R 贝叶斯有序逻辑-试图根据模式个体绘制预测y随时间的变化

我有一个数据集，它结合了多年来不同国家的几项调查。根据调查对象，我的因变量（lrparty）是一个政党的意识形态立场（范围从0到10）。我有几个自变量，比如年龄、性别、教育程度、党派偏见和受访者的收入

然后，对于每个政党和每个调查，我想根据模式个体（例如，年龄=31，女性=1，教育=2，收入=2，党派=1）绘制一段时间内lrparty的预测值。因此，图表看起来像：x轴=年；y轴=根据模态个体的LRP预测值

总而言之，以下是我尝试做的几个阶段： 1.评估模型： 该党（lrparty）在性别、年龄、教育程度、收入和受调查者的党派倾向上的有序逻辑回归

采取后抽签

预测模式响应方的位置（例如，500次抽签）

然后，我希望有一个数据集，它应该如下所示： 年份、调查、国家、缔约方（cmp代码），%缺少安置，x1:x500（来自抽签）

从这个数据集，我将生成我的绘图。以英国为例，根据CSE的调查

为了弄清楚代码，我开始只使用一项调查（cses）、一个国家（英国）和一个政党（保守党），正如您在下面代码中所看到的那样。但是我不知道如何从我在代码中的位置到我想要的情节（如上所述）

库（rstan）
图书馆（tidyverse）
图书馆（brms）
图书馆（GG2）
图书馆（主题）
图书馆（ggmcmc）
##数据：
加载（“pbrands.RData”）
##只保留国家=英国；调查=CSE；政党=保守派
uk_cses_con=pbrands%>%
选择（政党、女性、年龄、教育程度、收入、党派、年份、调查、，
国家，cmp，政党名称（简称，政党名称（英文，lrs）%>%
筛选（调查==“CSE”）%>%
过滤器（国家==“英国”）%>%
过滤器（cmp==51620）
##进行贝叶斯有序logit模型
适合正常。经过几次尝试和错误尝试，我找到了代码。由于其他人可能对此感兴趣，我将在下面发布代码

    ## Packages
    install.packages(c("bmrs", "coda", "mvtnorm", "devtools"))
    library(devtools)
    library(tidyverse)
    library(brms)

    ## Loading the data
    load('~/Data/mydata.RData')

    ## Keeping the variables of our interest
    mydata = mydata %>% 
    select(lrparty, female, age, education, income, partisan, year, survey, 
     country, cmp, party_name_short, party_name_english, lrs) 

    ## Function for calculating modes
    getmode <- function(v) {
    uniqv = unique(v)
    uniqv[which.max(tabulate(match(v, uniqv)))]
    }

    ## Finding Modal respondents by country, survey, and party:

    ## Modes by country 
    mode_by_country = mydata %>% 
    group_by(country) %>% 
    mutate(modal_age = getmode(na.omit(age))) %>% 
    mutate(modal_inc = getmode(na.omit(income))) %>% 
    mutate(modal_female = getmode(na.omit(female))) %>% 
    mutate(modal_edu = getmode(na.omit(education))) %>% 
    mutate(modal_partisan = getmode(na.omit(partisan))) %>% 
    filter(!duplicated(country))

    mode_by_country = mode_by_country[ , c(9, 14:18)]

    mode_by_country = mode_by_country[-40, ]

    ## Build object to receive the information we want to store
    runner <- c()
    pred = matrix(NA, 2000, 11)
    yhat = matrix(NA, 2000, 1)

    ###### Conducting the model for UK with two parties #########
    uk = mydata %>% 
    select(lrparty, female, age, education, income, partisan, year, survey,               
    country, cmp, party_name_short, party_name_english, lrs) %>% 
    filter(survey == "cses") %>% 
    filter(country == "uk") %>% 
    filter(cmp == 51320 | cmp == 51620)

    ## Finding how many regressions will be conducted
    reglength <- length(unique(uk$survey)) * length(unique(uk$year)) *  length(unique(uk$cmp))

    ## Creating our modal British individual based on mode_by_country
    mode_by_country[mode_by_country$country == "uk", c(2:6)]

    newavg <- data.frame(age = 35, income = 2, female = 1, education = 2,  partisan = 0)

