In September 2023, I taught a one-week course Statistics workflow in Nelson Mandela African Institute of Science and Technology (NM-AIST)a public postgraduate research university located in Arusha, Tanzania, founded in 2009.
This course was sponsored by the Dean Professor Ernest Rashid Mbega And that Center for African Research, Agricultural Advancement, Excellence and Sustainability Education (CREATES) through the leader Professor Hulda Swai and manager rose mosha.
Our case study was an experiment designed and conducted at the NM-AIST campus. Dr. Arjun Potter and Charles Luchagra study Effects of drought, fire, and herbivory on the growth of different acacia tree species. The focus was on Pre-data workflow steps, that is, an experimental design. This week's goal was to learn a common statistical language so that scientists can collaborate with statisticians.
Along with Arjun and Charles, we also received input from Dr. Emmanuel Muporia, Anna Treide, Andrew Gellman, michael betancourt, Avi Ferrer, Daphna Harrelland Joe BlitzsteinI made teaching materials Lots of activities. We asked the participants to: hand drawn Collaborate with teammates to create an experimental plan and its pre-planning. I also did paper-and-pencil calculations and coded in R.
Course participants were students and staff from across NM-AIST. Over the course of five days, he was attended by 15 to 25 participants on any given day.
Leveraging the ecological expertise of our participants, built the model To tell the mathematical story of how acacia tree height changes with drought, fire, herbivory, species, and plot location.we Simulated parameters and data from this model,for example beta_fire = rnorm(n = 1, mean = -2, sd = 1) after that Simulated_data …= rnorm(n, beta_0 + beta_fire*Fire +… beta_block(block), sd_tree).At that time we Fit the model to simulated data.
Because fire is difficult to handle, Fire was assigned at block levelOn the other hand, drought and herbivory were assigned at the sub-block level. We have seen how this reduces the accuracy when estimating the impact of a fire.
Rerunning the simulation assuming a smaller block effect improved accuracy. This confirmed the researcher's intuition that we should work diligently to reduce differences between blocks.
To focus on concepts rather than code, we only performed one simulation from the model. A complete design analysis includes many simulations from the model.In Section 16.6 ROS They are Fix one parameter value Simulate multiple datasets.in Gellman and Karlin (2014) They are Consider reasonable parameter ranges Use advance information. Betancourt Workflow Simulate parameters from previous parameters.
Fourteen participants completed the course evaluation survey. When asked, „What part of the class was most helpful in helping you understand the concepts?“ respondents chose the instructor's explanations, diagrams, and activities over the R code. However, participants also expressed their enthusiasm to learn R and analyze real data in their next course.
The hand-drawn course materials and activities were inspired by: Brendan Leonard's illustration bears don't care about your problems and I hate running, but you can too. Brendan wrote me a letter.
Don't you think hand-drawing is more fun and less scary?
Agree.
What I've been reading lately: Overview of modern causal inference Alejandro Schuler and Mark van der Laan said:
It’s easy to feel like you don’t belong or aren’t good enough to participate…
Yeah.
The voices we use throughout this book to address this issue are informal and distinctly non-academic. The illustrations are hand-drawn in a cartoon style.
We look forward to returning to NM-AIST and continuing our workflow steps using the data collected by Dr. Arjun Potter and Dr. Charles Luchagula. Using real data, we find that: Is our model realistic enough? How to achieve scientific goals?