R/relrisk_analysis.R

In spsurvey: Spatial Sampling Design and Analysis

################################################################################
# Function: relrisk_analysis (exported)
# Programmer: Tom Kincaid
# Date: July 16, 2020
# Revised: August 14, 2020 to allow use of an sf object as the input data
#          argument (dframe)
# Revised: December 15, 2020 to use a new function named survey_design to
#          create the survey design object
# Revised: March 2, 2021 to revise the process for creating unique site ID
#          values
# Revised: April 7, 2021 to ensure that the dframe argument does not contain
#          zero rows
# Revised: April 29, 2021 to ensure that the dframe argument only belongs to
#          class "data.frame"
# Revised: May 4, 2021 to avoid warning messages being generated during creation
#          of help files
# Revised: May 6, 2021 to ensure that sf objects do not belong to class tbl_df
# Revised: June 8, 2021 to simplify specification of the values required for
#          calculation of the finite population correction factor and to
#          eliminate use of the finite population correction factor with the
#          local mean variance estimator
# Revised: June 11, 2021 to modify the approach for specification of arguments
#          response_levels and stressor_levels
# Revised: July 28, 2021 to improve documentation
# Revised: September 9, 2021 to revise the documentation for argument popsize
#
#' Relative risk analysis
#'
#' This function organizes input and output for relative risk analysis (of
#' categorical variables).  The analysis data,
#' \code{dframe}, can be either a data frame or a simple features (\code{sf}) object.  If an
#' \code{sf} object is used, coordinates are extracted from the geometry column in the
#' object, arguments \code{xcoord} and \code{ycoord} are assigned values
#' \code{"xcoord"} and \code{"ycoord"}, respectively, and the geometry column is
#' dropped from the object.
#'
#' @inherit attrisk_analysis params
#'
#' @section Details:
#' Relative risk measures the relative strength of association between
#' conditional probabilities defined for a response variable and a stressor
#' variable, where the response and stressor variables are classified as either
#' good (i.e., reference condition) or poor (i.e., different from reference
#' condition). Relative risk is defined as the ratio of two conditional
#' probabilities.  The numerator of the ratio is the probability that the
#' response variable is in poor condition given that the stressor variable is in
#' poor condition.   The denominator of the ratio is the probability that the
#' response variable is in poor condition given that the stressor variable is in
#' good condition. A relative risk value equal to one indicates that the
#' response variable is independent of the stressor variable.  Relative risk
#' values greater than one measure the extent to which poor condition of the
#' stressor variable is associated with poor condition of the response variable.
#'
#' @return The analysis results. A data frame of population estimates for all combinations of
#'   subpopulations, categories within each subpopulation, response variables,
#'   and categories within each response variable.  Estimates are provided for
#'   proportion and size of the population plus standard error, margin of error,
#'   and confidence interval estimates. The data frame contains the following
#'   variables:
#'   \describe{
#'     \item{Type}{subpopulation (domain) name}
#'     \item{Subpopulation}{subpopulation name within a domain}
#'     \item{Response}{response variable}
#'     \item{Stressor}{stressor variable}
#'     \item{nResp}{sample size}
#'     \item{Estimate}{relative risk estimate}
#'     \item{Estimate_num}{relative risk numerator estimate}
#'     \item{Estimate_denom}{relative risk denominator estimate}
#'     \item{StdError}{relative risk standard error}
#'     \item{MarginofError}{relative risk margin of error}
#'     \item{LCBxxPct}{xx\% (default 95\%) lower confidence bound}
#'     \item{UCBxxPct}{xx\% (default 95\%) upper confidence bound}
#'     \item{WeightTotal}{sum of design weights}
#'     \item{Count_RespPoor_StressPoor}{number of observations in the poor response and poor stressor group}
#'     \item{Count_RespPoor_StressGood}{number of observations in the poor response and good stressor group}
#'     \item{Count_RespGood_StressPoor}{number of observations in the good response and poor stressor group}
#'     \item{Count_RespGood_StressGood}{number of observations in the good response and good stressor group}
#'     \item{Prop_RespPoor_StressPoor}{weighted proportion of observations in the poor response and poor stressor group}
#'     \item{Prop_RespPoor_StressGood}{weighted proportion of observations in the poor response and good stressor group}
#'     \item{Prop_RespGood_StressPoor}{weighted proportion of observations in the good response and poor stressor group}
#'     \item{Prop_RespGood_StressGood}{weighted proportion of observations in the good response and good stressor group}
#'   }
#'
#' @author Tom Kincaid \email{Kincaid.Tom@@epa.gov}
#'
#' @keywords survey univar
#'
#' @seealso
#'   \describe{
#'   \item{\code{\link{attrisk_analysis}}}{ for attributable risk analysis}
#'   \item{\code{\link{diffrisk_analysis}}}{ for risk difference analysis}
#'   }
#'
#' @examples
#' dframe <- data.frame(
#'   siteID = paste0("Site", 1:100),
#'   wgt = runif(100, 10, 100),
#'   xcoord = runif(100),
#'   ycoord = runif(100),
#'   stratum = rep(c("Stratum1", "Stratum2"), 50),
#'   RespVar1 = sample(c("Poor", "Good"), 100, replace = TRUE),
#'   RespVar2 = sample(c("Poor", "Good"), 100, replace = TRUE),
#'   StressVar = sample(c("Poor", "Good"), 100, replace = TRUE),
#'   All_Sites = rep("All Sites", 100),
#'   Resource_Class = rep(c("Agr", "Forest"), c(55, 45))
#' )
#' myresponse <- c("RespVar1", "RespVar2")
#' mystressor <- c("StressVar")
#' mysubpops <- c("All_Sites", "Resource_Class")
#' relrisk_analysis(dframe,
#'   vars_response = myresponse,
#'   vars_stressor = mystressor, subpops = mysubpops, siteID = "siteID",
#'   weight = "wgt", xcoord = "xcoord", ycoord = "ycoord",
#'   stratumID = "stratum"
#' )
#' @export
################################################################################

