Residual and Predicted

This function was used to adjust some variables by other covariate(s) using generalized linear or additive model (spline smoothing), calculate the predicted value and residual value.  The predicted and/or residual value will be used in further analysis.

For example, we want to look the association of a SNP marker (SNP1) with blood pressure (SBP and DBP), we know SBP and DBP were affected by age, body mass index (BMI), cigarette smoking and education.  We want to adjust SBP and DBP by age, BMI, smoke and education and calculate SBP and DBP residuals first.

A screen shot of sample input window:

ScreenHunter_02 Jan. 02 18.14.gif

Drag and drop variables from variables list to relevant box (Dependent, Independent linear terms, Smoothing terms).

Technical notes:

1)      If you want derive the regression model from a subset of population (called reference population), you can specify the reference population by type in the conditions.  For example, your study population include random selected subjects (POP=0), and affected subjects (POP=1), you want to generate the predictive model from random selected subjects only, you just type in “POP=1” in the text box.

2)      You can apply generalized additive model to generate the predicted value and residuals.  Drag and drop the smoothing term (eg. age and body mass index) to smoothing terms list.  The default degree of freedom for each smoothing term is using GCV method to determine.  You also can specify the degree of freedom manually by right click the variable and then click “change degree of freedom”

3)      For each dependent variable, the default new variable name for holding the predicted value is the original name plus “.pred” for holding the residual value is the original name plus “.resid”.  You can change the name by click the “.pred” and/or “.resid” text.

4)      You can specify a stratified variable to get separate predictive model for each stratum.

 

Click “Save”, new variables will be added to the variables list.