# ddm model fit
Compare model-predicted data and observed data.
plots below: observed behavior against model-predicted behavior
- panels 1, 2, 3, 4: accuracy `acc`, overall rt `rt`, incorrect rt `rt0`, correct rt `rt1`
## no-pooling
1 model per subject. **1 dot is one subject**. red dashed line is diagonal (dots on the line = perfect fit). line is correlation/regression line. Model fit looks quite good for individual subjects.
![[s20220407_201212.png|900]]
When 1 dot is the mean of each country... Model predictions are systematically off (dots deviate systematically from the red diagonal line), though the relative ordering across countries is preserved.
![[s20220408_123011.png|900]]
## partial pooling (shrinkage towards country mean)
Bayesian hierarchical models fitted to subjects within countries (i.e., 1 model per country; 16 models). **1 dot is one subject**. red dashed line is diagonal (dots on the line = perfect fit). line is correlation/regression line. Some shrinkage of predicted accuracy (first panel)?
![[s20220407_201749.png|900]]
When 1 dot is the mean of each country... Excellent fit, unlike no-pooling above.
![[s20220408_122910.png|900]]