    ## Loop to conduct the ordered logit in rstan, using iter=1000, and           chains=4

    for(p in na.omit(unique(uk$cmp))){

    hold <- uk[uk$cmp == p, ]
    country <- hold$country[1]

    for(s in na.omit(unique(hold$survey))){

        hold1 <- hold[hold$survey == s, ]

        for(y in na.omit(unique(hold1$year))){

            mod <- brm(lrparty ~ age + female + education + income + partisan, data = hold1[hold1$year == y, ], family = "cumulative", chains = 4, iter = 1000)

            for(i in 1:2000) {
              pred[i,] <- predict(mod, newdata = newavg, probs = c(0.025, 0.975), summary=TRUE) 
              yhat[i] <- sum(pred[i, ] * c(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11))
            }

              newData <- data.frame(country, p, s, y, pred, yhat)
              newData$m <- mean(newData$yhat)
              newData$sd <- sd(newData$yhat)
              newData$lower <- newData$m - 1.96*newData$sd 
              newData$upper <- newData$m + 1.96*newData$sd   

            runner <- rbind(runner, newData)
        }
      }
   }

    ## Keeping unique values within dataset
    uniqdata = runner %>% 
    filter(!duplicated(m))

    ## Creating the Figure
    uniqdata2 <- uniqdata[, c(1:4, 17:20)]

    uniqdata3 <- uniqdata2 %>% 
    gather(variable, value, -(y:p)) %>%
    unite(temp, p, variable) %>%
    spread(temp, value)

    uniqdata3 = uniqdata3[ , -c(3,6,8,11)]

    names(uniqdata3)[3:8] = c("lower_lab", "m_lab", "upper_lab", "lower_con", "m_con", "upper_con")

    uniqdata3[3:8] = as.numeric(unlist(uniqdata3[3:8]))

    ## Plot: Predicted Party Ideological Placement for Modal British Respondent

    ggplot(uniqdata3, aes(x = (y))) + geom_line(aes(y = m_lab, colour = "Labor")) + geom_ribbon(aes(ymin = lower_lab,ymax = upper_lab,
              linetype=NA), alpha = .25) +
    geom_line(aes(y = m_con, color = "Conservatives")) +
    geom_ribbon(aes(ymin = lower_con,
              ymax = upper_con,
              linetype=NA), alpha = .25) +
    theme_bw() + 
    theme(legend.position = "bottom", plot.title = element_text(hjust = 0.5)) + labs(title = "Predicted Party Ideological Placement for Modal British Respondent \n Survey = CSES") + theme(plot.title = element_text(color="black", size=20, face="bold.italic"), axis.title.x = element_text(color="black", size=15, face="italic"), axis.title.y = element_text(color="black", size=15, face="italic")) + 
    theme(legend.title = element_blank()) +
    theme(axis.text.x = element_text(color="black", size= 12.5), axis.text.y = element_text(color="black", size=12.5)) + theme(legend.text = element_text(size=15)) + scale_x_continuous(name="Year", breaks=seq(1997, 2005, 2)) + scale_y_continuous(name="Left-Right Party Position", limits=c(0, 10))

##软件包
安装包（c（“BMR”、“coda”、“mvtnorm”、“devtools”））
图书馆（devtools）
图书馆（tidyverse）
图书馆（brms）
##加载数据
加载（“~/Data/mydata.RData”）
##保留我们感兴趣的变量
mydata=mydata%>%
选择（政党、女性、年龄、教育程度、收入、党派、年份、调查、，
国家、cmp、缔约方名称（简称、缔约方名称、英语、lrs）
##计算模式的函数
getmode%
按（国家）划分的组别%>%
变异（模式年龄=获取模式（不适用，省略年龄））%>%
突变（模式inc=getmode（不适用于省略（收入）））%>%
突变（模态_雌性=getmode（na.省略（雌性）））%>%
变异（模态_edu=getmode（na.省略（教育）））%>%
mutate（modal_partisan=getmode（na.omit（partisan）））%>%
筛选器（！重复（国家/地区））
按国家划分的模式=按国家划分的模式[，c（9，14:18）]
按国家划分的模式=按国家划分的模式[-40，]
##构建对象以接收我们要存储的信息
跑步者%
选择（政党、女性、年龄、教育程度、收入、党派、年份、调查、，
国家，cmp，政党名称（简称，政党名称（英文，lrs）%>%
筛选（调查==“CSE”）%>%
过滤器（国家==“英国”）%>%
过滤器（cmp==51320 | cmp==51620）
##找出将进行多少次回归
reglength
    ## Packages
    install.packages(c("bmrs", "coda", "mvtnorm", "devtools"))
    library(devtools)
    library(tidyverse)
    library(brms)

    ## Loading the data
    load('~/Data/mydata.RData')

    ## Keeping the variables of our interest
    mydata = mydata %>% 
    select(lrparty, female, age, education, income, partisan, year, survey, 
     country, cmp, party_name_short, party_name_english, lrs) 

    ## Function for calculating modes
    getmode <- function(v) {
    uniqv = unique(v)
    uniqv[which.max(tabulate(match(v, uniqv)))]
    }