relrisk_analysis <- function(dframe, vars_response, vars_stressor, response_levels = NULL,
                             stressor_levels = NULL, subpops = NULL, siteID = NULL, weight = "weight",
                             xcoord = NULL, ycoord = NULL, stratumID = NULL, clusterID = NULL,
                             weight1 = NULL, xcoord1 = NULL, ycoord1 = NULL, sizeweight = FALSE,
                             sweight = NULL, sweight1 = NULL, fpc = NULL, popsize = NULL,
                             vartype = "Local", conf = 95, All_Sites = FALSE) {

  # Create a vector for error messages

  error_ind <- FALSE
  error_vec <- NULL

  # Create a data frame for warning messages

  warn_ind <- FALSE
  warn_df <- NULL
  fname <- "relrisk_analysis"

  # Ensure that the dframe argument was provided

  if (missing(dframe) | is.null(dframe)) {
    stop("\nThe dframe argument must be provided.\n")
  }

  # If the dframe argument is an sf object, extract coordinates from the
  # geometry column, assign values "xcoord" and "ycoord" to arguments xcoord and
  # ycoord, respectively, and drop the geometry column from the object

  if ("sf" %in% class(dframe)) {
    temp <- st_coordinates(dframe)
    xcoord <- "xcoord"
    dframe$xcoord <- temp[, "X"]
    ycoord <- "ycoord"
    dframe$ycoord <- temp[, "Y"]
    dframe <- st_set_geometry(dframe, NULL)
  }

  # If the dframe argument is a tibble or does not belong to class
  # "data.frame", coerce the argument to class "data.frame"

  if ("tbl_df" %in% class(dframe) | !("data.frame" %in% class(dframe))) {
    dframe <- as.data.frame(dframe)
  }

  # Ensure that the dframe argument does not contain zero rows

  if (nrow(dframe) == 0) {
    stop("\nThe dframe argument contains zero rows.\n")
  }

  # Ensure that unused levels are dropped from factor variables in the dframe
  # data frame

  dframe <- droplevels(dframe)

  # If no siteID is provided, set one that assumes each row is a unique site

  if (is.null(siteID)) {
    siteID <- "siteID"
    dframe$siteID <- paste("site", seq_len(nrow(dframe)), sep = "-")
  }

  # Ensure that the dframe data frame contains the site ID variable

  if (!(siteID %in% names(dframe))) {
    ind <- FALSE
    error_ind <- TRUE
    msg <- paste0("The name provided for the siteID argument, \"", siteID, "\", does not occur among \nthe names for the dframe data frame.\n")
    error_vec <- c(error_vec, msg)
  } else {
    ind <- TRUE
  }

  # Check site IDs for repeat values and, as necessary, create unique site IDs
  # and output a warning message

  if (ind) {
    IDs <- dframe[, siteID]
    temp <- sapply(split(IDs, IDs), length)
    if (any(temp > 1)) {
      warn_ind <- TRUE
      temp.str <- vecprint(names(temp)[temp > 1])
      warn <- paste("The following site ID values occur more than once among the values that were \ninput to the function:\n", temp.str)
      act <- "Unique site ID values were created.\n"
      warn_df <- rbind(warn_df, data.frame(
        func = I(fname), subpoptype = NA,
        subpop = NA, indicator = NA, stratum = NA, warning = I(warn), action = I(act)
      ))
      dframe[, siteID] <- uniqueID(dframe[, siteID])
    }
  }

  # Ensure that the dframe data frame contains the survey weight variable

  if (!(weight %in% names(dframe))) {
    error_ind <- TRUE
    msg <- paste0("The name provided for the weight argument, \"", weight, "\", does not occur among \nthe names for the dframe data frame.\n")
    error_vec <- c(error_vec, msg)
  }

  # Assign names to the variables required for calculation of the finite
  # population correction factor

  if (is.null(fpc)) {
    fpcfactor_ind <- FALSE
    fpcsize <- NULL
    Ncluster <- NULL
    stage1size <- NULL
  } else {
    fpcfactor_ind <- TRUE
    if (is.null(clusterID)) {
      fpcsize <- "fpcsize"
      Ncluster <- NULL
      stage1size <- NULL
    } else {
      fpcsize <- NULL
      Ncluster <- "Ncluster"
      stage1size <- "stage1size"
    }
  }

