Deck 20: Categorical Outcomes: Logistic Regression

Question 1

A researcher was interested in predicting whether a person would attempt to commit suicide (score = 1) or not (score = 0) from their depression scores. They found that the value of Exp(B) (the odds ratio) was 2.56. How should this value be interpreted?

A)2.56 times more people who attempted suicide scored highly on depression.
B)If two people have depression scores that differ by 1 unit, then the odds of the person with the higher score attempting suicide are is 2.56 lower than for the other person.
C)If two people have depression scores that differ by 1 unit, then the odds of the person with the higher score attempting suicide are 2.56 higher than for the other person.
D)25.6% of the variance in suicide attempts is explained by depression.

Answer 1

C

Question 2

The statistical implication of using a parsimony heuristic is:

A)Predictors are not included unless they have explanatory benefit.
B)Models should be as complex as possible.
C)Model parameters should be estimated using maximum likelihood methods.
D)Iterative estimation should be used.

Answer 2

A

Question 3

The assumption of linearity in logistic regression:

A)Assumes that there is a non-linear relationship between the logit of the predictor variables and the outcome variable.
B)Is the same as the assumption of linearity in linear regression.
C)Is tested by looking for a significant interaction between all pairs of predictor variables.
D)Assumes that there is a linear relationship between any continuous predictors and the logit of the outcome variable.

Answer 3

D

Question 4

Imagine we were trying to predict whether someone will pass (score = 1), or fail (score = 0) their driving test based on: (1) the number of hours of lessons they'd had; (2) whether or not they owned a car; (3) how many previous tests they had taken; and (4) their score on a test of spatial awareness. What analysis could we use on these data?

A)Multinomial logistic regression
B)Bimodal logistic regression
C)Binary logistic regression
D)Multiple regression

Question 5

Which of the following statements about the Wald statistic is false?

A)The Wald statistic assesses the individual contribution of a predictor to a logistic regression model.
B)The Wald statistic tends to be biased when the regression coefficient is large.
C)The Wald statistic is analogous to the t-statistic in linear regression.
D)The Wald statistic has a t-distribution.

Question 6

The ____ the value of the log-likelihood statistic, the ______ unexplained observations there are.

A)smaller, more
B)None of these are true.
C)larger, more
D)larger, fewer

Question 7

In the following table, how many mistakes are made when an animal is predicted to be a dog? $$\begin{array} { | l | l | l | } \hline & \text { Actually DoB } & \text { Actually Cat } \\\hline \text { Predicted DeB } & 89 & 8 \\\hline \text { Predicted Cat } & 4 & 55 \\\hline\end{array}$$

A)84
B)8
C)55
D)4

Question 8

The interpretation of the odds ratio, Exp(B), can be generalized to the population if:

A)The confidence interval of Exp(B) does not cross 0.
B)The confidence interval of Exp(B) does cross 1.
C)The confidence interval of Exp(B) does cross 0.
D)The confidence interval of Exp(B) does not cross 1.

Question 9

It is useful to compare a logistic regression model against some kind of baseline state. Which of the following is the baseline state that is usually used in logistic regression?

A)The model when all predictors and interactions are entered included
B)The model when only the constant is included
C)The mean of all scores
D)The log of all scores

Question 10

Multinomial logistic regression can be used on:

A)Ordinal predictor variables only.
B)Both categorical and continuous predictor variables.
C)Categorical predictor variables only.
D)Continuous predictor variables only.

Question 11

Complete separation is:

A)When the observed variance is bigger than expected from the logistic regression model.
B)When the outcome variable can be perfectly predicted by one variable or a combination of variables.
C)When the outcome variable is completely separate from all other variables.
D)When all predictor variables are perfectly uncorrelated.

Question 12

Which of the following sentences regarding logistic regression is false?

A)The linearity assumption is the same as for linear regression.
B)In logistic regression we assume the same things as ordinary regression.
C)If the outcome variable can be predicted perfectly from one predictor variable (or a combination of predictor variables) then we have complete separation.
D)Violating the assumption of independence can cause overdispersion.

Question 13

A researcher collected some data to look at tiredness (measured on a 10-point scale ranging from 0 (I'm wide awake) to 10 (I can't keep my eyes open) as a predictor of the following mood states (people were classified into each mood based on a diary): more depressed than normal (score = 2), more anxious than normal (score = 1) or normal (score = 0). Which of the following tests would be the most appropriate method for analysing these data?

A)Loglinear analysis
B)Multiple regression
C)Multinomial logistic regression
D)Binary logistic regression

Question 14

In logistic regression, a classification table:

A)Indicates how well a model has predicted group membership.
B)Allows you to check that your outcome variable has been coded correctly.
C)Classifies levels of your model in terms of statistical significance.
D)None of these.

Question 15

In logistic regression, what is the R-statistic?

A)It is the partial correlation between the outcome variable and each of the predictor variables.
B)It is the semi-partial correlation between the outcome variable and each of the predictor variables.
C)It is the polychoric correlation between the outcome variable and each of the predictor variables.
D)It is the correlation between the outcome variable and each of the predictor variables.

Question 16

With regard to logistic regression, which of the following sentences about the R-statistic is false?

A)The R-statistic is by no means an accurate measure and should be treated with some caution.
B)A negative value of the R-statistic implies that as the predictor variable decreases, the likelihood ratio of the outcome occurring decreases.
C)The R-statistic is the partial correlation between the outcome variable and each of the predictor variables.
D)The R-statistic can vary between -1 and 1.

Question 17

Imagine we had a sample of 15 couples and we wanted to predict which of the couples are likely to still be together in 1 year's time. In an attempt to do this, we filmed each couple having a 5-minute discussion about a recent 'difficult issue' that had arisen in their relationship . How could we analyse these data?

A)Independent t-test
B)Logistic regression
C)Multiple regression
D)One-way repeated-measures ANOVA

Question 18

Unlike multiple regression, logistic regression:

A)Does not have b weights.
B)Is not open to sources of bias.
C)Log-transforms the predictor variables.
D)Predicts a categorical outcome variable.

Question 19

Overdispersion:

A)All of these are correct.
B)Tends to limit standard errors.
C)Doesn't affect the model parameters (b-values) .
D)Biases our conclusions about the significance and population value of the model parameters.

Question 20

In multinomial logistic regression:

A)There must be more than one nominal predictor variable.
B)All predictors must be log-transformed.
C)The dependent variable is categorical with three or more levels.
D)There is more than one predictor variable.