6

Given a minimal dataset where am looking for the occurrence of a certain motif within a dataset of 500 observations. with_motif represents obervations with the specified motif and without_motif are observations without the motif.

with_motif <- 100
without_motif <- 400
dt <- data.frame(with_motif,without_motif)

The following code will plot a bar-chart using ggplot2 library,

bar_plot <- ggplot(melt(dt),aes(variable,value)) + geom_bar() + scale_x_discrete(name="with or without") + theme_bw() + opts( panel.grid.major = theme_blank(),title = "", plot.title=theme_text(size=14))

bar_plot

I would like to compute a standard error at 95% CI and attach a barchart to the plot. ggplot offers geom_errorbar() but I would be glad to know different ways for deriving the standard errors(deviation) so as to calculate the errorbar limits(CI).

eastafri
  • 543
  • 3
  • 9
  • 15
  • +1, but kindly avoid "plot" as an object name! – Ari B. Friedman Aug 11 '11 at 10:36
  • yep! edited and nice advice! – eastafri Aug 11 '11 at 10:38
  • 1
    @Biorelated There's a very good online help for ggplot2, e.g. http://had.co.nz/ggplot2/geom_errorbar.html. I also provided some example of use of `geom_errorbar()` in [this response](http://stats.stackexchange.com/questions/9477/r-interaction-plot-with-confidence-intervals/9480#9480). Is this what you're after? – chl Aug 11 '11 at 10:50
  • Thanks I had a look at it. The question is on how to arrive at the standard errors. for example the ones given at had.co.nz/ggplot2/geom_errorbar.html. – eastafri Aug 11 '11 at 11:03
  • @Biorelated In that case you need to provide more information about your statistical problem. There's no magic formula to get good CIs! – Ari B. Friedman Aug 11 '11 at 11:21
  • 1
    @Biorelated As can be seen in my response, you'll need to compute SD or SE or 95% CI yourself, and add them to your data frame (or use another one), then call `geom_errorbar()`. If your question is about how to compute those estimates, then you may consider updating your question and provide more information on your dataset. – chl Aug 11 '11 at 11:21
  • I have rephrased my question. It seems my challenge is actually on deriving the CI that is necessary for assigning the error bar limits. There maybe many ways but i will be glad with the most accepted approach. – eastafri Aug 11 '11 at 11:40
  • 1
    @Biorelated I don't quite follow your update. To compute any measure of precision of a given estimate, we need to know what distribution it follows, right? So, what do `with` and `without` stand for in your code? – chl Aug 11 '11 at 11:54
  • with represents observations that contain a certain motif. I have clarified this in the main question. Thanks! – eastafri Aug 11 '11 at 12:06

1 Answers1

6

Here's an example from the ggplot2 homepage: https://ggplot2.tidyverse.org/reference/geom_errorbarh.html as others have mentioned in the comments, you have to calculate SE on your own and append this information to the data.frame

df <- data.frame( 
  trt = factor(c(1, 1, 2, 2)), 
  resp = c(1, 5, 3, 4), 
  group = factor(c(1, 2, 1, 2)), 
  se = c(0.1, 0.3, 0.3, 0.2) 
 ) 
df2 <- df[c(1,3),]
p <- ggplot(df, aes(fill=group, y=resp, x=trt)) 
p + geom_bar(position="dodge", stat="identity") 
dodge <- position_dodge(width=0.9) 
p + geom_bar(position=dodge, stat="identity") + geom_errorbar(aes(ymax = resp + se, ymin=resp - se), position=dodge, width=0.25)

As a pointer, SE @ 95% CI usually looks something like this: 

df$se <- 1.96*(sd(your_data, na.rm=T)/sqrt(your_n))

Your upper and lower CI bounds will just be df$se +/- the response (as shown in the aes() for geom_errorbar(), above)

Brandon Bertelsen
  • 6,672
  • 9
  • 35
  • 46