R/LogSeq.R

In DySeq: Functions for Dyadic Sequence Analyses

#'LogSeq
#'
#'Implementation of Bakeman & Gottman (1997) for sequence analysis.
#'Kenny, Kashy & Cook (2006) provide further examples.
#'
#'\itemize{
#' \item Runs logit models over a multiple number of state-transition tables, see: \code{\link{StateTrans}} 
#' \item Aggregates coefficients of all logit models and tests them against zero. 
#' \item If subgroups are defined, coefficients are tested to be different between groups.
#' \item Print-function displays mean logit-coefficients and p-values. 
#'}
#'
#'@param x a state.trans object or a list of 4*2 state-transition tables
#'@param delta constant added to every cell, required if zero frequencies occur!
#'@param subgroups an optional vector containing groupmembership if estimates should be compared between groups
#'@param single.case should p-values be computed for single case analysis
#'
#'@references
#'\itemize{
#'  \item Bakeman, R., & Gottman, J. M. (1997) <DOI: 10.1017/cbo9780511527685 >
#'  \item Kenny, D. A., Kashy, D. A., & Cook, W. L. (2006) <DOI: 10.1177/1098214007300894>
#'}
#'
#'
#'
#'
#'@examples
#'\dontrun{
#'data(CouplesCope)
#'my.states<-StateExpand(CouplesCope, 2:49, 50:97)
#'my.trans<-StateTrans(my.states, FALSE)
#'my.logseq<-LogSeq(my.trans, single.case=TRUE)
#'my.logseq
#'
#'plot(my.logseq) # interaction can be plotted
#'
#'single.LogSeq(my.logseq, 41) # for single case analysis
#'}
#'@export


LogSeq<-function(x, delta=0.5, subgroups=NA, single.case=FALSE){

  if(class(x)[2]!="state.trans") warning("x should be a state.trans object. See Help(StateTrans)!")


  lambdas<-matrix(NA, length(x),4) # Empty table for lambdas

  p.values<-matrix(NA, length(x),4)

  for(i in 1:length(x)){

    x.long<-c(x[[i]][1:4,1],x[[i]][1:4,2])

    casearray<-array(x.long, c(2,2,2), list(c("seq2_1", "seq2_0"),c("seq1_1","seq2_0"), c("dep1","dep0")))

    caselog<-MASS::loglm(~1+2+3+1*2*3, data=(casearray+delta), fit=F)

    if(single.case==TRUE){
      caselog.b1<-MASS::loglm(~1+2+3+2*3,1*2, data=(casearray+delta), fit=F)
      caselog.b2<-MASS::loglm(~1+2+3+1*3+2*3, data=(casearray+delta), fit=F)
      caselog.b3<-MASS::loglm(~1+2+3+2*3+1*3+1*2, data=(casearray+delta), fit=F)

      p1<-stats::anova(caselog.b1,caselog.b3)[2,5]
      p2<-stats::anova(caselog.b2,caselog.b3)[2,5]
      p3<-unclass(summary(caselog.b3))$tests[2,3]

      p.values[i,1]<-NA
      p.values[i,3]<-unlist(p1)# Ist Partner daher Col==3 
      p.values[i,2]<-unlist(p2)# Ist Actor daher Col==2
      p.values[i,4]<-unlist(p3)
    }

    b0<-stats::coef(caselog)$"3"[1]-stats::coef(caselog)$"3"[2]   #mean
    b1<-stats::coef(caselog)$"1.3"[1,1]-stats::coef(caselog)$"1.3"[1,2]   #partner
    b2<-stats::coef(caselog)$"2.3"[1,1]-stats::coef(caselog)$"2.3"[1,2]   #actor
    b3<-stats::coef(caselog)$"1.2.3"[1,1,1]-stats::coef(caselog)$"1.2.3"[1,1,2]   #interaction

    lambdas[i,1]<-unlist(b0)
    lambdas[i,3]<-unlist(b1) # Ist Partner daher Col==3 
    lambdas[i,2]<-unlist(b2) # Ist Actor daher Col==2
    lambdas[i,4]<-unlist(b3)
  }

  output<-list()

  output[[1]]<-lambdas
  output[[2]]<-x
  output[[3]]<-subgroups
  output[[4]]<-p.values

  attr(output, "firstSeq")<-attr(x, "firstSeq")
  

  class(output)[1]<-"LogSeq"

  return(output)

}