Deck 16: Predictive Analytics Ii: Logistic Regression, Discriminate Analysis,

Question 1

The natural logarithm of the odds is called the logit.

Answer 1

True

Question 2

Linear discriminate analysis does not assume that for each particular category being considered, the joint probability distribution of the predictor variables is a multivariate normal distribution.

Answer 2

False

Question 3

The penalty weight used in the neural network models controls the trade-off between overfitting and underfitting.

Answer 3

True

Question 4

It is not possible to apply a logistic regression using a neural network algorithm.

Question 5

A regression technique for analyzing large data sets is neural network modeling.

Question 6

Forward selection with simultaneous validation is when one predictor variable is added at a time to a model where the variable added to the model has the smallest p-value when added to the model.

Question 7

Linear discriminate analysis can classify accurately even when the assumptions are not valid.

Question 8

A major drawback of neural network modeling is that its parameters are usually uninterpretable.

Question 9

In logistic regression the goal is to predict the true value of the independent variable.

Question 10

The odds of an event occurring is the probability that the event will not occur divided by the probability that the event will occur.

Question 11

The logistic regression equation involves the exponential function and it estimates the probability that an observation described by a specified set of predictor variable values will fall into a particular class.

Question 12

A neural network is nonlinear.

Question 13

A dummy variable is a continuous variable.

Question 14

A single-layer perceptron neural network model consists of only one layer called the hidden layer.

Question 15

The confusion matrix helps you assess your model's accuracy and avoid overfitting.

Question 16

Linear regression requires the assumptions of independence, linearity, normality, homogeneity of variance and non-multicollinearity.

Question 17

If the outcome variable is quantitative and all explanatory variables take values 0 or 1, a logistic regression model is appropriate.

Question 18

To avoid overfitting in the neural network model, the parameter estimates that are used minimize the least squares criterion.

Question 19

The AIC never penalizes the model for the number of model coefficients; therefore we do not favor a minimum AIC value.

Question 20

Neural network modeling represents the response variable as a linear function of the predictor variables.

Question 21

In logistic regression, ________ is used as a goodness-of-fit test.

A) deviance
B) Pearson
C) Hosmer-Lemeshow
D) All of the choices are correct.

Question 22

Neural networks are common for large data mining projects to analyze a data set involving millions of observations.

Question 23

After calculating the squared differences in linear discriminate analysis, you can estimate the probability of how that specific observation fares with regard to the categorical outcome.

Question 24

In linear discriminate analysis we calculate the means of the predictor variable values for the observations belonging to both categories that are in the data set.

Question 25

In logistic regression,

A) the dependent variable is continuous.
B) the dependent variable is divided into two equal subcategories.
C) the dependent variable is categorical.
D) there is no dependent variable.

Question 26

In logistic regression, deviance is a measure of difference between a ________ model and the ________ model.

A) saturated, fitted
B) fitted, saturated

Question 27

The dependent variable in logistic regression can take on values of 0 and 1.

Question 28

The probability of a new employee passing a test is .20. What are the odds of the employee passing the test?

A) .20
B) .25
C) .80
D) .75

Question 29

When we compare the squared differences in linear discriminate analysis we are looking for the larger value to predict the outcome of that data point.

Question 30

The odds of an event occurring is the probability that the event will occur divided by the probability that the event will not occur.

Question 31

Since the neural network model employs many parameters, there is a danger that we will overfit the model.

Question 32

In linear discriminate analysis we compare squared differences to see where a particular data point lies in relation to the categorical outcome.

Question 33

Squared distances can be used in two different ways in linear discriminate analysis.

Question 34

Which of the following methods do we use to best fit the data in logistic regression?

A) Least square error
B) Maximum likelihood
C) Exponential decay
D) Both least square error and exponential decay

Question 35

The ________ regression method is used when the response variable is a qualitative or a categorical variable.

A) quadratic
B) logistic
C) multiple
D) simple

Question 36

Which of the following tests can be used to assess whether the logistic regression model is a good fit?

A) Homer-Lemeshow test
B) Pearson
C) ROC curve
D) All of the choices are correct.

Question 37

In logistic regression, the goal is to find

A) the actual value of the dependent variable.
B) the odds of the dependent variable.
C) the log of the dependent variable.
D) None of the choices are correct.

Question 38

A soccer player takes a shot on goal 15 times and scores a goal 3 of those attempts. What are the odds of the soccer player scoring a goal?

