Count Data

Question 1

Problem with count data

Answer 1

• NV is violated
• NV is just for continuous variables
• NV allows values below 0 -> count not
• NV is symmetrical -> count not

Question 2

GLM

Answer 2

• link function = log
• log transform expected value
  -> expected value always positive

But: definition of residuum is no longer clear -> autoplot() automatically picks the right ones

Question 3

Overdispersion

Answer 3

If Residual Deviance&raquo_space; Df
-> too small p-values
-> probably some x are missing
-> p-value of X^2 test for overdispersion (in anova) will be very small

-> use Quasipoisson

Question 4

Underdispersion

Answer 4

If Residual Deviance < Df
-> too large p-values
-> Quasipoisson

Question 5

Zero-inflation

Answer 5

Overrepresentation of 0
-> use new model

Question 6

Dispersion

Answer 6

Residual Deviance : Df =
>1 = Overdispersion
<1 = Underdispersion

Question 7

Link function

Answer 7

Used to transform the expected values of the response variable (y)
NOT of explanatory variables (x)
NOT of observed values of response variable (y)