If the linear prediction rule is correct, what should be expected?
The error component should be entirely random
There are no systematic components (biases) in data
What are the different names given to residuals based on their position in residual plots?
Below the line
Above the line
How would a residual plot show that a straight line model was not appropriate?
If the residuals vary as across the plot, for instance being negative at the extremes and positive in the centre
This would suggest the data is curvilinear
When can residual plots be produced?
When both DV and IV are random variables
When can residual plots be used?
When there is no clear IV
Simply select one variable as the IV
What is the pure error (lack of fit) test divided into?
Lack of fit
How do we obtain the pure error?
Measure the residuals relative to the mean response value for each condition rather than relative to the predicted value on the regression line
What can be inferred if the mean values coincide precisely with the predicted values?
The means all lie on the regression line
The residual error equals the pure error and the lack of fit SS is 0
It must be considered, however that due to sampling error it is highly unlikely that all of the means will be in the regression line
What are replicates?
Measurements of Y at a given value of X
How can we guarantee the existence of replicates?
When participants are assigned to a given value of X
Replicates might not occur if the IV is random