Difference between revisions of "Goherr: Fish consumption study"

  
library(OpasnetUtils)
library(ggplot2)
library(reshape2)

objects.latest("Op_en7749", "preprocess")

for(i in c(5:6, 16, 29:30, 46:49, 85:86, 95:98, 135)) {
  temp <- survey[!is.na(survey[[i]]),]
    p <- ggplot(temp, aes(x = temp[[i]])) + geom_bar() +
    theme_gray(base_size = 18) + theme(axis.text.x=element_text(angle=90,hjust=1,vjust=0.4)) +
    labs(title = colnames(temp[i])) + xlab("") + facet_wrap(~ Country)
  print(p)
}

temp <- melt(survey, measure.vars = 145:153, variable.name = "Changing.factor", value.name = "Impact")
levs <- c("-5 strongly disagree", "-4", "-3", "-2", "-1", "0 Neutral", "1", "2", "3", "4", "5 strongly agree", "I don't know")
temp$Impact <- factor(temp$Impact, levels = levs, labels = levs, ordered = TRUE)

ggplot(temp, aes(x = Impact, fill = Country)) + geom_histogram(position = "dodge") + facet_wrap(~ Changing.factor)+
  theme_gray(base_size = 24) + theme(axis.text.x=element_text(angle=90,hjust=1,vjust=0.4))

surveytemp <- subset(survey, survey[[31]] == "Yes")

temp <- melt(surveytemp, measure.vars = 38:43, variable.name = "Variable", value.name = "Value")
ggplot(temp, aes(x = Value, fill = Country)) + geom_histogram(position = "dodge") + facet_wrap(~ Variable)+
   theme_gray(base_size = 24) + labs(title = "Baltic salmon sources")

temp <- melt(surveytemp, measure.vars = 50:59, variable.name = "Variable", value.name = "Value")
ggplot(temp, aes(x = Value, fill = Country)) + geom_histogram(position = "dodge") + facet_wrap(~ Variable)+
   theme_gray(base_size = 24) + labs(title = "Reasons to eat Baltic salmon")

temp <- melt(surveytemp, measure.vars = 73:82, variable.name = "Variable", value.name = "Value")
ggplot(temp, aes(x = Value, fill = Country)) + geom_histogram(position = "dodge") + facet_wrap(~ Variable)+
   theme_gray(base_size = 24) + theme(axis.text.x=element_text(angle=90,hjust=1,vjust=0.4)) + labs(title = "Baltic salmon actions")

surveytemp <- subset(survey, survey[[31]] == "No")

temp <- melt(surveytemp, measure.vars = 62:70, variable.name = "Variable", value.name = "Value")
ggplot(temp, aes(x = Value, fill = Country)) + geom_histogram(position = "dodge") + facet_wrap(~ Variable)+
   theme_gray(base_size = 24) + labs(title = "Reasons not to eat Baltic salmon")


surveytemp <- subset(survey, survey[[86]] == "Yes")

temp <- melt(surveytemp, measure.vars = 87:92, variable.name = "Variable", value.name = "Value")
ggplot(temp, aes(x = Value, fill = Country)) + geom_histogram(position = "dodge") + facet_wrap(~ Variable)+
   theme_gray(base_size = 24) + labs(title = "Baltic herring sources")

temp <- melt(surveytemp, measure.vars = 99:108, variable.name = "Variable", value.name = "Value")
ggplot(temp, aes(x = Value, fill = Country)) + geom_histogram(position = "dodge") + facet_wrap(~ Variable)+
   theme_gray(base_size = 24) + labs(title = "Reasons to eat Baltic herring")

temp <- melt(surveytemp, measure.vars = 123:132, variable.name = "Variable", value.name = "Value")
ggplot(temp, aes(x = Value, fill = Country)) + geom_histogram(position = "dodge") + facet_wrap(~ Variable)+
   theme_gray(base_size = 24) + theme(axis.text.x=element_text(angle=90,hjust=1,vjust=0.4)) + labs(title = "Baltic herring actions")

surveytemp <- subset(survey, survey[[86]] == "No")

temp <- melt(surveytemp, measure.vars = 111:120, variable.name = "Variable", value.name = "Value")
ggplot(temp, aes(x = Value, fill = Country)) + geom_histogram(position = "dodge") + facet_wrap(~ Variable)+
   theme_gray(base_size = 24) + labs(title = "Reasons not to eat Baltic herring")

library(OpasnetUtils)

############## Generic functions and objects are defined first.

