Nominal Variable Regression

Question 1

WHY CAN’T WE USE UNORDERED NOMINAL VARIABLES AS REGRESSORS?

Answer 1

• we assume a stat relation between DV/IV (ie. strong positive correlation between height/weight)
• makes sense to represent them as such as both variables are continuous ratio variables so there’s only 1 logical order in which scores for each variable should appear
• so can we fit a regression line through ordinal variable points? YES as there’s only 1 LOGICAL ORDER for coding (ie. 1st year = 19y; 2nd year = 20y etc.)

Question 2

UNORDERED (NOMINAL) IV

Answer 2

• changing order of coding radically changes results and can render a previously effective model useless
• we decide arbitrarily how to code unordered category variables (ie. subjects x IQ) aka. data is likely to be misleading

Question 3

CATEGORICAL VARIABLES

Answer 3

• ordered (ordinal) categories = OK as regressor/IV (ie. undergrad year x age)
• unordered (nominal) categories = NOT OK as regressor/IV
• if unordered category variable = regressor/IV -> no issue (2 lvls ie. sex)/use dummy variables (2+ lvls ie. subjects)