A) 3/15
B) 5
C) 4
D) 3/12

Question 39

The hidden node function in neural networks is usually linear.

Question 40

Logistic regression is used when you want to

A) predict a categorical variable from continuous or categorical variables.
B) predict a continuous variable from categorical variables.
C) predict any categorical variable from other categorical variables.
D) predict a continuous variable from a categorical variable.

Question 41

Maximum likelihood estimation is an advanced statistical procedure which provides point estimates, and it is typically

A) ahead of the logistic regression.
B) a part of the logistic regression.
C) after the logistic regression to ensure statistical accuracy.
D) a simple process, completed to provide comparison estimates.

Question 42

Logistic regression analysis can answer what three major questions about a dataset?

A) Causal analysis, forecasting of an outcome, and trend forecasting
B) Causal analysis, odds of success, and trend forecasting
C) Odds of success, trend forecasting, and the confusion matrix
D) Maximum likelihood, causal analysis, and odds of success

Question 43

Which method will we use to best fit the data in logistic regression?

A) Maximum likelihood
B) Least effect likelihood
C) Least square error
D) Jaccard distance

Question 44

An example of a dummy variable is used in categorical data applications and might include

A) assigning the number 1 to Democrats and 2 to Republicans.
B) assigning the number 1 to the first place runner, and 2 to the second place runner.
C) assigning a 0 to students who can run a 10-minute mile, and a number 1 to students who run a 7-minute mile.
D) assigning a 1 to the probability of the customer receiving a $10 gift card and a 0 to the customer receiving a $20 gift card, when 10 of each denomination are being sent to a random sample of customers.

Question 45

One of the very good methods to analyze the performance of logistic regression is AIC, which is similar to R-squared in Linear Regression. Which of the following is true about AIC?

A) Model with minimum AIC value
B) Model with maximum AIC value
C) Both but depends on the situation
D) Both regardless of the situation

Question 46

In a neural network, the penalty weight controls the trade-off between ________ and ________.

A) overfitting; underfitting
B) overfitting; not fitting
C) underfitting; fitting
D) fitting; not fitting

Question 47

Deviance, Pearson and Hosmer-Lemeshow are all ________ that can be used in ________.

A) goodness of fit tests; logistic regression
B) predictive analysis tests; logistic regression
C) probability tests; neural network modeling
D) goodness of fit tests; neural network modeling

Question 48

You want to add new features to data to which you applied a logistic regression model and got a training accuracy of X and a testing accuracy of Y. Which options are correct, if you consider the remaining parameters are the same?

A) Training accuracy increases and testing accuracy increases or remains the same.
B) Training accuracy increases or remains the same.
C) Testing accuracy decreases and training accuracy remains the same.
D) Testing accuracy increases or remains the same and training accuracy decreases.

Question 49

What is the natural logarithm of the odds called?

A) The logit
B) The odds logarithm
C) The odds logit
D) The natural odds

Question 50

In a neural network model, the ________ is a nonlinear function of the linear predictor values.

A) response variable
B) null variable
C) predictor variable
D) quantitative response variable

Question 51

Which type of predictive analytic would be used to determine whether players on a football team, either historical or future, would perform successfully or unsuccessfully in a certain position on the field?

A) Linear discriminate
B) Logistic regression
C) Neural networks
D) Any of the choices could be used.

Question 52

Consider the following model for logistic regression: P (y = 1|x, h) = g(h₀ + h₁x) where g(z) is the logistic function. In the preceding equation the P (y = 1|x; h) , viewed as a function of x, that we can get by changing the parameters h. What would be the range of P in such case?

A) (0, 1)
B) (0, infinity)
C) (−infinity, 0)
D) (−infinity, infinity)

Question 53

In logistic regression, the difference between a fitted model and the saturated model can be measured with what test?

A) Deviance
B) Pearson
C) Hosmer-Lemeshow
D) All of the choices are correct.

Question 54

A logistic curve initially grows exponentially and

A) is theoretical.
B) is a summary of the probabilities.
C) is a consistent curve regardless of the data.
D) reflects the confidence intervals.

Question 55

What are used to avoid overfitting in the neural network model, and are used to minimize a penalized least squares criterion?

A) Parameter estimates
B) Least squares estimates
C) Probability estimates
D) Any of the choice could be used.