### webropol.convert converts a csv file from Webropol into a useful data.frame.

webropol.convert <- function(
  data, # Data.frame created from a Webropol csv file. The first row should contain headings.
  rowfact, # Row number where the factor levels start (in practice, last row + 3)
  textmark = "Other open" # The text that is shown in the heading if there is an open sub-question.
) {
  out <- dropall(data[2:(rowfact - 3) , ])
  subquestion <- t(data[1 , ])
  subquestion <- gsub("\xa0", " ", subquestion)
  subquestion <- gsub("\xb4", " ", subquestion)
  subquestion <- gsub("\n", " ", subquestion)
  #  subquestion <- gsub("\\(", " ", subquestion)
  #  subquestion <- gsub("\\)", " ", subquestion)
  textfield <- regexpr(textmark, subquestion) != -1
  subquestion <- strsplit(subquestion, ":") # Divide the heading into a main question and a subquestion.
  subqtest <- 0 # The previous question name.
  for(i in 1:ncol(out)) {
    #print(i)
    if(subquestion[[i]][1] != subqtest) { # If part of previous question, use previous fact.
      fact <- as.character(data[rowfact:nrow(data) , i]) # Create factor levels from the end of Webropol file.
      fact <- fact[fact != ""] # Remove empty rows
      fact <- gsub("\xa0", " ", fact)
      fact <- gsub("\xb4", " ", fact)
      fact <- gsub("\n", " ", fact)
      fact <- strsplit(fact, " = ") # Separate value (level) and interpretation (label)
    }
    if(length(fact) != 0 & !textfield[i]) { # Do this only if the column is not a text type column.
      out[[i]] <- factor(
        out[[i]], 
        levels = unlist(lapply(fact, function(x) x[1])), 
        labels = unlist(lapply(fact, function(x) x[2])), 
        ordered = TRUE
      )
    }
    subqtest <- subquestion[[i]][1]
  }
  return(out)
}

# merge.questions takes a multiple checkbox question and merges that into a single factor.
# First levels in levs have priority over others, if several levels apply to a row.

merge.questions <- function(
  dat, # data.frame with questionnaire data
  cols, # list of vectors of column names or numbers to be merged into one level in the factor
  levs, # vector (with the same length as cols) of levels of factors into which questions are merged.
  name # text string for the name of the new factor column in the data.
) {
  for(i in length(cols):1) {
    temp <- FALSE
    for(j in rev(cols[[i]])) {
      temp <- temp | !is.na(dat[[j]])
    }
    dat[[name]][temp] <- levs[i]
  }
  dat[[name]] <- factor(dat[[name]], levels = levs, ordered = TRUE)
  return(dat)
}

############# Data preprocessing

# Get the data either from Opasnet or your own hard drive.

#Survey original file: N:/Ymal/Projects/Goherr/WP5/Goherr_fish_consumption.csv

survey <- opasnet.csv("5/57/Goherr_fish_consumption.csv", sep = ";", fill = TRUE, quote = "\"")
surcol <- t(survey[1,])

survey <- webropol.convert(survey, 2121, textmark = ":Other open") # Data file is converted to data.frame using levels at row 1269.

# Take the relevant columnames from the table on the page.
colframe <- opbase.data("Op_en7749", subset = "Questions in the Goherr questionnaire")
colnames(survey) <- colframe$Result

survey$Row <- 1:nrow(survey)
survey$Weighting <- as.double(levels(survey$Weighting))[survey$Weighting]
survey$Ages <- ifelse(as.numeric(as.character(survey$Age)) < 46, "18-45",">45")

freqlist <- c(
  "less than once a year",
  "A few times a year",
  "1 - 3 times per month",
  "once a week",
  "2 - 4 times per week",
  "5 or more times per week"
)