  # Create a list containing names of survey design variables

  design_names <- list(
    siteID = siteID,
    weight = weight,
    xcoord = xcoord,
    ycoord = ycoord,
    stratumID = stratumID,
    clusterID = clusterID,
    weight1 = weight1,
    xcoord1 = xcoord1,
    ycoord1 = ycoord1,
    sweight = sweight,
    sweight1 = sweight1,
    fpcsize = fpcsize,
    Ncluster = Ncluster,
    stage1size = stage1size
  )

  # Ensure that a value was provided for the vars_response (response variable
  # names) argument

  if (missing(vars_response)) {
    error_ind <- TRUE
    msg <- "A value must be provided for the vars_response (response variable names) argument.\n"
    error_vec <- c(error_vec, msg)
  }

  # Ensure that a value was provided for the vars_stressor (stressor variable
  # names) argument

  if (missing(vars_stressor)) {
    error_ind <- TRUE
    msg <- "A value must be provided for the vars_stressor (stressor variable names) argument.\n"
    error_vec <- c(error_vec, msg)
  }

  # If a value was not provided for the subpops (subpopulation names) argument,
  # assign the value "All_Sites" to the subpops argument and create a factor
  # named "All_Sites" in the dframe data frame that takes the value "All Sites"

  if (is.null(subpops)) {
    subpops <- "All_Sites"
    dframe$All_Sites <- "All Sites"
    dframe$All_Sites <- factor(dframe$All_Sites)
  }

  # If the user wants information for all sites together in addition to the
  # subpops, add the value "All_Sites" to the subpops argument and create a
  # factor named "All_Sites" in the dframe data frame that takes the value
  # "All Sites"

  if (!is.null(subpops) && All_Sites) {
    subpops <- c(subpops, "All_Sites")
    dframe$All_Sites <- "All Sites"
    dframe$All_Sites <- factor(dframe$All_Sites)
  }

  # Check input arguments

  temp <- input_check(dframe, design_names, vars_response, NULL, vars_stressor,
    NULL, subpops, sizeweight, fpc, popsize, vartype, NULL, conf,
    error_ind = error_ind, error_vec = error_vec
  )
  dframe <- temp$dframe
  vars_response <- temp$vars_cat
  vars_stressor <- temp$vars_stressor
  subpops <- temp$subpops
  popsize <- temp$popsize
  vartype <- temp$vartype
  error_ind <- temp$error_ind
  error_vec <- temp$error_vec

  # Check arguments response_levels and stressor_levels

  if (is.null(response_levels)) {
    response_levels <- rep(list(c("Poor", "Good")), length(vars_response))
    names(response_levels) <- vars_response
  } else {
    if (!is.list(response_levels)) {
      error_ind <- TRUE
      msg <- "Argument response_levels must be a list.\n"
      error_vec <- c(error_vec, msg)
    } else {
      if (length(response_levels) != length(vars_response)) {
        error_ind <- TRUE
        msg <- "Argument response_levels must be the same length as argument vars_response.\n"
        error_vec <- c(error_vec, msg)
      }
      if (is.null(names(response_levels))) { # set default names if none provided
        names(response_levels) <- vars_response
      }
      if (any(sapply(response_levels, function(x) length(x) != 2))) {
        error_ind <- TRUE
        msg <- "Each element of argument response_levels must contain only two values.\n"
        error_vec <- c(error_vec, msg)
      }
      if (any(sapply(response_levels, function(x) !is.character(x)))) {
        error_ind <- TRUE
        msg <- "Each element of argument response_levels must contain character values.\n"
        error_vec <- c(error_vec, msg)
      }
      if (all(vars_response %in% names(dframe))) {
        tst <- logical(length(vars_response))
        for (i in 1:length(vars_response)) {
          if (!all(response_levels[[i]] %in% levels(dframe[, vars_response[i]]))) {
            tst[i] <- TRUE
          }
        }
        if (any(tst)) {
          temp_str <- vecprint(vars_response[tst])
          error_ind <- TRUE
          msg <- paste0("\nCategory names for the following response variables do not match category names \nin the response_levels argument:\n", temp_str)
          error_vec <- c(error_vec, msg)
        } else {
          if (!all(names(response_levels) %in% vars_response)) {
            error_ind <- TRUE
            msg <- "Names for the response_levels list do not match the values in the vars_response \nargument."
            error_vec <- c(error_vec, msg)
          } else {
            indx <- match(vars_response, names(response_levels))
            response_levels <- response_levels[indx]
          }
        }
      }
    }
  }