    ## Finding Modal respondents by country, survey, and party:

    ## Modes by country 
    mode_by_country = mydata %>% 
    group_by(country) %>% 
    mutate(modal_age = getmode(na.omit(age))) %>% 
    mutate(modal_inc = getmode(na.omit(income))) %>% 
    mutate(modal_female = getmode(na.omit(female))) %>% 
    mutate(modal_edu = getmode(na.omit(education))) %>% 
    mutate(modal_partisan = getmode(na.omit(partisan))) %>% 
    filter(!duplicated(country))

    mode_by_country = mode_by_country[ , c(9, 14:18)]

    mode_by_country = mode_by_country[-40, ]

    ## Build object to receive the information we want to store
    runner <- c()
    pred = matrix(NA, 2000, 11)
    yhat = matrix(NA, 2000, 1)

    ###### Conducting the model for UK with two parties #########
    uk = mydata %>% 
    select(lrparty, female, age, education, income, partisan, year, survey,               
    country, cmp, party_name_short, party_name_english, lrs) %>% 
    filter(survey == "cses") %>% 
    filter(country == "uk") %>% 
    filter(cmp == 51320 | cmp == 51620)

    ## Finding how many regressions will be conducted
    reglength <- length(unique(uk$survey)) * length(unique(uk$year)) *  length(unique(uk$cmp))

    ## Creating our modal British individual based on mode_by_country
    mode_by_country[mode_by_country$country == "uk", c(2:6)]

    newavg <- data.frame(age = 35, income = 2, female = 1, education = 2,  partisan = 0)

    ## Loop to conduct the ordered logit in rstan, using iter=1000, and           chains=4

    for(p in na.omit(unique(uk$cmp))){

    hold <- uk[uk$cmp == p, ]
    country <- hold$country[1]

    for(s in na.omit(unique(hold$survey))){

        hold1 <- hold[hold$survey == s, ]

        for(y in na.omit(unique(hold1$year))){

            mod <- brm(lrparty ~ age + female + education + income + partisan, data = hold1[hold1$year == y, ], family = "cumulative", chains = 4, iter = 1000)

            for(i in 1:2000) {
              pred[i,] <- predict(mod, newdata = newavg, probs = c(0.025, 0.975), summary=TRUE) 
              yhat[i] <- sum(pred[i, ] * c(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11))
            }

              newData <- data.frame(country, p, s, y, pred, yhat)
              newData$m <- mean(newData$yhat)
              newData$sd <- sd(newData$yhat)
              newData$lower <- newData$m - 1.96*newData$sd 
              newData$upper <- newData$m + 1.96*newData$sd   

            runner <- rbind(runner, newData)
        }
      }
   }

    ## Keeping unique values within dataset
    uniqdata = runner %>% 
    filter(!duplicated(m))

    ## Creating the Figure
    uniqdata2 <- uniqdata[, c(1:4, 17:20)]

    uniqdata3 <- uniqdata2 %>% 
    gather(variable, value, -(y:p)) %>%
    unite(temp, p, variable) %>%
    spread(temp, value)

    uniqdata3 = uniqdata3[ , -c(3,6,8,11)]

    names(uniqdata3)[3:8] = c("lower_lab", "m_lab", "upper_lab", "lower_con", "m_con", "upper_con")

    uniqdata3[3:8] = as.numeric(unlist(uniqdata3[3:8]))

    ## Plot: Predicted Party Ideological Placement for Modal British Respondent

    ggplot(uniqdata3, aes(x = (y))) + geom_line(aes(y = m_lab, colour = "Labor")) + geom_ribbon(aes(ymin = lower_lab,ymax = upper_lab,
              linetype=NA), alpha = .25) +
    geom_line(aes(y = m_con, color = "Conservatives")) +
    geom_ribbon(aes(ymin = lower_con,
              ymax = upper_con,
              linetype=NA), alpha = .25) +
    theme_bw() + 
    theme(legend.position = "bottom", plot.title = element_text(hjust = 0.5)) + labs(title = "Predicted Party Ideological Placement for Modal British Respondent \n Survey = CSES") + theme(plot.title = element_text(color="black", size=20, face="bold.italic"), axis.title.x = element_text(color="black", size=15, face="italic"), axis.title.y = element_text(color="black", size=15, face="italic")) + 
    theme(legend.title = element_blank()) +
    theme(axis.text.x = element_text(color="black", size= 12.5), axis.text.y = element_text(color="black", size=12.5)) + theme(legend.text = element_text(size=15)) + scale_x_continuous(name="Year", breaks=seq(1997, 2005, 2)) + scale_y_continuous(name="Left-Right Party Position", limits=c(0, 10))