amlist <- c(
  "1/6 plate or below (50 grams)",
  "1/3 plate (100 grams)",
  "1/2 plate (150 grams)",
  "2/3 plate (200 grams)",
  "5/6 plate (250 grams)",
  "full plate (300 grams)",
  "overly full plate (500 grams)"
#  "Not able to estimate"
)

sidel <- c(
  "1/6 plate or below (50 grams)",
  "1/4 plate (100 grams)",
  "1/2 plate (150 grams)",
  "2/3 plate (200 grams)",
  "5/6 plate (250 grams)"
#  "Not able to estimate"
)

ansl <- list(
  `How often eat fish` = freqlist,
  `How often Baltic salmon` = c("Never", freqlist),
  `How much Baltic salmon` = amlist,
  `How often side Baltic salmon` = c("Never", freqlist),
  `How much side Baltic salmon` = sidel,
  `How often Baltic herring` = c("Never", freqlist),
  `How much Baltic herring` = amlist,
  `How often side Baltic herring` = c("Never", freqlist),
  `How much side Baltic herring` = sidel
)

for (i in names(ansl)) {
  survey[[i]] <- factor(survey[[i]], levels = ansl[[i]])
}

objects.store(survey, surcol)
cat("objects survey, surcol were stored.\n")

# This is code Op_en7749/calculate_amount on page [[Goherr: Fish consumption study]]

library(OpasnetUtils)
library(ggplot2)
library(reshape2)
library(rjags)

objects.latest("Op_en7749", "preprocess")
assumptions <- opbase.data("Op_en7749.assumptions_for_calculations")

for (i in 46:49) {
  survey[[i]] <- as.character(survey[[i]])
}

survey[,46:49][is.na(survey[,46:49])] <- ""

for (i in 95:98) {
  survey[[i]] <- as.character(survey[[i]])
}

survey[,95:98][is.na(survey[,95:98])] <- ""

survey$Row <- 1:nrow(survey)
survey$Weighting <- as.double(levels(survey$Weighting))[survey$Weighting]
survey$Ages <- ifelse(as.numeric(as.character(survey$Age)) < 46, "18-45",">45")
short <- survey[c("Country", "Gender", "Row", "Ages")]

salmon_dishtimes <- Ovariable( 
  "salmon_dishtimes", 
  data = data.frame(
    short,
    Value = survey[[46]],
    Result = 1
  )
) #salmon with dish times / year

salmon_dishamount <- Ovariable(
  "salmon_dishamount", 
  data = data.frame(
    short,
    Value = survey[[47]],
    Result = 1
  )
) #salmon with dish amount per one serving

salmon_wholetimes <- Ovariable("salmon_wholetimes", data = data.frame(
  short,
  Value = survey[[48]],
  Result = 1
)) #salmon as such times / year

salmon_wholeamount <- Ovariable("salmon_wholeamount", data = data.frame(
  short,
  Value = survey[[49]],
  Result = 1
)) #salmon as whole amount per one serving

salmon_dishtimesassump <- Ovariable("salmon_dishtimesassump", data = assumptions[assumptions$Variable == "V46" , c("Value", "Result")])

salmon_dishamountassump <- Ovariable("salmon_dishamountassump", data = assumptions[assumptions$Variable == "V47" , c("Value", "Result")])

salmon_ingridient <- Ovariable("salmon_ingridient", data = assumptions[assumptions$Variable == "salmon_ingridient", ]["Result"])

salmon_wholetimesassump <- Ovariable("salmon_wholetimesassump", data = assumptions[assumptions$Variable == "V48" , c("Value", "Result")])

salmon_wholeamountassump <- Ovariable("salmon_wholeamountassump", data = assumptions[assumptions$Variable == "V49" , c("Value", "Result")])

amount <- Ovariable("amount", 
                    dependencies = data.frame(Name = c(
                      "salmon_dishamount",
                      "salmon_dishamountassump",
                      "salmon_ingridient",
                      "salmon_dishtimes",
                      "salmon_dishtimesassump",
                      "salmon_wholeamount",
                      "salmon_wholeamountassump",
                      "salmon_wholetimes",
                      "salmon_wholetimesassump"
                    )),
                    formula = function(...) {
                      
