Doing more with less

ARS SEA IACUC workshop, July 18, 2024

Quentin D. Read

Who is this talk for?

Image (C) Anton Petrus/Getty Images

  • Practicing animal researchers without a strong statistics background
  • I’ll throw some ideas out there that might pique your curiosity
  • Use as a jumping-off point to learn more

Going beyond power analysis

  • You might remember me boring you with a talk on power analysis from last year
  • Power analysis is what you have to do to follow the rules, but it doesn’t encourage creative thinking
  • We can breathe new life into old methods and really elevate our research technique with these new ways of doing things (that actually aren’t that new)

How to get more mileage out of fewer animals

  • Creative experimental designs
  • Bayesian statistical analysis

1. Creative experimental designs

Reproducibility crisis

  • In many scientific fields there is concern about reproducibility
  • Gold standard: conducting the same study in multiple laboratories
  • If the effect is real, it will be detected under different experimental conditions
  • But this is often not feasible in practice

Image (C) Maki Naro

Blocks in experimental design

  • Within a laboratory, we can maximize statistical power by using a blocked experimental design
  • Example: cages, rooms, plots
  • Usually blocks are physically separated from one another in space

The problem with blocks

  • But these blocks are still not really independent of each other
  • They share a lot of environmental conditions among them, not just within each block
  • Paradoxically, our conclusions are still limited by the lack of heterogeneity among the blocks
  • Our experiments are “too controlled”

A potential solution: “Mini-experiment” design

  • We can make the blocks more independent of each other by blocking in both time and space
  • This mimics a multi-laboratory study because conditions change more between the time points, than they would if all blocks were run at the same time
  • Introducing environmental heterogeneity can increases realism
  • If the effect is detectable across a wider range of environmental conditions, it may indicate a more general pattern

Drawbacks of mini-experiment approach

  • May be more costly because it is less efficient to set up the experimental setup multiple times over a long period
  • Slows down the all-important time to publication

Image (C) Pixel-Shot/Adobe Stock

2. Bayesian statistics

photo of a neon sign of Bayes’ Theorem

Difference between Bayesian and frequentist probability

  • Classical statistics most of you are familiar with are called “frequentist”
  • Bayesian statistics and frequentist statistics are based on different interpretations of probability
  • Frequentists see probability only as the randomness of events in the world and their long-term frequencies, not our knowledge about them
  • Bayesians see probability as a combination of the randomness in the world and our knowledge about the world

Bayesian vs. frequentist probability

Usain Bolt, the GOAT

  • Probability, in the Bayesian interpretation, includes how uncertain our knowledge of an event is
  • Example Before the 2016 Olympics, saying “The probability that Usain Bolt will win the gold medal in the men’s 100 meter dash is 75%.”
  • In frequentist analysis, one single event does not have a probability. Either Bolt wins or Bolt loses
  • In frequentist analysis, probability is a long-run frequency
    • We could predict if the 2016 men’s 100m final was repeated many times Bolt would win 75% of them
  • But Bayesian probability sees the single event as an uncertain outcome, given our imperfect knowledge
  • Calculating Bayesian probability = giving a number to a belief that best reflects the state of your knowledge

Prior knowledge