Field Ch. 20 - Logistic Regression

Question 1

Why use Logistic Regression?

Answer 1

to predict binary categorical outcomes

from categorical & continuous predictors

Question 2

What is the possible range of model outcome values?

Answer 2

values between 0 (outcome will not occur) and 1 (outcome will occur)

ex: 0.45

Question 3

R and R^2 tell us what?

Answer 3

• express model fit
• pos value = as predictor goes up, likelihood of outcome occurrence goes up
• neg value = as predictor goes down, likelihood of outcome occurrence goes down

Question 4

What’s the Wald statistic do?

Answer 4

determines significance of individual predictors

Question 5

Odds ratio tells us what?

Answer 5

change in odds due to one unit change in predictor

Question 6

What to look for before analysis?

Answer 6

• missing data (predictor & outcome): bad

- complete separation (graph with no overlap on x-axis): bad

Question 7

6 assumptions for BLR

Answer 7

1. binary outcome
2. independent observations
3. no multicollinearity (run correlation matrix, pearson’s less than 0.7)
4. no extreme outliers
5. linear reln btwn ind. variables and logit of dep variable (Eric did not discuss)
6. large sample size

Question 8

SPSS step 1

Answer 8

Analyze > Regression > Binary Logistic

• enter independent variables in blocks
• assign reference group (‘categorical’ button)

(less about memorizing and more about familiarity with boxes to check during analysis)

Question 9

SPSS step 2: ‘Save’ button

Answer 9

select:

• predicted probabilities
• group membership
• Cook’s (should be less than 1)
• standardized residuals (few above 1.96, none > 3)
• include covariance matrix

Question 10

SPSS step 3: ‘Options’ button

Answer 10

select:

• hosmer lemeshow
• CI @ 95%
• include constant in model

(less about memorizing and more about familiarity with boxes to check during analysis)

Question 11

SPSS: How to transform variables

Answer 11

Transform > Recode into different variable

• move old variable to center block
• type new variable name, ‘Change’

Old and New Values

• enter old and new values, ‘Add’
• ‘Continue’

Question 12

Reporting results: where do you find X^2?

Answer 12

Omnibus table

X^2 (df, N) = ###, p < Sig

• X^2 (value)
• df (number)
• model significance (p value)

Question 13

Reporting results: where do you find odds ratio?

Answer 13

‘Variables in the Equation’ table

• Exp (B)
• remember your reference group (set in the ‘Categorical’ menu)

Question 14

Reporting results: where can you find R^2 values?

Answer 14

‘Model Summary’ table

• Cox & Snell -or- Nagelkerke

Question 15

Reporting results: what is Exp(B)?

Answer 15

Odds ratio

Question 16

What does X^2 tell us?

Answer 16

difference between observed frequencies and expected frequencies if an effect didn’t exist

Question 17

What does Cook’s D(istance) tell us?

Answer 17

• the influence of a data point

- should be less than 1

Question 18

What does Hosmer-Lemeshow tell us?

Answer 18

• R^2

- another model fit indicator (others are Cox & Snell, Nagelkerke)

Question 19

What do standardized residuals tell us?

Answer 19

• strength of the difference between observed and expected values
• few > 1.96
• none > 3