Load data.
Descriptives
How many participants in each condition?
## Source: local data frame [2 x 2]
##
## training_condition n()
## (chr) (int)
## 1 active_active 20
## 2 passive_passive 20
How long did each condition take?
## Source: local data frame [2 x 3]
##
## training_condition m_train_time m_exp_time
## (chr) (dbl) (dbl)
## 1 active_active 0.6112942 4.434711
## 2 passive_passive 0.4713317 3.475504

Visualization
Sanity check on relational pretest

Seems reasonable – participants are performing above chance for most questions.
The hardest questions were:
- Is a parallelogram and rectangle?
- Is a square a rhombus?
Sanity check on entity pretest for all 16 question_shape combinations

- People are good at selecting the shape we ask for: PaPa, ReRe, RhRh, SqSq.
- They are not likely to select other shapes for parallelograms or squares.
- There is more disagreement between ss about rectangles and rhombi.
- For rectangles, people also select rhombi.
- For rhombi, people select rectangles.
Sanity check on entity pretest for all 4 questions

How do we define chance performance on the entity test? Chance level changes for each question:
Rectangle: 2/4 are rectangles Parallelogram: 1/4 are paralleograms Square: 4/4 are squares Rhombus: 2/4 are rhombi
Overall accuracy analysis for both entity and relational tests

Within subjects change scores all shapes

Accuracy on the learned shape
Overall accuracy analysis for both entity and relational tests

Within subjects change scores for learned shape tests

Models (todo)
Predict overall accuracy based on condition and test type
Predict accuracy for learned shape based on condition and test type
Predict change scores for learned shape based on condition and test type
Exploratory Analysis
Training time distribution for each condition

Median split on training time for active learning