R: ggplot2: plot initiating at the axes - r

Question

I'm trying to respond to a reviewer that wants some changes to a figure... I am using ggplot2 to generate Kaplan-Meier curves, and the reviewer wants the X-axis to start at 0. The default in ggkmTable adds some space between 0 and the Y-axis. I can't figure out what to change to make it look right.
Here is my code:
ggkmTable <- function(sfit, table=TRUE,returns = FALSE,
xlabs = "Time in Years", ylabs = "Survival Probability",
ystratalabs = NULL, ystrataname = NULL,
timeby = 100, main = "Kaplan-Meier Plot",
pval = TRUE, ...) {
require(plyr)
require(ggplot2)
require(survival)
require(gridExtra)
if(is.null(ystratalabs)) {
ystratalabs <- as.character(levels(summary(sfit)$strata))
}
m <- max(nchar(ystratalabs))
if(is.null(ystrataname)) ystrataname <- "Strata"
times <- seq(0, max(sfit$time), by = timeby)
.df <- data.frame(time = sfit$time, n.risk = sfit$n.risk,
n.event = sfit$n.event, surv = sfit$surv, strata = summary(sfit, censored = T)$strata,
upper = sfit$upper, lower = sfit$lower)
levels(.df$strata) <- ystratalabs
zeros <- data.frame(time = 0, surv = 1, strata = factor(ystratalabs, levels=levels(.df$strata)),
upper = 1, lower = 1)
.df <- rbind.fill(zeros, .df)
d <- length(levels(.df$strata))
p <- ggplot(.df, aes(time, surv, group = strata)) +
geom_step(aes(linetype = strata), size = 0.7) +
theme_bw() +
theme(axis.title.x = element_text(vjust = 0.5)) +
scale_x_continuous(xlabs, breaks = times, limits = c(0, max(sfit$time))) +
scale_y_continuous(ylabs, limits = c(0, 1)) +
theme(panel.grid.minor = element_blank()) +
theme(legend.position = "bottom") +
theme(legend.key = element_rect(colour = NA)) +
labs(linetype = ystrataname) +
theme(plot.margin = unit(c(0, 1, .5, ifelse(m < 10, 1.5, 2.5)), "lines")) +
ggtitle(main)
if(pval) {
sdiff <- survdiff(eval(sfit$call$formula), data = eval(sfit$call$data))
pval <- pchisq(sdiff$chisq, length(sdiff$n)-1, lower.tail = FALSE)
pvaltxt <- paste("p =", signif(pval, 3))
p <- p + annotate("text", x = 0.6 * max(sfit$time), y = 0.1, label = pvaltxt)
}
## Create a blank plot for place-holding
## .df <- data.frame()
blank.pic <- ggplot(.df, aes(time, surv)) +
geom_blank() +
theme_bw() +
theme(axis.text.x = element_blank(), axis.text.y = element_blank(),
axis.title.x = element_blank(), axis.title.y = element_blank(),
axis.ticks = element_blank(), panel.grid.major = element_blank(),
panel.border = element_blank())
if(table) {
## Create table graphic to include at-risk numbers
risk.data <- data.frame(strata = summary(sfit, times = times, extend = TRUE)$strata,
time = summary(sfit, times = times, extend = TRUE)$time,
n.risk = summary(sfit, times = times, extend = TRUE)$n.risk)
data.table <- ggplot(risk.data, aes(x = time, y = strata, label = format(n.risk, nsmall = 0))) +
#, color = strata)) +
geom_text(size = 3.5) +
theme_bw() +
scale_y_discrete(breaks = as.character(levels(risk.data$strata)), labels = ystratalabs) +
# scale_y_discrete(#format1ter = abbreviate,
# breaks = 1:3,
# labels = ystratalabs) +
scale_x_continuous("Numbers at risk", limits = c(0, max(sfit$time))) +
theme(axis.title.x = element_text(size = 10, vjust = 1), panel.grid.major = element_blank(),
panel.grid.minor = element_blank(), panel.border = element_blank(),
axis.text.x = element_blank(), axis.ticks = element_blank(),
axis.text.y = element_text(face = "bold", hjust = 1))
data.table <- data.table + theme(legend.position = "none") +
xlab(NULL) + ylab(NULL)
data.table <- data.table +
theme(plot.margin = unit(c(-1.5, 1, 0.1, ifelse(m < 10, 2.5, 3.5)-0.28 * m), "lines"))
## Plotting the graphs
## p <- ggplotGrob(p)
## p <- addGrob(p, textGrob(x = unit(.8, "npc"), y = unit(.25, "npc"), label = pvaltxt,
## gp = gpar(fontsize = 12)))
grid.arrange(p, blank.pic, data.table,
clip = FALSE, nrow = 3, ncol = 1,
heights = unit(c(2, .1, .25),c("null", "null", "null")))
if(returns) {
a <- arrangeGrob(p, blank.pic, data.table, clip = FALSE,
nrow = 3, ncol = 1, heights = unit(c(2, .1, .25),c("null", "null", "null")))
return(a)
}
}
else {
## p <- ggplotGrob(p)
## p <- addGrob(p, textGrob(x = unit(0.5, "npc"), y = unit(0.23, "npc"),
## label = pvaltxt, gp = gpar(fontsize = 12)))
print(p)
if(returns) return(p)
}
}

Answer 1

The answer first, then an explanation.
Add the line:
coord_cartesian(xlim=c(0,max(sfit$time)))
to your ggplot object.
A simple example:
df <- data.frame(c(runif(10,0,1)),runif(10,0,1))
names(df) <- c("x","y")
p <- ggplot(df, aes(x,y)) +
geom_point() +
scale_x_continuous(breaks=c(0,0.25,0.5,0.75,1.0),
labels=c("0","0.25","0.5","0.75","1.0"))
p
gives you
whereas, if you add to the above code
p <- p + coord_cartesian(xlim=c(0,1))
p
you get
coord_cartesian() is a friend of yours (and anyone who uses your code after you) IFF you are absolutely certain any [visually important] data will never fall beyond the bounds you set within that very function. This is well documented; see Hadley's doc on this useful creature
To apply it to your code above and visualize it myself... I need to know what "sfit" is (as per user Pascal's insightful inquiry), likely among other idiosyncratic things. But the gist is:
scale_x_continuous (and its sisters scale_y_continuous et al.) don't hard-cut bounds to the EXACT parameters you specify. They do as the name implies, yes, "scale" axes according to a function (e.g. log10). But they always leave a cute little buffer around the limits, for assumed aesthetically preferential defaults.
'coord_cartesian`, on the other hand, DOES set axes limits EXACTLY as you specify, cutting out all space and data falling outside those bounds without altering the analyses pertaining to that entire data field. But don't take my word for it: read Hadley's apt description:
"The Cartesian coordinate system is the most familiar, and common, type of coordinate system. Setting limits on the coordinate system will zoom the plot (like you're looking at it with a magnifying glass), and will not change the underlying data like setting limits on a scale will."
To be clear, you can use both scale_x_continuous (and y) and coord_cartesian in the same ggplot object, because they do different things. The former sets breaks and labels for those breaks, the latter the frame (i.e. visual bounds) of the plot.
The other (dirty, dirty) solution is... photoshop. :(

Answer 2

Use:
scale_x_continuous("Numbers at risk", limits = c(0, max(sfit$time)), expand = c(0, 0))
The expand can also be used on your y-axis if needed.