Setting up R and RStudio for ecology

A first R session for ecologists: RStudio projects instead of setwd, installing packages once, and why a script run in a clean session is the only reproducible analysis.

Most ecology tutorials start with the data already loaded, as if the counts arrive in your R session by magic. They do not. Before any diversity index or ordination, you have to open R, point it at a folder of field data, load the packages you need, and run your code in a way that you (or a reviewer, or you in six months) can reproduce. That last part is where most beginner workflows quietly break, and it has nothing to do with statistics.

This post sets up a working session the way a careful analysis needs it: an RStudio project instead of a hard-coded path, packages installed once and loaded every time, and a script rather than a stream of console commands. The payoff is concrete, so we finish by watching a leftover value in the workspace silently change a diversity estimate.

The console is for trying things; the script is the analysis

When you open RStudio you get four panes. The one that matters first is the console (bottom left), where you type a line and R answers immediately. It is ideal for trying something out. It is a poor place to keep an analysis, because the console has a memory you cannot see and a colleague cannot reproduce.

The script editor (top left) is where the analysis lives. You write your commands in a .R file (or a Quarto document), save it, and run it top to bottom. The difference is not cosmetic. A script is a record: anyone with the script and the data can get your numbers. A console session is a conversation that vanishes when you close R.

A useful habit from day one: type exploratory lines in the console, but the moment a line becomes part of the analysis, move it into the script. If it is not in the script, it did not happen.

Projects, not setwd()

The single most common reason a colleague cannot run your script is a line like this:

setwd("C:/Users/alice/Documents/fieldwork/2025/")

That path exists on exactly one computer. On anyone else’s machine the script fails on line 1, and even on your own machine it breaks the moment you move the folder or rename it.

The fix is an RStudio project. In RStudio choose File then New Project, and make a project folder for your study (say, meadow-survey/). RStudio drops a small .Rproj file in it and, importantly, sets the working directory to that folder every time you open the project. Now your data lives at a relative path:

counts <- read.csv("data/plot-counts.csv")

There is no drive letter and no username. The same line works on your laptop, your supervisor’s desktop, and the machine that renders the analysis. Keep the raw data in a data/ subfolder, your scripts in the project root, and never write setwd() again. (For the broader habits that make an analysis rerun cleanly, see the reproducible-workflow tutorial linked at the end.)

Install once, load every session

Two verbs trip up every beginner because they look interchangeable and are not.

install.packages("vegan") downloads a package from CRAN and copies it onto your computer. You do this once (and again when you want a newer version). It does not belong in your analysis script, because rerunning it re-downloads the package every time.

library(vegan) makes an already-installed package available in the current session. R starts each session with only the base packages loaded, so you call library() at the top of every script, for every package you use. Group the library() calls at the top; leave install.packages() out of the script, or comment it out.

# install.packages(c("vegan", "dplyr"))   # run once, by hand, then comment out
library(vegan)
library(dplyr)

If a script errors with there is no package called 'vegan', you skipped the install. If it errors with could not find function "diversity", you skipped the library().

Why the clean session matters: a leftover value

Here is the measurable point, and it is the reason the console-versus-script distinction is more than tidiness. We compute Shannon diversity for one plot. The data is four species and their counts:

counts <- c(`Carex flava` = 12, `Juncus effusus` = 5,
            `Poa pratensis` = 8, `Festuca rubra` = 1)
N <- sum(counts)              # this plot's total: 26 individuals
p <- counts / N               # relative abundances
H_correct <- -sum(p * log(p)) # Shannon diversity
round(H_correct, 3)

[1] 1.162

The correct total is 26 individuals and the Shannon value is 1.162. Now picture the ordinary way beginners work. Earlier in the console, for a different plot, you had typed N <- 40. You then write the script above but forget the line N <- sum(counts). Because N is still sitting in the workspace as 40, the script runs without complaint:

# the workspace still holds N <- 40 from an earlier, different plot
N_stale <- 40
p_stale <- counts / N_stale         # divides by the wrong total
H_stale <- -sum(p_stale * log(p_stale))
round(H_stale, 3)                   # a Shannon value, but the wrong one

[1] 1.035

The script produced 1.035 instead of the correct 1.162, a 10.9 percent error, with no warning of any kind. The only hint that something is off is that the relative abundances now sum to 0.65 rather than 1, and nothing forces you to notice.

Run the same script in a fresh session, where N was never defined, and R stops you:

clean <- new.env()               # an empty workspace, as a fresh session would have
clean$counts <- counts
result <- tryCatch(
  eval(quote(counts / N), envir = clean),   # N does not exist here
  error = function(e) conditionMessage(e)
)
result

   Carex flava Juncus effusus  Poa pratensis  Festuca rubra 
    0.46153846     0.19230769     0.30769231     0.03846154 

The clean session turns a silent wrong answer into a loud, fixable error. This is the whole argument for the workflow above: “it ran on my machine” is not the same as “it is correct”, because your machine remembers things your script never wrote down. In RStudio the check is one menu item, Session then Restart R, followed by running the script from the top (Run then Run All). If it fails, you found a hidden dependency. If it runs, your numbers came from the script and nothing else.

Figure 1: Relative abundance of the four species in the plot. This community profile is what a Shannon value summarises into one number, and dividing it by the wrong total (a stale workspace value) distorts every bar.

Where to go next

You now have a project, a script, and a session you can restart with confidence. The next step is getting real field data into that session, where the first surprise is that a blank cell is not a zero. The reading-field-data tutorial picks up exactly there.

Related tutorials

• Reading field data into R
• Your first ggplot: data, aes and geom
• A reproducible statistical workflow
• From a field sheet to your first analysis

References

Wickham H, Grolemund G 2017. R for Data Science. O’Reilly. ISBN 978-1-491-91039-9.

Bryan J 2018. The American Statistician 72(1):20 (10.1080/00031305.2017.1399928).

Sandve GK, Nekrutenko A, Taylor J, Hovig E 2013. PLoS Computational Biology 9(10):e1003285 (10.1371/journal.pcbi.1003285).