geom_smooth won't work with facet_wrap when factors differ?

Question

This is my first time posting, so I apologize for any weird formatting issues, especially with my R code.

First, I am trying to get 95% confidence interval curves around the two curves in each panel of the figure reproduced with the code below. It seems like when I have ggplot(look, aes(x=treatment, y=fit, color=generation)) and facet_wrap(~population), I can't get the 95% confidence intervals to work with geom_smooth (not included in the code below because it doesn't do anything). geom_smooth only works when I have ggplot(look, aes(x=treatment, y=fit, color=population)) and facet_wrap(~population).

Please let me know if anything needs clarification!

Here's the visual output I get. I am trying to get geom_smooth to add confidence intervals to each curve.

These are the types of intervals I'm trying to add:

library(ggplot2)

germData <- read.csv("https://dl.dropbox.com/s/h75rzzavpoxb8q1/processed_germData%20-%20no%20GA%20-%20continuous.csv")

totalViable    <- germData$totalGoodSeeds   
totalGerm      <- germData$totalGerm        

y <- cbind(totalGerm, totalViable-totalGerm)

germData$germRate <- germData$totalGerm/germData$totalGoodSeeds

q1  <- glm(y ~ population + generation + treatment + population:treatment + population:generation + generation:treatment + population:generation:treatment, germData, family=quasibinomial)

xv <- seq(0, 20, .2)
pfl <- length(xv)

look <- data.frame(population = factor(rep(levels(germData$population), each = 2*pfl)), generation = factor(rep(levels(germData$generation), each = pfl)), treatment = xv)

pred <- predict(q1, newdata = look, type= "response", se.fit = TRUE)
look$fit <- pred$fit
look$se.fit <- pred$se.fit

GAintercepts <- read.csv("https://dl.dropbox.com/s/jeei9yst7h4l37q/predict_germData.csv")

GAinterceptpoints <- GAintercepts[GAintercepts$treatment=="GA3",]
GAinterceptpoints$intercept <- "0" 
GAinterceptpoints$intercept <- as.numeric(GAinterceptpoints$intercept)

bigPlot <- ggplot(look, aes(x=treatment, y=fit, color=generation)) + 
    geom_line() + 
    geom_jitter(width=0.02, inherit.aes = FALSE, data=germData, 
    aes(x=treatment, y=germRate, color=generation), show.legend = FALSE) 

bigPlotPlus <- bigPlot + 
    geom_smooth() + 
    scale_color_manual(name="generation", values=c("#0f71bc", "#a3e1ff")) + 
    geom_pointrange(inherit.aes=FALSE, data=GAinterceptpoints, 
    aes(x=intercept, y=fit.germRate, ymin=fit.germRate-se, 
    ymax=fit.germRate+se, color=generation)) + 
    scale_color_manual(name="generation", values=c("#0f71bc", "#a3e1ff")) + 
    facet_wrap(~population) 

bigPlotPlus

Answer 1

Without seeing a sketch of your desired output, I may be misunderstanding what you're going for. If these don't match it, then please sketch what you're going for. I think the problem is that you need to explicitly define the group parameter to be a unique combination of population and generation.

library(magrittr)
library(ggplot2)

germDataOriginal <- read.csv("https://dl.dropbox.com/s/h75rzzavpoxb8q1/processed_germData%20-%20no%20GA%20-%20continuous.csv")

germData <- germDataOriginal %>% 
  dplyr::select(
    population, 
    generation, 
    treatment,
    totalGoodSeeds,  
    totalGerm,       
    totalColdGerm   
  ) %>% 
  dplyr::mutate(
    difference  = totalViable - totalGerm,
    germRate = totalGerm / totalGoodSeeds
  )

q1 <- glm(
  cbind(germData$totalGerm, germData$difference) 
  ~ 
  population + generation + treatment + population*generation*treatment, germData, family=quasibinomial
)

look <- tidyr::crossing(
  population  = germData$population, 
  generation  = germData$generation,
  treatment   = seq(0, 20, by=5)
) %>% 
dplyr::mutate(
  pop_gen    = paste0(population, "-", generation)
)

pred <- predict(q1, newdata = look, type= "response", se.fit = TRUE)
look$fit <- pred$fit
look$se.fit <- pred$se.fit

GAinterceptpoints <- read.csv("https://dl.dropbox.com/s/jeei9yst7h4l37q/predict_germData.csv") %>% 
  dplyr::filter(treatment=="GA3") %>% 
  dplyr::mutate(
    intercept = 0
  )

bigPlot <- ggplot(look, aes(x=treatment, y=fit, color=generation)) + 
  geom_line(aes(group=pop_gen)) +
  geom_point(aes(x=treatment, y=germRate, colour=generation), data=germData, inherit.aes = FALSE, shape=1)
bigPlot

bigPlot + 
  geom_pointrange(
    aes(x=intercept,y=fit.germRate, ymin=fit.germRate-se, ymax=fit.germRate+se, color=generation),
    inherit.aes=FALSE, data=GAinterceptpoints
  ) 

last_plot() + 
  facet_wrap(~population) 

For reference, your original bigPlot looks like

Notice I removed more cosmetics. I liked what you had better, but they're likely distracting from your fundamental question. Also, I'd separate that question about color scales, because it appears mostly unrelated. Also, in the future, simplify the incoming datasets to just the minimal amount of rows and columns. In order to address your fundamental question, there's no real need to calculate the germRate in the question's code; it's distracting.