5b. Logistic Regression

Question 1

What is a logistic regression used for?

Answer 1

To predict the probability of a certain class existing or an event occurring (yes/no) depending on one or several independent variables

Question 2

What measurement level can the independent variables be (in a logistic regression)?

Answer 2

SAME as the IVs in a LINEAR regression

• continuous (eg. age)
• categorical (eg. gender)

Question 3

What maesurement level can the dependent variables be? (in a logistic regression)

Answer 3

Categorical (nominal)

binary (0, 1) or dummy-coded (eg. Yes/ No)

Question 4

What shape does the logistic curve have?

Answer 4

S curve

Question 5

What does the dependent variable represent? (S-curve)

Answer 5

Represents probabilities, with values ranging from 0 to 1

Question 6

At what level of probability is the turning point?

Answer 6

p = 0.5

Question 7

How are regression coefficients estimated?

Answer 7

Using maximum likelihood estimation (highest possible for y = 1, lowest possible for y = 0)

Question 8

What is the same size requirement for logistic regression?

Answer 8

To allow meaningful interpretation each category should contain at least 25 observations. Assuming normal distribution, minimum sample of n = 50.

Question 9

How do you interpret the logistic regression coefficients?

Answer 9

Logistic is NOT linear. You can only interpret based on whether the coefficient is positive or negative..

eg. a positive regression coefficient will lead to increasing profitability that an event (DV) will be occurring.

Question 10

What is the “logit value”?

Answer 10

Calculated by taking the logarithm of the odds.

An odds value less than 1 will have a negative logit value, and an odds ratio greater than 1 will have a positive logit value.

Question 11

What is the “odds”?

Answer 11

p/(1-p)

[The ratio of an event happening to it not happening.]

Question 12

What are the extreme values of probability (p)?

Answer 12

1 = 100%
0 = 0%

Question 13

This shows the odds and the logit value of combinations of p:

Answer 13

Question 14

What was the movie streaming case of logistic regression?

Answer 14

Does access to movie streaming services (e.g., Netflix, Amazon Prime) lead to lower student success of the marketing analytics subject (pass vs. fail)?
Does the “self-control” of students increase student success?

Question 15

What does p in this case refer to? (movie streaming case)

Answer 15

The probability of passing the exam

Question 16

What does successExam variable say?

Answer 16

What actually happened, whether they, 1 = passed, or 0 = failed

Question 17

Why should you ignore Block 0 in SPSS? (logistic regression)

Answer 17

Because Block 0 is the “null” (constant-only) model:

The classification/table and fit stats in Block 0 are just a baseline and aren’t evidence your predictors work.

Question 18

What is important in this model summary?

Answer 18

R-Square can be interpreted in the same way. So according to Nagelkerke R Square we can infer 11% of the variance

Question 19

What does the -2 Log likelihood tell us?

Answer 19

Not much, it is often reported but is not an informative number by itself, it can however be used to compare different models. The smaller, the better.

Question 20

What do the model results show?

Answer 20

The model correctly predicts 169 out of 192 students who passed the exam.

Question 21

What is problematic about these results?

Answer 21

The model is actually really poor. The overall accuracy is noted as 63.7%. However, the model has a clear overaccuracy towards passing the exam. Even though that 88% looks great, it really is not, because in the context, the model ALREADY overcompensates on believing people will pass, == because its terrible in predicting people will FAIL, it’ll obviously be great at predicting people will PASS.

Question 22

What does the logistic regression coefficient tell you?

Answer 22

REMEMBER, it can only be interpreted using positive/ negative sign. It is NOT linear.

Question 23

What does the significance of the coefficients tell you?

Answer 23

Self control is significant at the 1% level. The access to movie streaming service is MARGINALLY significant (between 5-10%).

Question 24

What does the odds-ratios tell you about self-control?

Answer 24

Increasing self-control by one unit, will lead to a 42% increase in the odds of passing the exam.