1. Binary Response
    1. Logistic Regression
    2. Inference
    3. Diagnostics
    4. Model Selection
    5. Goodness of Fit
    6. Estimation
    7. Prediction
  2. Binomial and Proportion Responses
    1. Binomial Regression
      1. link
        1. logit
          1. make retrospective equivalent to prospective
        2. probit
        3. cloglog
        4. cauchy
      2. Inference
      3. Goodness of Fit
      4. Prediction and Effective Dose
        1. prediction
          1. No distinction between future observation and the mean response
        2. effective dose
      5. large deviance
        1. structual deviance : need transformation
        2. outliers
        3. sparse data
        4. overdispersion
          1. causes
          2. dependece
          3. different pi
          4. solution
          5. when the covariate classes are roughly equal in size
          6. introduce a dispersion parameter
          7. estimate the dispersion parameter
          8. redo the test
          9. drop1(m1, scale = s2, test='Chisq')
          10. summary(m1, dispersion = s2)
          11. when the covariate classes are not roughly equal in size
          12. use dispmod package
          13. use F test
        5. check these two before trying dispersion parameter
    2. Beta Regression
  3. Count Responses
    1. Poisson Regression
      1. when to use
        1. probability depends on the length of time
        2. probability small, approximates binomial
      2. use chisq test
    2. Dispersed Poisson Model
      1. when to use
        1. deviance is large
        2. no apparent outliers
        3. no sparsity
      2. mechanism known
        1. negative binomial
      3. mechanism unknown
        1. dispersion parameter
        2. use F test
    3. Rate Model
      1. use log(num) as predictor
    4. Negative Binomial
      1. can use two parameters to depict mean and variance
      2. glm.nb
  4. Model Selection
    1. Cross Validation
    2. Stability