Been reading a lot about ggplot2 and experimenting with its very elegant graphing capabilities. Underneath the beautiful designs are even more beautiful statistical principles.
The goal is to really see what the data look like and build on analyses approaches from there. The analyses is not lost in a mesh of numbers that the brain has trouble processing. Accompanying each step is a visual cue towards what is going on.
The way I process it, for my dissertation it is:
Describe
Describe in numbers.
Describe in pictures.
Describe in words.
Model
Is this model appropriate? Does it "fit" the data?
Are there influential data distorting the risk estimates?
What are the risk estimates?
What is the confidence I can place on these estimates?
Build the model
Throw in more variables in the model.
Are these changing anything?
Can these changes be explained by biology?
Is it interaction?
Is it confounding?
Explain the universe
Throw everything together?
What does it all tell me?
Approach data analyses in layers.
Simplify and harvest the maximum information from each layer.
Unify these harvest and view the bigger picture.
The goal is to really see what the data look like and build on analyses approaches from there. The analyses is not lost in a mesh of numbers that the brain has trouble processing. Accompanying each step is a visual cue towards what is going on.
The way I process it, for my dissertation it is:
Describe
Describe in numbers.
Describe in pictures.
Describe in words.
Model
Is this model appropriate? Does it "fit" the data?
Are there influential data distorting the risk estimates?
What are the risk estimates?
What is the confidence I can place on these estimates?
Build the model
Throw in more variables in the model.
Are these changing anything?
Can these changes be explained by biology?
Is it interaction?
Is it confounding?
Explain the universe
Throw everything together?
What does it all tell me?
0 Comments:
Post a Comment