Dynamic State-Space Models of Coded Political Texts

Description

The function latpos allows to specify and estimate a dynamic state-space model of political texts and returns an object from which the (latent) positions, which actors take in their texts, can be predicted.

Usage

  latpos(formula,data,subset,id,time,
         unfold.method="Schoenemann",start=NULL,
         sampler=mvt.sampler(df=7*length(latent.dims)),
         ...)

  ## S4 method for signature 'latpos'
predict(object, newdata = NULL, id=NULL, time=NULL,
          type=c("posterior modes","posterior means","simulate"),
          se.fit=FALSE, interval=c("none","normal","percentile"), level=0.95,
          sample.size = object$sample.size,
          sampler=object$sampler,
          maxiter=100,...)

Arguments

formula

a formula describing the model.

data

an optional data frame that contains data to which the model is fitted.

subset

a logical vector; an optional condition that defines a subset of the optiona data frame to which the analysis is to be restricted.

id

name of a variable that identifies the actors.

time

name of a variable that identifies the occasions on which actors publish their political texts.

unfold.method

a text string, name of the method which is used in the unfolding procedure to obtain starting values.

start

an optional list giving starting values.

sampler

an object that generates random numbers to be used in the MCEM algorithm.

...

further optional arguments, that are passed to latpos.control and latpos.start by the function latpos and ignored by the predict method.

object

an object of class “latpos”.

newdata

an optional data frame to predict actors positions from, based on the fitted model contained in object.

type

a character string identifying the type of predictions about the latent positions, either “posterior modes”,”posterior means”, or “simulate”. If the type is “posterior means” or “posterior modes”, only one prediction is generated per coded text. If the type is “simulate”, then sample.size predictions are generated from the empirical Bayes posterior distribution of the positions.

se.fit

logical, should standard errors be provided with the posterior means?

interval

a character vector. If “none”, then no prediction intervals are returned, if “normal” than prediction intervals are contstructed based on normal quantiles, if “percentile”, then predicion intervals are constructed based on the empirical percentiles of samples from the posterior.

level

a number between zero and one, the nominal coverage level of the prediction intervals.

sample.size

a positive number, the sample size for the simulated posterior means, or the number of imputed values generated from the posterior.

maxiter

a positive number, the maximum number of iterations used to compute posterior modes.

Value

latpos returns an object of class “latpos” that contains parameter estimates etc.

predict.latpos returns a vector or a matrix depending on the arguments provided to the function.

Examples

load(file="manifesto-counts.RData")
options(latpos.chunk.size=4*1024*1024) # determines the size of
# data chunks used in internal computations.

econ.latpos <- latpos(
   nationalize+controlecon+econplan+marketregul+incentives+econorthodox+freeenterp~econlr,
   data=manifesto.counts,
   id=party,
   time=year*100 + month,
   free.beta=TRUE,
   initial.size=50
 )

summary(econ.latpos)

nonmat.latpos <- latpos(
    laworder+tradmoralpos+tradmoralneg+natwaylifepos+
    democracy+freedmhumrights~authlib+tradmod,
    data=manifesto.counts,
    id=party,
    time=year*100 + month,
    initial.size=50,
    restrictions=set.parms.free(
      authlib=c("natwaylifepos","laworder","democracy","freedmhumrights"),
      tradmod=c("tradmoralpos","tradmoralneg","democracy","freedmhumrights")
    ),
    start = list(
          A=list(
              authlib=c(natwaylifepos=1.4,laworder=.7,democracy=-1),
              tradmod=c(tradmoralpos=1,tradmoralneg=-1)
            )
        )
)
summary(nonmat.latpos)
© Copyright 2022, Martin Elff. Last updated on 28 Aug 2024. Created using Sphinx 7.2.6. Page source