STATISTICAL WORKFLOW:
Experimental Design and Model Building
Dr. Shira Mitchell and Dr. Arjun Potter
September 1-5 (Mon.-Fri.), 2025. Each day in NM-AIST Room L1 from 9:00-12:00 (including chai and mandazi).
Course Plan
This short course introduces a statistical workflow for an experiment here on the NM-AIST campus about the effect of drought, fire, and herbivory on acacia trees. The goal for the week is to learn some shared statistical language so you can work with statisticians (like me) on your research.
The first 4 steps help to design an experiment:
- What are our scientific goals ?
- How will we get data that informs our scientific goals ? What is our experimental design ?
- How do we model (i.e. tell a mathematical story about) our data ?
- Assuming this model, will we get precise enough estimates to achieve our scientific goals ?
The next step analyzes our data:
- Is our model realistic enough to achieve our scientific goal ?
Team activity (2 minutes): Throughout class, we will have activities in green. You will work in pairs, and then share your work with the class. A statistical workflow is best done in teams, not alone. Please introduce yourself to your teammate !
Reading
Students are welcome to read these course materials ahead of class. Maoni karibu !
Other reading can include the online textbook Regression and Other Stories, by Gelman et al..
Other references
The course materials are based mainly on Regression and Other Stories, by Gelman et al. Chapter 16 about design, Gelman (2023), and Towards A Principled Bayesian Workflow, by Betancourt. For more about step 5 and beyond, see also Bayesian Workflow, by Gelman et al.. Learning statistics with R, by Danielle Navarro is another good resource. We include further references below.Scientific Goals
We want to estimate the effects of drought, fire, and herbivory on African acacia species. Many outcomes are interesting, but in this workflow we focus on height.
Team activity (5 minutes): Why is it important to estimate these effects ? What would you want to know ?
An answer
The height of a tree is a simple way to measure the size and growth rate of a tree. As a tree grows, its height usually increases. Trees that are in poor health or struggling to get key resources will grow very little and thus change little in height. Height may even decrease if upper branches die! So tree growth rate is a useful indicator of the health of a tree.
Tree height is not just a quick way to measure size; the height of a tree itself matters for many reasons. Taller trees can capture more light, which allows them to compete with other plants. Once a tree is tall enough, its leaves are above the height of fire and animals, and so it will escape these stresses. The height of a tree also matters aesthetically; tall trees are beautiful!
We expect that different species will have different heights, on average. We can keep this in mind for our statistical modeling later.
Perhaps you can add another reason why tree height is worth measuring?Acacia Trees
Acacia trees (Miti Migunga) are a very important group of trees across the world, but particularly in Africa. These include such iconic species as the umbrella thorn and the fever tree.
Currently, the former genus Acacia has been divided into two genera: Vachellia and Senegalia.
The twelve Acacia trees in this study are as follows:
Vachellia nilotica, V. gerardii, V. robusta, V. seyal, V. sieberiana, V. tortilis, V. xanthophloea
Senegalia brevispica, S. mellifera, S. polyacantha, S. senegal, V. drepanolobium
These are referred to using a 6-letter code, created from the first
three letters of the genus and species. For instance, Vachellia
nilotica is “vacnil” and Senegalia brevispica
is “senbre”.
Experimental design
We will get data from a factorial experiment (factors: drought, fire, herbivory) on the NM-AIST campus done by Arjun Potter and Charles Luchagula. They transplanted seedlings of 12 Acacia species from Serengeti National Park to the experimental site on campus.
Your ideas
Before we review the design of Arjun’s and Charles’ experiment, I want to hear your ideas !
Team activity (35 minutes): How would you design a study to estimate the effects of drought, fire, and herbivory on African acacia species heights ?
Layout of the site
The experimental site consists of 10 blocks (12 m × 12 m), 2 m apart. Drought was implemented via rainout shelters over each block.
Each block contains 9 subplots.
Each subplot contains 6 plant positions.
Team activity (13 minutes): Try to draw a possible layout like this on your paper. What questions do you have ? What have I not told you that you would like to know ? Why divide the rainout shelter blocks into subplots ?
An answer
Here is the layout for this experiment’s design: