16. How not to lie with statistics

Question 1

What are ways to lie with statistical analysis and inferences…

to do with experimental design (1)

Answer 1

Bias, replication, sample sizes

Question 2

What are ways to lie with statistical analysis and inferences…

to do with performing statistical analysis (3)

Answer 2

Data visualization

Principles of modeling

Overfitting (and avoiding it)

Question 3

What are ways to lie with statistical analysis and inferences…

to do with how we make inferences and interpret statistics (3)

Answer 3

Rejecting/accepting hypotheses

A bit of Bayers

Base rate fallacy

Question 4

How can we avoid bias?

Answer 4

increase precision
increase power

Question 5

What is pseudoreplication?

Answer 5

the process of artificially inflating the number of replicates

Question 6

How do we avoid errors when statically modeling?

Answer 6

Formulate a model that is commensurate with the design/data

Fit model to data; evaluate hypotheses

Critique the model

using residuals/outlier analysis

Consider transformations

Repeat….

Question 7

What types of statical models are there?

Answer 7

Full models

Maximal Models

Minimal Adequate Model (MAM)

Null model

Question 8

What is overfitting?

Answer 8

When there are more parameters describing a relationship than what is giving us an appropriate
inference

f-ratios can be sued to penalize complicated models

Question 9

What is the base rate fallacy?

Answer 9

The prior probability that something was true before new evidence occurred.

prior is also known as the base rate