  if (is.null(stressor_levels)) {
    stressor_levels <- rep(list(c("Poor", "Good")), length(vars_stressor))
    names(stressor_levels) <- vars_stressor
  } else {
    if (!is.list(stressor_levels)) {
      error_ind <- TRUE
      msg <- "Argument stressor_levels must be a list.\n"
      error_vec <- c(error_vec, msg)
    } else {
      if (length(stressor_levels) != length(vars_stressor)) {
        error_ind <- TRUE
        msg <- "Argument stressor_levels must be the same length as argument vars_stressor.\n"
        error_vec <- c(error_vec, msg)
      }
      if (is.null(names(stressor_levels))) { # set default names if none provided
        names(stressor_levels) <- vars_stressor
      }
      if (any(sapply(stressor_levels, function(x) length(x) != 2))) {
        error_ind <- TRUE
        msg <- "Each element of argument stressor_levels must contain only two values.\n"
        error_vec <- c(error_vec, msg)
      }
      if (any(sapply(stressor_levels, function(x) !is.character(x)))) {
        error_ind <- TRUE
        msg <- "Each element of argument stressor_levels must contain character values.\n"
        error_vec <- c(error_vec, msg)
      }
      if (all(vars_stressor %in% names(dframe))) {
        tst <- logical(length(vars_stressor))
        for (i in 1:length(vars_stressor)) {
          if (!all(stressor_levels[[i]] %in% levels(dframe[, vars_stressor[i]]))) {
            tst[i] <- TRUE
          }
        }
        if (any(tst)) {
          temp_str <- vecprint(vars_stressor[tst])
          error_ind <- TRUE
          msg <- paste0("\nCategory names for the following stressor variables do not match category names \nin the stressor_levels argument:\n", temp_str)
          error_vec <- c(error_vec, msg)
        } else {
          if (!all(names(stressor_levels) %in% vars_stressor)) {
            error_ind <- TRUE
            msg <- "Names for the stressor_levels list do not match the values in the vars_stressor \nargument."
            error_vec <- c(error_vec, msg)
          } else {
            indx <- match(vars_stressor, names(stressor_levels))
            stressor_levels <- stressor_levels[indx]
          }
        }
      }
    }
  }

  # As necessary, output a message indicating that error messages were generated
  # during execution of the program

  if (error_ind) {
    error_vec <<- error_vec
    if (length(error_vec) == 1) {
      message("During execution of the program, an error message was generated.  The error \nmessage is stored in a vector named 'error_vec'.  Enter the following command \nto view the error message: errorprnt()\n")
    } else {
      message(paste("During execution of the program,", length(error_vec), "error messages were generated.  The error \nmessages are stored in a vector named 'error_vec'.  Enter the following \ncommand to view the error messages: errorprnt()\n"))
    }

    if (warn_ind) {
      warn_df <<- warn_df
      if (nrow(warn_df) == 1) {
        message("During execution of the program, a warning message was generated.  The warning \nmessage is stored in a data frame named 'warn_df'.  Enter the following command \nto view the warning message: warnprnt()\n")
      } else {
        message(paste("During execution of the program,", nrow(warn_df), "warning messages were generated.  The warning \nmessages are stored in a data frame named 'warn_df'.  Enter the following \ncommand to view the warning messages: warnprnt() \nTo view a subset of the warning messages (say, messages number 1, 3, and 5), \nenter the following command: warnprnt(m=c(1,3,5))\n"))
      }
    }
    stop("See the preceding message(s).")
  }

  # Assign a logical value to the indicator variable for a stratified sample

  stratum_ind <- !is.null(stratumID)

  # For a stratified sample, remove strata that contain a single site

  if (stratum_ind) {
    dframe[, stratumID] <- factor(dframe[, stratumID])
    stratum_levels <- levels(dframe[, stratumID])
    nstrata <- length(stratum_levels)
    ind <- FALSE
    for (i in 1:nstrata) {
      tst <- dframe[, stratumID] == stratum_levels[i]
      if (sum(tst) == 1) {
        warn_ind <- TRUE
        warn <- paste("The stratum named \"", stratum_levels[i], "\" contains a single value and was removed from the analysis.\n")
        act <- "Stratum was removed from the analysis.\n"
        warn_df <- rbind(warn_df, data.frame(
          func = I(fname), subpoptype = NA,
          subpop = NA, indicator = NA, stratum = NA, warning = I(warn), action = I(act)
        ))
        dframe <- dframe[!tst, ]
        ind <- TRUE
      }
    }
    if (ind) {
      dframe[, stratumID] <- factor(dframe[, stratumID])
      stratum_levels <- levels(dframe[, stratumID])
      nstrata <- length(stratum_levels)
    }
  }

  # Assign a logical value to the indicator variable for a two-stage sample

  cluster_ind <- !is.null(clusterID)

  # Create the survey design object

  design <- survey_design(
    dframe, siteID, weight, stratum_ind, stratumID, cluster_ind, clusterID,
    weight1, sizeweight, sweight, sweight1, fpcfactor_ind, fpcsize, Ncluster,
    stage1size, vartype, NULL
  )