                      out1 <- (salmon_dishamount * salmon_dishamountassump * salmon_ingridient) 
                      out1@output$Value <- NULL
                      out1 <- ((salmon_dishtimes * salmon_dishtimesassump) * out1) / 365
                      out1@output$Value <- NULL
                      out2 <- (salmon_wholeamount * salmon_wholeamountassump)
                      out2@output$Value <- NULL
                      out2 <- ((salmon_wholetimes * salmon_wholetimesassump) * out2) / 365
                      out2@output$Value <- NULL
                      out <- out1 + out2
                      
                      # Per vuosi -> per d
                      
                      #result(out)[result(out) == 0] <- 0.01 # Ei jätetä nollia saantiin
                      return(out)
                    }
)
#amount <- EvalOutput(amount)
#### Draw N rows from the questionnaire subset filling the condition (in this case from the Country+Gender-group)

condition <- unique(survey[c("Country", "Gender")])

rows <- data.frame()
for(i in 1:nrow(condition)) {
  temp <- survey[
    survey$Gender == condition$Gender[i] & survey$Country == condition$Country[i] , 
    c("Gender", "Country", "Weighting")
    ] # Eikö tässä voisi yksinkertaistaa ja käyttää vain rivinumeroa eikä koko tempiä?
  temp <- temp[sample(1:nrow(temp), get("N", envir = openv), replace = TRUE, prob = temp$Weighting) , ]
  rows <- rbind(rows, data.frame(
    condition[i , ],
    Iter = 1:get("N", envir = openv),
    Row = as.character(floor(as.numeric(rownames(temp)))),
    Result = 1
  ))
}

# New version with same functionality
# However, the whole approach is inefficient. With N=1000 only ca 200 iterations per country and sex results from here.
# Instead, some probability distributions should be fitted based on the survey.
rows <- data.frame()
for(i in 1:nrow(condition)) {
  temp <- (1:nrow(survey))[survey$Gender == condition$Gender[i] & 
                             survey$Country == condition$Country[i]]
  temp <- sample(temp, openv$N, replace = TRUE, prob = survey$Weighting[temp])
  rows <- rbind(rows, data.frame(
    Iter = 1:openv$N,
    Row = temp
  ))
}


#rows <- Ovariable(output = rows, marginal = c(TRUE, TRUE, TRUE, TRUE, FALSE))

########## create ovariable for impact assessment model with the drawn individuals

#row <- rows@output
#row$Result <- NULL

salmon_dishamount@data <- merge(salmon_dishamount@data, rows) # One merge is enough to choose the rows wanted
#salmon_dishtimes@data <- merge(salmon_dishtimes@data, row)
#salmon_wholeamount@data <- merge(salmon_wholeamount@data, row)
#salmon_wholetimes@data <- merge(salmon_wholetimes@data, row)

amountSalmon <- EvalOutput(amount, forceEval = TRUE)

temp <- amountSalmon@output 
temp <- aggregate(temp$amountResult, by=list(temp$Country, temp$Ages, temp$Gender), FUN = mean)
colnames(temp) <- c("Country","Age","Gender","Salmon amount")
oprint(temp)

#################Bayesian approach

mod <- textConnection("
  model{
    for(j in 1:4) {
      for(i in 1:N) {
        sal[i , j] ~ dbin(p[j], 9)
      }
      p[j] ~ dunif(0,1)
    }
  }
")

# This row takes data and converts it to numbers. But what should be done is to take
# each variable, remove NA, order answers meaningfully and then convert to numbers.
# A binomial distribution is assumed for bins of answer choices.
salm <- as.data.frame(lapply(survey[46:49], FUN = function(x) as.numeric(as.factor(x))))
jags <- jags.model(mod,
  data = list(N = nrow(survey),
    sal = salm),
  n.chains = 4,
  n.adapt = 100
)

update(jags, 1000)

samps <- jags.samples(jags, 'p', 1000)

plot(samps[[1]])


