Small-sample research is everywhere, but the methods actually designed for it? Basically ghosts. 👻📊
We need to learn these methods faster, and the best way is together. What if we treated statistical exploration like citizen science? (Think eBird, but for stats 🦅🔬).
In this thread, we use a Replext app from mightymetrika to crowdsource, replicate, and extend a method called "informative hypothesis testing."
Let’s dive in! 🧵👇
🌐: https://t.co/zSit9oRgK5
📊: https://t.co/8onefQzXmK
#rstats #statistics #math #simulation #citizenscience #science #data
1/
Group averages can lie. 📉
In small-sample longitudinal studies, looking only at the mean can hide the real story.
Visualizing individual trajectories reveals hidden heterogeneity, non-linear trends, and critical outliers. Here’s a practical tool from Carrig, Wirth, & Curran (2004; 10.1207/S15328007SEM1101_9) to fix that.
🧵👇 1/
#RStats #Statistics #Math #Simulation #Data #DataScience #MachineLearning #Graphics
P-values can hold your small-sample studies back. 📉
“Not significant” ≠ “No effect.”
Bayes factors fix this. They let you weigh the evidence for multiple plausible hypotheses at once—including your theory vs. the null. It’s time to upgrade how we measure evidence.
👇🧵 1/
Hard to collect a large sample size for your #clinical intervention #research? 🧐
Single-Case Experimental Designs (SCEDs) are a powerful alternative! Participants serve as their own controls, measured repeatedly over time [Maric, Marija & Werff, Vera. (2020). Single-Case Experimental Designs in Clinical Intervention Research. 10.4324/9780429273872-9]
Note: To meet strict WWC 5.0 standards, a well-designed SCED needs built-in replication (like an ABAB #design), not just a simple A-B phase. [https://t.co/yJ6CmZ4wq6]
But regardless of the specific design, how do we validate that our data shows a meaningful, reliable effect? Enter: The Non-Overlap of All Pairs (NAP).
#RStats #Math #Statistics #Programming #ggplot2 #Simulation
🧵👇 1/12
Does an effect hold for everyone the sample, or just on average? 🤔 Detecting a group-level effect doesn't guarantee it applies to every individual. Klaassen (10.4324/9780429273872-11) proposed a method to evaluate hypotheses case-by-case and aggregate the results. 🧵👇 #RStats
This game dev is making a game where you fly through a semi-open world as a paper plane. Wow!
- Transform between a plane, ball, or boat in real-time
- Realistic physics
- Unlockable upgrades
The game is called Paper Sky. Release date TBA.
🔬 Help us better understand statistical methods tailored for smaller datasets by contributing to our citizen science projects.
📊 Submit data: https://t.co/4ICFchaYSy💻 Contribute code & bug reports: https://t.co/4TYs3g7mjt
Every contribution counts! 💪 #math#RStats
Want to build #RStats#CitizenScience tools? 🧑💻
At https://t.co/fqrnuJ7bkj, we use mmints to streamline our REPLEXT Shiny apps.
It includes PostgreSQL database modules and more.
Contribute today:
https://t.co/azeOmckuds
#statistics#data
Ever read a brilliant statistics paper, but realized its simulation parameters don't quite match your real-world project? 🤔
Meet the REPLEXT (Replication & Extension) framework!
🚀 Let's bridge the gap between published research and your actual data.
1/4
Sometimes theory says "positive effect " but regression output says "p-value for not zero". Avoidable?
Wolak's 87 paper (10.1080/01621459.1987.10478499) gives exact tests for inequality constraints.
R package: restriktor
#Statistics#Rstats