i02.
Stan Development Team. 2022a. “RStan: The R Interface to Stan.” https://mc-stan.org/.
———. 2022b. The Stan C++ Library. https://mc-stan.org/.
Wickham, Hadley. 2016. Ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag New York.
https://ggplot2.tidyverse.org.

On this page
Introduction

The Model

Defining The Model

Example of Multichannel Model

References

   
ABOUT

About rOpenSci
Software Review
Our Team
Jobs
Donate
Contact Us
dynamite
COMMUNITY
1.1.0

https://docs.ropensci.org/dynamite/articles/dynamite.html 15/16
2/2/23, 1:41 PM dynamite • dynamite

Our Community
Community calls
Events
Join the Discussion
Code of conduct
RESOURCES

Packages
Use Cases
Talks & Publications
Documentation
Newsletter
Cite rOpenSci

rOpenSci is a fiscally sponsored project of NumFOCUS.

https://docs.ropensci.org/dynamite/articles/dynamite.html 16/16