#### Calculate herring amount

herring_dishtimes <- Ovariable("herring_dishtimes", data = data.frame(
  short,
  Value = survey[[95]],
  Result = 1
)) #herring with dish times / year

herring_dishamount <- Ovariable("herring_dishamount", data = data.frame(
  short,
  Value = survey[[96]],
  Result = 1
)) #herring with dish amount per one serving

herring_wholetimes <- Ovariable("herring_wholetimes", data = data.frame(
  short,
  Value = survey[[97]],
  Result = 1
)) #herring as such times / year

herring_wholeamount <- Ovariable("herring_wholeamount", data = data.frame(
  short,
  Value = survey[[98]],
  Result = 1
)) #herring as whole amount per one serving

herring_dishtimesassump <- Ovariable("herring_dishtimesassump", data = assumptions[assumptions$Variable == "V95" , c("Value", "Result")])

herring_dishamountassump <- Ovariable("herring_dishamountassump", data = assumptions[assumptions$Variable == "V96" , c("Value", "Result")])

herring_ingridient <- Ovariable("herring_ingridient", data = assumptions[assumptions$Variable == "herring_ingridient", ]["Result"])

herring_wholetimesassump <- Ovariable("herring_wholetimesassump", data = assumptions[assumptions$Variable == "V97" , c("Value", "Result")])

herring_wholeamountassump <- Ovariable("herring_wholeamountassump", data = assumptions[assumptions$Variable == "V98" , c("Value", "Result")])

amount <- Ovariable("amount", 
                    dependencies = data.frame(Name = c(
                      "herring_dishamount",
                      "herring_dishamountassump",
                      "herring_ingridient",
                      "herring_dishtimes",
                      "herring_dishtimesassump",
                      "herring_wholeamount",
                      "herring_wholeamountassump",
                      "herring_wholetimes",
                      "herring_wholetimesassump"
                    )),
                    formula = function(...) {
                      
                      out1 <- (herring_dishamount * herring_dishamountassump * herring_ingridient) 
                      out1@output$Value <- NULL
                      out1 <- ((herring_dishtimes * herring_dishtimesassump) * out1) / 365
                      out1@output$Value <- NULL
                      out2 <- (herring_wholeamount * herring_wholeamountassump)
                      out2@output$Value <- NULL
                      out2 <- ((herring_wholetimes * herring_wholetimesassump) * out2) / 365
                      out2@output$Value <- NULL
                      out <- out1 + out2
                      
                      # Per vuosi -> per d
                      
                      return(out)
                    }
)


#### Draw N rows from the questionnaire subset filling the condition (in this case from the Country+Gender-group)

rows <- data.frame()
for(i in 1:nrow(condition)) {
  temp <- survey[
    survey$Gender == condition$Gender[i] & survey$Country == condition$Country[i] , 
    c("Gender", "Country", "Weighting")
    ] # Eikö tässä voisi yksinkertaistaa ja käyttää vain rivinumeroa eikä koko tempiä?
  temp <- temp[sample(1:nrow(temp), get("N", envir = openv), replace = TRUE, prob = temp$Weighting) , ]
  rows <- rbind(rows, data.frame(
    condition[i , ],
    Iter = 1:get("N", envir = openv),
    Row = as.character(floor(as.numeric(rownames(temp)))),
    Result = 1
  ))
}
rows <- Ovariable(output = rows, marginal = c(TRUE, TRUE, TRUE, TRUE, FALSE))

########## create ovariable for impact assessment model with the drawn individuals

row <- rows@output
row$Result <- NULL

herring_dishamount@data <- merge(herring_dishamount@data, row)
herring_dishtimes@data <- merge(herring_dishtimes@data, row)
herring_wholeamount@data <- merge(herring_wholeamount@data, row)
herring_wholetimes@data <- merge(herring_wholetimes@data, row)

amountHerring <- EvalOutput(amount, forceEval = TRUE)

temp <- amountHerring@output
temp <- aggregate(temp$amountResult, by=list(temp$Country,temp$Ages,temp$Gender), FUN = mean)
colnames(temp) <- c("Country","Age","Gender","Herring amount")
oprint(temp)

objects.store(amountHerring, amountSalmon)
cat("objects amountHerring, amountSalmon were stored.\n")