Regression

Question 1

What is regression analysis

Answer 1

Analysis allowing for adjustment of confounders

Question 2

Types of regression to know

Answer 2

linear - continuous numeric outcome which can be used for curves despite being called linear
logistic - binary outcome, transformation of outcome is modelled

Question 3

simple vs multivariable regression

Answer 3

simple has one predictor so no accounting for confounders, multivariable adjusts for other factors

Question 4

Linear regression coefficients

Answer 4

y is outcome
a shows position of line (doesn’t always have practical meaning), shows as (constant) on SPSS
b is regression coefficient, measures association between predictor and outcome
b1, b2 etc are partial regression coefficients for multiple regression coefficients for each variable, assuming all other confounders remain constant
P-values for all coefficients

Question 5

dummy variables use

Answer 5

Used for categoric confounders
One less dummy variable than number of categories
Set one as reference category
Dummy variable ‘b’ values are compared to reference variable

Question 6

Archaic/Bayes information criterion

Answer 6

AIC/BIC, lower the better

Measures how well the model fits the data

Question 7

Residuals

Answer 7

Differences between observed and predicted values for model

Minimising residuals overall finds line of best fit

Question 8

Linear regression assumptions

Answer 8

continuous outcome
Predictor variables not dependent on each other - multicollinearity
Residuals normally distributed around 0
Residuals have constant variance

Question 9

Multicollinearity measure

Answer 9

Variance inflation factor (VIF), e.g. 10 means 90% of variability explained by other predictors
Lower is better, >5 should be investigated

Question 10

Homoscedasticity

Answer 10

Constant variance of residuals for all values of predictors and outcome, one of the necessary assumptions for linear regression

Question 11

Transformation for logistic regression

Answer 11

Binary outcomes interpreted as odds then log(odds) used as this is linear, spss fits log(odds) in same way as linear regression

Question 12

Interpreting logistic regression outcome

Answer 12

exponential of (b coefficient (in binary variables) x category value) gives to back transform, gives odds of outcome
The odds ratio tells you how the odds of the outcome
changes for every one unit increase in the predictor variable for numeric variable confounders

Question 13

Logistic regression assumptions

Answer 13

No multicollinearity (VIF can be used to check)
Binary variable
No need to check residuals, don't have to have normal distribution