ggplot2: multiple variables on x-axis at multiple times - r

Question

I have a data frame for observation numbers (3 observations for same id), height, weight and fev that looks like this (just for example):
id obs height weight fev
1 1 160 80 90
1 2 150 70 85
1 3 155 76 87
2 1 140 67 91
2 2 189 78 71
2 3 178 86 89
I need to plot this data using ggplot2 such that on x-axis there are 3 variables height, weight, fev; and the observation numbers are displayed as 3 vertical lines for each variable (color coded), where each lines show a median as a solid circle, and 25th and 75th percentiles as caps at the upper and lower extremes of the line (no minimum or maximum needed). I have so far tried many variations of box plots but I am not even getting close. Any suggestion(s) how to approach or solve this?
Thanks

Answer 1

OK instead what I did below was make three graphs then piece together with gridExtra. Read more about package here: http://www.sthda.com/english/wiki/wiki.php?id_contents=7930
I took the common legend code from this site to produce the following, starting with our existing longdf2. By piecing together the graphs, the information about corresponding observation is within the title of the graph
id <- rep(1:12, each = 3)
obs <- rep(1:3, 12)
height <- seq(140,189, length.out = 36)
weight <- seq(67,86, length.out = 36)
fev <- seq(71,91, length.out = 36)
df <- as.data.frame(cbind(id,obs,height, weight, fev))
obsonly <- melt(df, id.vars = c('id'), measure.vars = 'obs')
obsonly <- rbind(obsonly,obsonly,obsonly)
newvars <- melt(df[-2],id.vars = 'id')
longdf2 <- cbind(obsonly,newvars)
longdf2 <- longdf2[-4] #dropping second id column
colnames(longdf2)[c(2:5)] <- c('obs', 'obsnum', 'variable', 'value')
#Make graph 1 of observation 1
g1 <- longdf2 %>%
dplyr::filter(obsnum == 1) %>%
ggplot(aes(x = variable, y = value, color = variable)) +
stat_summary(fun.data=median_hilow) +
labs(title = "Observation 1") +
theme(plot.title = element_text(hjust = 0.5)) #has a legend
g2 <- longdf2 %>%
dplyr::filter(obsnum == 2) %>%
ggplot(aes(x = variable, y = value, color = variable)) +
stat_summary(fun.data=median_hilow) +
labs(title = "Observation 2") +
theme(plot.title = element_text(hjust = 0.5), legend.position =
'none')
#specified as none to make common legend at end
g3 <- longdf2 %>%
dplyr::filter(obsnum == 3) %>%
ggplot(aes(x = variable, y = value, color = variable)) +
stat_summary(fun.data=median_hilow) +
labs(title = "Observation 3") +
theme(plot.title = element_text(hjust = 0.5), legend.position =
'none')
library(gridExtra)
get_legend<-function(myggplot){
tmp <- ggplot_gtable(ggplot_build(myggplot))
leg <- which(sapply(tmp$grobs, function(x) x$name) == "guide-box")
legend <- tmp$grobs[[leg]]
return(legend)
}
# Save legend
legend <- get_legend(g1)
# Remove legend from 1st graph
g1 <- g1 + theme(legend.position = 'none')
# Combine graphs
grid.arrange(g1, g2, g3, legend, ncol=4, widths=c(2.3, 2.3, 2.3, 0.8))
Plenty of other little tweaks you could make along the way

Answer 2

Try putting the data into long format prior to graphing. I generated some more data, 12 subjects, each with 3 observations.
id <- rep(1:12, each = 3)
obs <- rep(1:3, 12)
height <- seq(140,189, length.out = 36)
weight <- seq(67,86, length.out = 36)
fev <- seq(71,91, length.out = 36)
df <- as.data.frame(cbind(id,obs,height, weight, fev))
library(reshape2) #use to melt data from wide to long format
longdf <- melt(df,id.vars = c('id', 'obs'))
Don't need to define measure variables here since the id.vars are defined, the remaining non-id.vars automatically default to measure variables. If you have more variables in your data set, you'll want to define measure variables in that same line as: measure.vars = c("height,"weight","fev")
longdf <- melt(df,id.vars = c('id', 'obs'), measure.vars = c("height", "weight", "fev"))
Apologies, haven't earned enough votes to put figures into my responses
ggplot(data = longdf, aes(x = variable, y = value, fill = factor(obs))) +
geom_boxplot(notch = T, notchwidth = .25, width = .25, position = position_dodge(.5))
This does not produce the exact graph you described-- which sounded like it was geom_linerange or something similar? -- those geoms require an x, ymin, and ymax to draw. Otherwise a regular, 'ole boxplot has your 1st and 3rd IQRs and median marked. I adjusted parameters of the boxplot to make it thinner with notches and widths, and separated them slightly with the position_dodge(.5)
after reading your response, I edited my original answer
You could try facet_wrap -- and watch the exchanging of "fill" vs. "color" in ggplot. If an object can't be "filled" with a color, like a boxplot or distribution, then it has to be "colored" with a color. Use color instead in the original aes()
ggplot(data = longdf, aes(x = variable, y = value, color = factor(obs))) +
stat_summary(fun.data=median_hilow) + facet_wrap(.~obs)
This gives you observation 1 - height, weight, fev side by side, observation 2- height, ....
If that still isn't what you want perhaps more like height observation 1,2,3; weight observation 1,2,3...then you'll need to modify your melting to have two variable and two value columns. Essentially make two melted dataframes, then cbind. Annnnd because each observation has three variables, you'll need to rbind to make sure both data frames have the same number of rows:
obsonly <- melt(df, id.vars = c('id'), measure.vars = 'obs')
obsonly <- rbind(obsonly,obsonly,obsonly) #making rows equal
longvars <- melt(df[-2],id.vars = 'id') #dropping obs from melt
longdf2 <- cbind(obsonly,longvars)
longdf2 <- longdf2[-4] #dropping second id column
colnames(longdf2)[c(2:5)] <- c('obs', 'obsnum', 'variable', 'value')
ggplot(data = longdf2, aes(x = obsnum, y = value,
color = factor(variable))) +
stat_summary(fun.data=median_hilow) +
facet_wrap(.~variable)
From here you can play around with the x axis marks (probably isn't useful to have a 1.5 observation marked) and the spacing of the lines from each other