Threshold analysis

Learn me by follow me exercises

This module runs two-piece-wise regression models to examine the threshold effect of risk factor on outcome.

Threshold effect means when the risk factor (X) reaches to certain point, its relationship with outcome (Y) starts to change, could be from no relation to positive or negative relation, or from positive or negative relation to no relation, or from positive relation to negative relation, etc.

If you already have the turning point (the point where the relationship starts to change) is, you can input it in the “Enter turning points” box. 

To let this module find the turning point for you, choose “Auto determine 2-pieces-wise regression”.  This module uses maximized log likelihood method.  It first runs multiple two-piece-wise regression models.  Each model uses a percentile (between 10% and 90%) of risk factor (X) as turning point, and then picks the model with the turning point that provides maximum log likelihood.

The output table includes one-line linear regression (model I), two-piece-wise regression (model II) that gives maximum log likelihood, and p-value from log likelihood ratio test comparing model I versus model II.

The module automatically detects whether the outcome variable is binary or continuous and selects logistic or linear regression as appropriate.  You also can specify the distribution and link function for each outcome manually, right click the variable and then select “Change functions”. In the popup window, select the distribution and link function.

What it’s used for

This module is used to determine whether the risk factor(s) has any threshold effect or saturate effect on the outcome(s). It also describes the relationship between risk factor(s) and outcome(s) if such effect exists.

A screen shot of sample design:

t8_input.gif

A sample output table:

t8_pwreg.gif