A near-equivalent of wbm() that instead uses Stan, via rstan and brms.

wbm_stan(formula, data, id = NULL, wave = NULL, model = "w-b",
detrend = FALSE, use.wave = FALSE, wave.factor = FALSE,
min.waves = 2, model.cor = FALSE, family = gaussian,
fit_model = TRUE, balance.correction = FALSE, dt.random = TRUE,
dt.order = 1, chains = 3, iter = 2000, scale = FALSE,
save_ranef = FALSE, interaction.style = c("double-demean", "demean",
"raw"), weights = NULL, offset = NULL, ...)

## Arguments

formula Model formula. See details for crucial info on panelr's formula syntax. The data, either a panel_data object or data.frame. If data is not a panel_data object, then the name of the individual id column as a string. Otherwise, leave as NULL, the default. If data is not a panel_data object, then the name of the panel wave column as a string. Otherwise, leave as NULL, the default. One of "w-b", "within", "between", "contextual". See details for more on these options. Adjust within-subject effects for trends in the predictors? Default is FALSE, but some research suggests this is a better idea (see Curran and Bauer (2011) reference). Should the wave be included as a predictor? Default is FALSE. Should the wave variable be treated as an unordered factor instead of continuous? Default is FALSE. What is the minimum number of waves an individual must have participated in to be included in the analysis? Default is 2 and any valid number is accepted. "all" is also acceptable if you want to include only complete panelists. Do you want to model residual autocorrelation? This is often appropriate for linear models (family = gaussian). Default is FALSE to be consistent with wbm(), reduce runtime, and avoid warnings for non-linear models. Use this to specify GLM link families. Default is gaussian, the linear model. Fit the model? Default is TRUE. If FALSE, only the model code is returned. Correct between-subject effects for unbalanced panels following the procedure in Curran and Bauer (2011)? Default is FALSE. Should the detrending procedure be performed with a random slope for each entity? Default is TRUE but for short panels FALSE may be better, fitting a trend for all entities. If detrending using detrend, what order polynomial would you like to specify for the relationship between time and the predictors? Default is 1, a linear model. How many Markov chains should be used? Default is 3, to leave you with one unused thread if you're on a typical dual-core machine. How many iterations, including warmup? Default is 2000, leaving 1000 per chain after warmup. For some models and data, you may need quite a few more. Standardize predictors? This can speed up model fit. Default is FALSE. Save random effect estimates? This can be crucial for predicting from the model and for certain post-estimation procedures. On the other hand, it drastically increases the size of the resulting model. Default is FALSE. The best way to calculate interactions in within models is in some dispute. The conventional way ("demean") is to first calculate the product of the variables involved in the interaction before those variables have their means subtracted and then subtract the mean of the product from the product term (see Schunk and Perales (2017)). Giesselmann and Schmidt-Catran (2018) show this method carries between-entity differences that within models are designed to model out. They suggest an alternate method ("double-demean") in which the product term is first calculated using the de-meaned lower-order variables and then the subject means are subtracted from this product term. Another option is to simply use the product term of the de-meaned variables ("raw"), but Giesselmann and Schmidt-Catran (2018) show this method biases the results towards zero effect. The default is "double-demean" but if emulating other software is the goal, "demean" might be preferred. If using weights, either the name of the column in the data that contains the weights or a vector of the weights. this can be used to specify an a priori known component to be included in the linear predictor during fitting. This should be NULL or a numeric vector of length equal to the number of cases. One or more offset terms can be included in the formula instead or as well, and if more than one is specified their sum is used. See model.offset. Additional arguments passed on to brms::brm(). This can include specification of priors.

## Value

A wbm_stan object, which is a list containing a model object with the brm model and a stan_code object with the model code.

If fit_model = FALSE, instead a list is returned containing a stan_code object and a stan_data object, leaving you with the tools you need to run the model yourself using rstan.

## Details

See wbm() for details on the formula syntax, model types, and some other stuff.

wbm()

## Examples

# NOT RUN {
data("WageData")
wages <- panel_data(WageData, id = id, wave = t)
model <- wbm_stan(lwage ~ lag(union) + wks | blk + fem | blk * lag(union),
data = wages, chains = 1, iter = 2000)
summary(model)
# }