09 Techniques: Regression & Qualitative Variables Flashcards
What are Probit & Logit models and how do they differ?
- types of generalized linear models
- modeling the relationship between predictor variables and a categorical outcome
- have versions for binary, ordinal, or multinomial categorical outcomes
Real difference is theoretical: different link functions
- > logistic regression uses a logit link function
- > probit regression uses an inverse normal link function
How many columns does a model with 1 qualitative regressor have?
d₁ = number of levels of the regressor
How many columns does a model with 2 qualitative regressors have?
d₁ + d₂ + 1
d₁ = number of levels of the first regressor d₂ = number of levels of the second regressor
How many columns does a model with 2 qualitative regressors with interactions have?
d₁ ⋅ d₂
d₁ = number of levels of the first regressor d₂ = number of levels of the second regressor
What are the steps to build data matrix X for a model with two qualitative regressors with interactions?
a) 1 column for constant
b) for the first regressor, make (d₁-1) columns.
- if d₁ = 2: make 1 column. code one level as 1, one level as -1
- if d₁ = 3: make 2 columns. code one level as 1, one level as 0, and the third as -1. switch 1 and 0 for the second column.
c) repeat step b) for the second regressor
d) (d₁-1) ⋅ (d₂-1) columns for interactions, where the relevant columns are multiplied element by element
Find Y* as a linear combination of vector X!
β = (β₁; β₂; β₃)ᵀ X = (X₁;X₂;X₃)ᵀ
Y* = β₁ X₁ + β₂ X₂ + ; β₃ X₃ + ε
Interpret β₁ :
Y* = β₁ X₁ + β₂ X₂ + ; β₃ X₃ + ε
If X₁ changes by one unit, we expect Y* to change by β₁, ceteris paribus.