  # If popsize is not equal to NULL, then call either the postStratify or
  # calibrate function, as appropriate

  if (!is.null(popsize)) {
    if (all(class(popsize) %in% c("data.frame", "table", "xtabs"))) {
      if ("data.frame" %in% class(popsize)) {
        pnames <- names(popsize)[-ncol(popsize)]
      } else {
        pnames <- names(dimnames(popsize))
      }
      design <- postStratify(design, make.formula(pnames), popsize)
    } else {
      cnames <- cal_names(make.formula(names(popsize)), design)
      pop_totals <- numeric(length(cnames))
      names(pop_totals) <- cnames
      pop_totals[1] <- sum(popsize[[1]])
      k <- 2
      for (i in names(popsize)) {
        temp <- popsize[[i]]
        for (j in 2:length(temp)) {
          pop_totals[k] <- temp[j]
          k <- k + 1
        }
      }
      design <- calibrate(design, make.formula(names(popsize)), pop_totals)
    }
  }

  # If popsize is not equal to NULL and vartype equals "Local", then assign
  # adjusted weights to the appropriate weight variable(s) in the
  # design$variables data frame

  if (!is.null(popsize) && vartype == "Local") {
    if (cluster_ind) {
      design$variables$wgt2 <- weights(design) / design$variables$wgt1
    } else {
      design$variables$wgt <- weights(design)
    }
  }

  # For variables that exist in the design$variables data frame, assign survey
  # design variables

  dframe <- design$variables
  for (i in names(design_names)) {
    if (is.null(design_names[[i]])) {
      eval(parse(text = paste0(i, " <- NULL")))
    } else {
      eval(parse(text = paste0(i, " <- dframe[, \"", design_names[[i]], "\"]")))
    }
  }

  # Assign values to weight variables

  if (cluster_ind) {
    wgt1 <- dframe$wgt1
    wgt2 <- dframe$wgt2
  } else {
    wgt <- dframe$wgt
  }

  # Create the rrsum (results) data frame

  rrsum <- NULL

  # Assign the confidence bound multiplier

  mult <- qnorm(0.5 + (conf / 100) / 2)

  # Loop through all subpopulations (domains)

  for (itype in subpops) {
    lev_itype <- levels(dframe[, itype])
    nlev_itype <- length(lev_itype)

    # Loop through all response variables (vars_response)

    for (ivar_r in vars_response) {

      # Loop through all stressor variables (vars_stressor)

      for (ivar_s in vars_stressor) {

        # Loop through all levels of the subpopulation

        for (isubpop in lev_itype) {
          tst <- !is.na(dframe[, itype]) & dframe[, itype] == isubpop &
            !is.na(dframe[, ivar_r]) & !is.na(dframe[, ivar_s])

          # Assign response variable values

          response <- dframe[, ivar_r]
          nresp <- length(response)

          # Assign stressor variable values

          stressor <- dframe[, ivar_s]

          # Create the warn_vec object

          warn_vec <- c(itype, isubpop, paste(ivar_r, "and", ivar_s))

          #
          # Branch to handle stratified and unstratified data
          #

          if (stratum_ind) {

            #
            # Begin the section for stratified data
            #

            # Initialize variables for all strata combined

            wgt_total <- 0
            varest <- 0

            #
            # Begin the subsection for individual strata
            #

            for (i in 1:nstrata) {

              # Calculate required values

              stratum_i <- tst & stratumID == stratum_levels[i]
              response_st <- response[stratum_i]
              stressor_st <- stressor[stratum_i]
              if (cluster_ind) {
                wgt1_st <- wgt1[stratum_i]
                wgt2_st <- wgt2[stratum_i]
              } else {
                wgt_st <- wgt[stratum_i]
              }

              # Compute the 2x2 table of weight totals

              wgt_total_st <- svytable(make.formula(paste(
                ivar_r, "+",
                ivar_s
              )), design = subset(design, stratum_i))

              # Calculate the variance-covariance estimate for the cell and
              # marginal totals

              if (cluster_ind) {
                temp <- relrisk_var(
                  response[stratum_i], stressor[stratum_i],
                  response_levels[[ivar_r]], stressor_levels[[ivar_s]],
                  wgt2[stratum_i], xcoord[stratum_i], ycoord[stratum_i],
                  stratum_ind, stratum_levels[i], cluster_ind,
                  clusterID[stratum_i], wgt1[stratum_i], xcoord1[stratum_i],
                  ycoord1[stratum_i], vartype, warn_ind, warn_df, warn_vec
                )
              } else {
                temp <- relrisk_var(response[stratum_i], stressor[stratum_i],
                  response_levels[[ivar_r]], stressor_levels[[ivar_s]],
                  wgt[stratum_i], xcoord[stratum_i], ycoord[stratum_i],
                  stratum_ind, stratum_levels[i], cluster_ind,
                  vartype = vartype, warn_ind = warn_ind, warn_df = warn_df,
                  warn_vec = warn_vec
                )
              }
              varest_st <- temp$varest
              warn_ind <- temp$warn_ind
              warn_df <- temp$warn_df

              # Add estimates to the variables for all strata combined

              wgt_total <- wgt_total + wgt_total_st
              varest <- varest + varest_st

              #
              # End the subsection for individual strata
              #
            }

            # Add margins to the wgt_total table

            wgt_total <- addmargins(wgt_total)

