GLM Model Validation and Checks

Assess Output Reasonableness
– Standard errors < estimated values
– Base reflects categories with highest exposure
– Factors with similar estimates should be grouped
– Relativities for factors make intuitive sense
– ensuring continuous curves are only fitted to reasonable factors
– ensuring interaction effects have been allowed for
– assess changes to model one change at a time

Compare Actual vs Modelled

Deviance
– how far the actual data differs from the fitted model

Degrees of Freedom
– number of places the actual data could differ from model
– number of observations less parameters of model

Statistical tests
– F-Test: [(D2 – D1)/(DF2 – DF1)] / [D1/DF1]
.    checks overall goodness of fit
– t/z-Test

Use Residual Plots to assess:
– Heteroskedasticity
– Outliers
– Biasness

QQ Plot
– to assess residual quantiles against expected residuals of normal distribution
– should be a straight 45 degree line, otherwise:
.    if greater then tail heavier than assumed
.    if less then tail is not as heavy as assumed

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s