Please enable JavaScript.
Coggle requires JavaScript to display documents.
Statistical Modelling (Partial & Part Correlation (exploring…
Statistical Modelling
Partial & Part Correlation
(exploring confounders)
Partial correlation
The direct correlation
between 2 va's
Implicitly accounts for a 3rd va in both va's
(ie value of 3rd va is held constant)
Part correlation
(=semi-partial correlation)
Correlation between 2 va's
where a 3rd va is
accounted for in 1 of the va's
Ordinary correlation
(=zero-order correlation)
The correlation between 2 va's
whether direct or
indirect (ie through another va)
Modelling strategies
Hierarchical
regression approach
Each block is sequentially fit
in the model using either the
enter method or stepwise methods
Works by collating va's
into thematic blocks
Combination of heuristic &
contextual (scientific)
model selection
Enter method
Analyst decides!
Empirical or Heuristic
methods
statistical software
selected va's
Stepwise
Strategy
(3 possible methods)
Backward Stepwise
approach
If not all va's are significant
then remove 'weakest' va
Refit model with revised va set
Add all va's into model
Repeat steps 2 & 3 until
all va's in model are significant
Full Stepwise
approach
Same as forward stepwise
except as 3rd va entered
it looks backwards to check that
all va's in model are significant
Hybrid between
forward & backward stepwise
Forward Stepwise
approach
Enter va with
highest
significant correlation
into model
Repeat steps 3 & 4
until no significant correlations remain
Calc the partial correlation after
adjusting for explanatory va's
already in model
Enter the va with the
highest significant correlation
into the model
Obtain correlations for each
y vs x pair