            # Assign cell and marginal weight totals

            total1 <- wgt_total[
              response_levels[[ivar_r]][1],
              stressor_levels[[ivar_s]][1]
            ]
            total2 <- wgt_total["Sum", stressor_levels[[ivar_s]][1]]
            total3 <- wgt_total[
              response_levels[[ivar_r]][1],
              stressor_levels[[ivar_s]][2]
            ]
            total4 <- wgt_total["Sum", stressor_levels[[ivar_s]][2]]

            # Calculate the estimate of relative risk for all strata combined

            if (total2 == 0 || total4 == 0) {
              rr <- NA
              rr_num <- NA
              rr_denom <- NA
              warn_ind <- TRUE
              temp <- ifelse(total2 == 0, stressor_levels[[ivar_s]][1],
                stressor_levels[[ivar_s]][2]
              )
              warn <- paste0("Since there are no observations for level \"", temp, "\" of the stressor \nvariable, the relative risk estimate and its standard error cannot be \ncalculated for stratum \"", stratum_levels[i], "\".  Also, the stratum \nwas removed from the analysis.\n")
              act <- paste0("The relative risk estimate and its standard error were not calculated for \nstratum \"", stratum_levels[i], "\".  Also, the stratum was removed from \nthe analysis.\n")
              warn_df <- rbind(warn_df, data.frame(
                func = I(fname),
                subpoptype = warn_vec[1], subpop = warn_vec[2],
                indicator = warn_vec[3], stratum = NA, warning = I(warn),
                action = I(act)
              ))
            } else if (total1 == 0 && total3 != 0) {
              rr <- 0
              rr_num <- 0
              rr_denom <- total3 / total4
              warn_ind <- TRUE
              warn <- paste0("Since there are no observations for the cell defined by level \"", response_levels[[ivar_r]][1], "\" \nof the response variable and level \"", stressor_levels[[ivar_s]][1], "\" of the stressor \nvariable, the relative risk estimate is zero and standard error of the relative \nrisk estimate cannot be calculated for stratum \"", stratum_levels[i], "\".  \nAlso, the stratum was removed from the analysis.\n")
              act <- paste0("Standard error of the relative risk estimate was not calculated for stratum \n\"", stratum_levels[i], "\".  Also, the stratum was removed from the \nanalysis.\n")
              warn_df <- rbind(warn_df, data.frame(
                func = I(fname),
                subpoptype = warn_vec[1], subpop = warn_vec[2],
                indicator = warn_vec[3], stratum = NA, warning = I(warn),
                action = I(act)
              ))
            } else if (total1 == 0 && total3 == 0) {
              rr <- NA
              rr_num <- total1 / total2
              rr_denom <- total3 / total4
              warn_ind <- TRUE
              warn <- paste0("Since there are no observations for the cell defined by level \"", response_levels[[ivar_r]][1], "\" \nof the response variable and level \"", stressor_levels[[ivar_s]][1], "\" of the stressor \nvariable and for the cell defined by level \"", response_levels[[ivar_r]][1], "\" of the \nresponse variable and level \"", stressor_levels[[ivar_s]][2], "\" of the stressor variable, \nthe relative risk estimate and its standard error cannot be calculated for \nstratum \"", stratum_levels[i], "\".  Also, the stratum was removed from \nthe analysis.\n")
              act <- paste0("The relative risk estimate and its standard error were not calculated for \nstratum \"", stratum_levels[i], "\".  Also, the stratum was removed from \nthe analysis.\n")
              warn_df <- rbind(warn_df, data.frame(
                func = I(fname),
                subpoptype = warn_vec[1], subpop = warn_vec[2],
                indicator = warn_vec[3], stratum = NA, warning = I(warn),
                action = I(act)
              ))
            } else if (total3 == 0) {
              rr <- NA
              rr_num <- total1 / total2
              rr_denom <- total3 / total4
              warn_ind <- TRUE
              warn <- paste0("Since there are no observations for the cell defined by level \"", response_levels[[ivar_r]][1], "\" \nof the response variable and level \"", stressor_levels[[ivar_s]][2], "\" of the stressor \nvariable, the relative risk estimate and its standard error cannot be \ncalculated for stratum \"", stratum_levels[i], "\".  Also, the stratum \nwas removed from the analysis.\n")
              act <- paste0("The relative risk estimate and its standard error were not calculated for \nstratum \"", stratum_levels[i], "\".  Also, the stratum was removed from \nthe analysis.\n")
              warn_df <- rbind(warn_df, data.frame(
                func = I(fname),
                subpoptype = warn_vec[1], subpop = warn_vec[2],
                indicator = warn_vec[3], stratum = NA, warning = I(warn),
                action = I(act)
              ))
            } else {
              rr <- (total1 * total4) / (total2 * total3)
              rr_num <- total1 / total2
              rr_denom <- total3 / total4
            }

            # Calculate the standard error estimate of the log of relative risk
            # for all strata combined

            if (any(c(total1, total2, total3, total4) == 0)) {
              rrlog_se <- NA
            } else {
              pder <- 1 / c(total1, -total2, -total3, total4)
              rrlog_se <- sqrt(t(pder) %*% varest %*% pder)
            }

