Lecture 8 - Logistic Regression

Question 1

What is logistic regression?

Answer 1

The prediction of group membership from a continuous independent variable

Question 2

What is binary logistic regression?

Answer 2

The prediction of two groups from a continuous variable.

Question 3

What is binary logistic regression also called?

Answer 3

Binomial logistic regression

Question 4

What is multinomial logistic regression?

Answer 4

Predicting more than two groups from a continuous variable.

Question 5

The odds of an event can be calculated by…?

Answer 5

Dividing the probability of the event occurring by the probability of the event not occurring.

Question 6

Why are log odds used rather than normal odds?

Answer 6

Log odds are symmetrical around 0, which is better for statisticians.

Question 7

Why is logistic regression so-called?

Answer 7

It is based on natural logarithms.

Question 8

Logistic regressions aren’t actually used for predictions, they are used for…?

Answer 8

Modelling data.

Question 9

What level of data is needed for logistic regression to work well/best?

Answer 9

High quantity of data.

Question 10

What does Block 0: Beginning Block predict about your data?

Answer 10

All cases are predicted to result in the most common outcome - the accuracy of this prediction is then reported in a later table.

Question 11

The first classification table tells us what?

Answer 11

How good the (first) model predicted in Block 0 is.

Question 12

What does Block 1 method: Enter tell us?

Answer 12

• Iteration history: summary of how the model has been approximated and improved.
• Reports the results of logistic regression analysis.
• Whether it is a functional model, being better than block 0 or not.

Question 13

-2 log likelihood has to be as ____ as possible to reflect a more accurate/effective model?

Answer 13

Small

Question 14

What does the Omnibus Test of Model Coefficients tell us?

Answer 14

Whether the model is significant - tests general fit of the model.

Question 15

What does the model summary tell us in a logistic regression?

Answer 15

• The final -2log likelihood of the model

- An approximation of R^2 for the model

Question 16

Cox and Snell R Square and Nagelkerke R Square represents what?

Answer 16

An approximation/estimation of R Square for the model.

Question 17

Can an equation be used to predict data points in logistic regressions, like for linear regressions?

Answer 17

No.

Question 18

In Logistic Regression outputs under ‘Variables in the Equation’, what does the WALD statistic and Sig. represent?

Answer 18

Usefulness of each predictor - whether or not each predictor is significant in the model.

Question 19

In Logistic Regression outputs under ‘Variables in the Equation’, what does the B represent?

Answer 19

Coefficients for each predictor variable in the model.

Question 20

In Logistic Regression outputs under ‘Variables in the Equation’, what does the Exp (B) represent?

Answer 20

The change in predicted odds for each unit change in the predictor variable.

E.g. an increase of 1.898 in the odds of re-offending for every extra year/day/month that someone is convicted for.

Question 21

The constant is the same as what?

Answer 21

The intercept.

Question 22

What are 4 limitations of logistic regressions?

Answer 22

• Calculation only works if the relationship between the predictors/probabilities and the dependent variable is sigmoidal
• Very sensitive to outliers
• The ratio of sample size to variables needs to be quite high (needs large numbers of participants)
• SPSS assumes that relationships are being described, rather than being predicted.

Question 23

The classification tables in the output of logistic regressions relate to what?

Answer 23

The sample only.

Question 24

Why can logistic regressions not provide an equivalent of R Square Adjusted?

Answer 24

Logistic regression provides a model of the data from the sample, not a formula or prediction and not for the wider population.

Question 25

In logistic regressions, the significance of the model generated is calculated using what?

Answer 25

An x squared-like measure.

Question 26

Why is significance measured using an x squared-like measure in logistic regressions?

Answer 26

Because the DV is categorical.

Question 27

In logistic regressions, what type of variable is the DV?

Answer 27

Categorical

Question 28

The second classification table tell us what?

Answer 28

How good the (second) prediction in Block 1 was.