Question 56

The odds of an event occurring is: A. The probability that the event will occur divided by the probability that the event will not occur.
B. The probability that the event will not occur divided by the probability that the event will occur.
C. The probability that the event will occur divided by the total of the probability that it will occur, plus the probability of the event not occurring.

A) Option A
B) Option B
C) Option C
D) Either option A or B would create the same outcome.

Question 57

A ________ consists of three layers: input, hidden, and output.

A) single-layer perceptron neural network model
B) double-layer perceptron neural network model
C) triple-layer perceptron neural network model
D) neural network model

Question 58

If the odds of a student passing a class are .25, what is the probability that the student will pass the class?

A) .20
B) .25
C) .75
D) .80

Question 59

Parametric predictive analysis can be used to classify qualitative variables into several groups. The processes might include:

A) logistic regression.
B) simple regression.
C) vertical discriminate analysis.
D) stratification and clustering techniques.

Question 60

What is a major drawback of the neural network modeling?

A) The parameters are typically uninterpretable.
B) The model is not easily utilized with existing data sets.
C) The parameters minimize the least squares criterion.
D) This model only uses a single-layer perceptron.

Question 61

Even if a residual may be unusually large, the standardized residual rule might not identify the observation as being an outlier. How can this difficulty be circumvented?

A) by using studentized deleted residuals
B) by performing a logistic regression
C) through linear discriminate analysis
D) It cannot be circumvented and should be considered a component of the validation process.

Question 62

The chart provided shows the probability that a customer will use a coupon offered in a local paper by Lucky Shirts Store. Compute the odds ratio for the BankOne variable x₂ = 1, holding annual spending constant at x₁ = 4000. Annual Spending at Lucky Shirts Store

A) 1.3787
B) .4594
C) 1.4845
D) .3148

Question 63

Below gives the data concerning (1) the dependent variable Default which equals 1 if a customer defaults on their loan and 0 if they do not; (2) the independent variable Price of Home, which is the price of the home (in tens) and (3) the independent variable First Purchase which equals 0 if the customer has owned a home before and 1 if this is their first home. Identify and interpret the odds ratio estimate for First Purchase.

A) Odds ratio: 10.8675; a first-time home buyer is 10 times less likely to default than a buyer who has bought a home before.
B) Odds ratio: 10.8675; a first-time home buyer is 11 times more likely to default than a buyer who has bought a home before.
C) Odds ratio: 10.8675; a first-time home buyer is 10% times more likely to default than a buyer who has bought a home before.
D) Odds ratio: 10.8675; a first-time home buyer is 11% times more likely to default than a buyer who has bought a home before.

Question 64

Below gives the data concerning (1) the dependent variable Default which equals 1 if a customer defaults on their loan and 0 if they do not; (2) the independent variable Price of Home, which is the price of the home (in tens) and (3) the independent variable First Purchase which equals 1 if the customer has owned a home before and 0 if this is their first home. Estimate the probability that a buyer who spent $135,000 on their first home purchase will default.

A) .8529
B) .8501
C) .8674
D) .8754

Question 65

Below gives the data concerning (1) the dependent variable Default which equals 1 if a customer defaults on their loan and 0 if they do not; (2) the independent variable Price of Home, which is the price of the home (in hundred tens) and (3) the independent variable First Purchase which equals 0 if the customer has owned a home before and 1 if this is their first home. Evaluate the significance of each independent variable in the model.

A) Intercept = −3.4592 significant, price of home = .70935 significant
B) Intercept = −3.492 not significant, first purchase of home = .70935 not significant
C) Price of home is significant p-value = .0201, first purchase is significant p-value = .0136
D) Price of home is significant p-value = .0136, first purchase is significant p-value = .201

Question 66

Below gives the data concerning (1) the dependent variable Default which equals 1 if a customer defaults on their loan and 0 if they do not; (2) the independent variable Price of Home, which is the price of the home (in tens) and (3) the independent variable First Purchase which equals 1 if the customer has owned a home before and 0 if this is their first home. Estimate the probability that a buyer who spent $435,000 on their first home purchase will default.

A) .99636
B) .9899
C) .9768
D) .9695

Question 67

Below gives the data concerning (1) the dependent variable Default which equals 1 if a customer defaults on their loan and 0 if they do not; (2) the independent variable Price of Home, which is the price of the home (in tens) and (3) the independent variable First Purchase which equals 0 if the customer has owned a home before and 1 if this is their first home. Identify and interpret the odds ratio estimate for Price of Home.