            #
            # End the section for stratified data
            #
          } else {

            #
            # Begin the section for unstratified data
            #

            # Check whether the vector of response values contains a single
            # element

            if (nresp == 1) {
              stop("\nEstimates cannot be calculated since the vector of response values contains a \nsingle element.")
            }

            # Compute the 2x2 table of weight totals

            wgt_total <- addmargins(svytable(
              make.formula(paste(ivar_r, "+", ivar_s)),
              design = subset(design, tst)
            ))

            # Assign cell and marginal weight totals

            total1 <- wgt_total[
              response_levels[[ivar_r]][1],
              stressor_levels[[ivar_s]][1]
            ]
            total2 <- wgt_total["Sum", stressor_levels[[ivar_s]][1]]
            total3 <- wgt_total[
              response_levels[[ivar_r]][1],
              stressor_levels[[ivar_s]][2]
            ]
            total4 <- wgt_total["Sum", stressor_levels[[ivar_s]][2]]

            # Calculate the estimate of relative risk

            if (total2 == 0 || total4 == 0) {
              rr <- NA
              rr_num <- NA
              rr_denom <- NA
              warn_ind <- TRUE
              temp <- ifelse(total2 == 0, stressor_levels[[ivar_s]][1],
                stressor_levels[[ivar_s]][2]
              )
              warn <- paste0("Since there are no observations for level \"", temp, "\" of the stressor \nvariable, the relative risk estimate and its standard error cannot be \ncalculated.\n")
              act <- "The relative risk estimate and its standard error were not calculated.\n"
              warn_df <- rbind(warn_df, data.frame(
                func = I(fname),
                subpoptype = warn_vec[1], subpop = warn_vec[2],
                indicator = warn_vec[3], stratum = NA, warning = I(warn),
                action = I(act)
              ))
            } else if (total1 == 0 && total3 != 0) {
              rr <- 0
              rr_num <- 0
              rr_denom <- total3 / total4
              warn_ind <- TRUE
              warn <- paste0("Since there are no observations for the cell defined by level \"", response_levels[[ivar_r]][1], "\" \nof the response variable and level \"", stressor_levels[[ivar_s]][1], "\" of the stressor variable, \nthe relative risk estimate is zero and standard error of the relative risk \nestimate cannot be calculated.\n")
              act <- "Standard error of the relative risk estimate was not calculated.\n"
              warn_df <- rbind(warn_df, data.frame(
                func = I(fname),
                subpoptype = warn_vec[1], subpop = warn_vec[2],
                indicator = warn_vec[3], stratum = NA, warning = I(warn),
                action = I(act)
              ))
            } else if (total1 == 0 && total3 == 0) {
              rr <- NA
              rr_num <- total1 / total2
              rr_denom <- total3 / total4
              warn_ind <- TRUE
              warn <- paste0("Since there are no observations for the cell defined by level \"", response_levels[[ivar_r]][1], "\" \nof the response variable and level \"", stressor_levels[[ivar_s]][1], "\" of the stressor \nvariable and for the cell defined by level \"", response_levels[[ivar_r]][1], "\" of the \nresponse variable and level \"", stressor_levels[[ivar_s]][2], "\" of the stressor variable, \nthe relative risk estimate and its standard error cannot be calculated.\n")
              act <- "The relative risk estimate and its standard error were not calculated.\n"
              warn_df <- rbind(warn_df, data.frame(
                func = I(fname),
                subpoptype = warn_vec[1], subpop = warn_vec[2],
                indicator = warn_vec[3], stratum = NA, warning = I(warn),
                action = I(act)
              ))
            } else if (total3 == 0) {
              rr <- NA
              rr_num <- total1 / total2
              rr_denom <- total3 / total4
              warn_ind <- TRUE
              warn <- paste0("Since there are no observations for the cell defined by level \"", response_levels[[ivar_r]][1], "\" \nof the response variable and level \"", stressor_levels[[ivar_s]][2], "\" of the stressor \nvariable, the relative risk estimate and its standard error cannot be \ncalculated.\n")
              act <- "The relative risk estimate and its standard error were not calculated.\n"
              warn_df <- rbind(warn_df, data.frame(
                func = I(fname),
                subpoptype = warn_vec[1], subpop = warn_vec[2],
                indicator = warn_vec[3], stratum = NA, warning = I(warn),
                action = I(act)
              ))
            } else {
              rr <- (total1 * total4) / (total2 * total3)
              rr_num <- total1 / total2
              rr_denom <- total3 / total4
            }

            # Determine whether the standard error can be calculated

            if (any(c(total1, total2, total3, total4) == 0)) {
              rrlog_se <- NA
            } else {

