2.2 Flashcards
Assessing Predictors: The Wald Statistic
- Similar to t-statistic in regression.
- Tests the null hypothesis that π = 0.
- Is biased when π is large.
- Better to look at Odds Ratio statistics.
ππππ πππ‘ππ = 3.42 example meaning
Patients that get the intervention are 3.42 times more likely to get cured than patients without the intervention.
odds ratio
odds after a unit change in the predictor / odds before a unit change in the predictor
the ratio of your chances of getting cured when taking the medication / the chances of getting cured without taking the medication
-> that is actually the effect of the treatment that you want to investigate
high / low: log β likelihood
The higher the value, the better a model fits a dataset.
log-likelihood isa measure the goodness of fit for a model
Model deviance
- It is an indicator of how much unexplained information there is after the model has been fitted.
- Large values indicate poorly fitting statistical models. When our model is precise, the LL will be close to 0.
deviance = -2LL
Things That Can Go Wrong
Unique problems
β’ Incomplete information
β’ Complete separation
Incomplete information from the predictors
Categorical predictors:
β’ Predicting cancer from smoking and eating tomatoes.
β’ We donβt know what happens when non-smokers eat tomatoes because we have no data in this cell of the design.
Complete Separation
- When the outcome variable can be perfectly predicted by an X variable, you cannot run the model because there is complete separation in the data.
- If you think about it: This might actually not be a problem, because if X indeed perfectly predicts Y, this might be very helpful to you.