              # Calculate the variance-covariance estimate for the cell and
              # marginal totals

              if (cluster_ind) {
                temp <- relrisk_var(
                  response[tst], stressor[tst],
                  response_levels[[ivar_r]], stressor_levels[[ivar_s]],
                  wgt2[tst], xcoord[tst], ycoord[tst], stratum_ind, NULL,
                  cluster_ind, clusterID[tst], wgt1[tst], xcoord1[tst],
                  ycoord1[tst], vartype, warn_ind, warn_df, warn_vec
                )
              } else {
                temp <- relrisk_var(response[tst], stressor[tst],
                  response_levels[[ivar_r]], stressor_levels[[ivar_s]],
                  wgt[tst], xcoord[tst], ycoord[tst], stratum_ind, NULL,
                  cluster_ind,
                  vartype = vartype, warn_ind = warn_ind,
                  warn_df = warn_df, warn_vec = warn_vec
                )
              }
              varest <- temp$varest
              warn_ind <- temp$warn_ind
              warn_df <- temp$warn_df

              # Calculate the standard error estimate of the log of relative
              # risk

              pder <- 1 / c(total1, -total2, -total3, total4)
              rrlog_se <- sqrt(t(pder) %*% varest %*% pder)
            }

            #
            # End section for unstratified data
            #
          }

          # Calculate confidence limits for the estimate of relative risk

          if (is.na(rrlog_se)) {
            cl <- NA
          } else {
            cl <- c(exp(log(rr) - rrlog_se * mult), exp(log(rr) + rrlog_se *
              mult))
          }

          # Calculate the table of cell and margin counts

          cc <- addmargins(table(list(
            response = response[tst],
            stressor = stressor[tst]
          )))

          # Calculate the table of cell and margin proportion estimates

          cp <- wgt_total / wgt_total["Sum", "Sum"]

          # Append results to the rrsum data frame

          rrsum <- rbind(rrsum, data.frame(
            Type = itype,
            Subpopulation = isubpop,
            Response = ivar_r,
            Stressor = ivar_s,
            nResp = cc[3, 3],
            Estimate = rr,
            Estimate_num = rr_num,
            Estimate_denom = rr_denom,
            StdError_log = rrlog_se,
            MarginofError_log = mult * rrlog_se,
            LCB = cl[1],
            UCB = cl[2],
            WeightTotal = wgt_total["Sum", "Sum"],
            CellCounts_11 = cc[
              response_levels[[ivar_r]][1],
              stressor_levels[[ivar_s]][1]
            ],
            CellCounts_12 = cc[
              response_levels[[ivar_r]][1],
              stressor_levels[[ivar_s]][2]
            ],
            CellCounts_21 = cc[
              response_levels[[ivar_r]][2],
              stressor_levels[[ivar_s]][1]
            ],
            CellCounts_22 = cc[
              response_levels[[ivar_r]][2],
              stressor_levels[[ivar_s]][2]
            ],
            CellProportions_11 = cp[
              response_levels[[ivar_r]][1],
              stressor_levels[[ivar_s]][1]
            ],
            CellProportions_12 = cp[
              response_levels[[ivar_r]][1],
              stressor_levels[[ivar_s]][2]
            ],
            CellProportions_21 = cp[
              response_levels[[ivar_r]][2],
              stressor_levels[[ivar_s]][1]
            ],
            CellProportions_22 = cp[
              response_levels[[ivar_r]][2],
              stressor_levels[[ivar_s]][2]
            ]
          ))

          # End of the loop for levels of the subpopulation
        }

        # End of the loop for stressor variables
      }

      # End of the loop for response variables
    }

    # End of the loop for subpopulations
  }

  # As necessary, output a message indicating that warning messages were
  # generated during execution of the program

  if (warn_ind) {
    warn_df <<- warn_df
    if (nrow(warn_df) == 1) {
      message("During execution of the program, a warning message was generated.  The warning \nmessage is stored in a data frame named 'warn_df'.  Enter the following command \nto view the warning message: warnprnt()\n")
    } else {
      message(paste("During execution of the program,", nrow(warn_df), "warning messages were generated.  The warning \nmessages are stored in a data frame named 'warn_df'.  Enter the following \ncommand to view the warning messages: warnprnt() \nTo view a subset of the warning messages (say, messages number 1, 3, and 5), \nenter the following command: warnprnt(m=c(1,3,5))\n"))
    }
  }

  # Assign dimension names to the rrsum data frame

  cell_lbl <- character(4)
  cell_lbl[1] <- "RespPoor_StressPoor"
  cell_lbl[2] <- "RespPoor_StressGood"
  cell_lbl[3] <- "RespGood_StressPoor"
  cell_lbl[4] <- "RespGood_StressGood"
  dimnames(rrsum) <- list(1:nrow(rrsum), c(
    "Type", "Subpopulation", "Response", "Stressor", "nResp", "Estimate",
    "Estimate_num", "Estimate_denom", "StdError_log", "MarginofError_log",
    paste0("LCB", conf, "Pct"), paste0("UCB", conf, "Pct"), "WeightTotal",
    paste0("Count_", cell_lbl[1]), paste0("Count_", cell_lbl[2]),
    paste0("Count_", cell_lbl[3]), paste0("Count_", cell_lbl[4]),
    paste0("Prop_", cell_lbl[1]), paste0("Prop_", cell_lbl[2]),
    paste0("Prop_", cell_lbl[3]), paste0("Prop_", cell_lbl[4])
  ))

  # Return the rrsum data frame

  rrsum
}