A) Odds ratio: 4.56789; for each addition $10 spent on a home, the odds of a person defaulting increased 4.56%
B) Odds ratio: 4.56789; for each addition $100 spent on a home, the odds of a person defaulting increased 4.56%
C) Odds ratio: 4.56789; for each addition $10 spent on a home, a default is 4 times more likely
D) Odds ratio: 4.56789; for each addition $1000 spent on a home, a default is 4 times less likely for each additional $1,000 spent

Question 68

The chart provided shows the probability that a customer will use a coupon offered in a local paper by Lucky Shirts Store. Using this information, compute the odds that a customer will use the coupon, who spends $4,000 and does NOT bank at BankOne. Annual Spending at Lucky Shirts Store

A) .4594
B) .3148
C) 1.4845
D) .7742

Question 69

Below gives the data concerning (1) the dependent variable Default which equals 1 if a customer defaults on their loan and 0 if they do not; (2) the independent variable Price of Home, which is the price of the home (in tens) and (3) the independent variable First Purchase which equals 0 if the customer has owned a home before and 1 if this is their first home. Interpret the confusion matrix.

A) The misclassification rate for defaulters is .08 and the misclassification for non-defaulters is .1290.
B) The misclassification rate for defaulters is .08 and the misclassification for non-defaulters is .1482.
C) The misclassification rate for defaulters is .1290 and the misclassification for non-defaulters is .1111.
D) The misclassification rate for defaulters is .1482 and the misclassification for non-defaulters is .0870.

Question 70

What is a major drawback of neural network modeling?

A) The parameters are typically uninterpretable.
B) The parameters are typically interpretable.
C) The independent variables cannot be defined.
D) The level of significance is below .05, making the results invalid.

Question 71

Which predictive analytic technique would be best suited to an analysis of Facebook users, age 50-59, who purchase items on Amazon?

A) Neural network modeling due to the large data set with many observations.
B) Logistic regression, since the dependent variable is age.
C) Logistic regression, since the dependent variable is purchase items on Amazon, or does not purchase items on Amazon.
D) Either neural network modeling or logistic regression would work, depending on how close you want your p-value to .05.

Question 72

Below gives the data concerning (1) the dependent variable Default which equals 1 if a customer defaults on their loan and 0 if they do not; (2) the independent variable Price of Home, which is the price of the home (in tens) and (3) the independent variable First Purchase which equals 0 if the customer has owned a home before and 1 if this is their first home. Estimate the probability that a buyer who spent $678,700 on their first home will default.

A) .9832
B) .99636
C) .9998
D) .9345

Question 73

Below gives the data concerning (1) the dependent variable Default which equals 1 if a customer defaults on their loan and 0 if they do not; (2) the independent variable Price of Home, which is the price of the home (in hundred tens) and (3) the independent variable First Purchase which equals 0 if the customer has owned a home before and 1 if this is their first home. Identify the independent variables in this equation.

A) Price of Home and Default
B) Default
C) Price of Home, First Purchase, Default
D) Price of Home and First Purchase

Question 74

Neural networks have developed due to:

A) large amounts of data with huge numbers of observations.
B) increasingly unavailable data sets at reasonable costs for researchers.
C) work more effectively with data sets having fewer than 1000 observations.
D) have a clear and simple equation to use with very large data sets.

Question 75

Your supervisor has asked you to analyze all of the sales receipts for your company for the entire year to determine the average sales per customer, depending on whether the customer used a credit card, or another form of payment. What would your dependent variable be in this case and how would you label it?

A) Credit card = 1, other payment method = 0
B) Average sales = 1 and credit card = 0
C) Credit card = 1, cash = 2, debit card = 0
D) We cannot determine which variable is dependent from the information provided.

Question 76

Below gives the data concerning (1) the dependent variable Default which equals 1 if a customer defaults on their loan and 0 if they do not; (2) the independent variable Price of Home, which is the price of the home (in tens) and (3) the independent variable First Purchase which equals 1 if the customer has owned a home before and 0 if this is their first home. Estimate the probability that a buyer who spent $535,600 on their fourth home purchase will default.

A) .5842
B) .5674
C) .9834
D) .9465

Question 77

The number of dummy variables required if a categorical variable has k levels is:

A) k − 1
B) b.k
C) k + 1
D) !

Question 78

A confusion matrix helps:

A) To give misclassification rates of the model
B) To assign p-values for independent variables in the model
C) Both "To give misclassification rates of the model" and "To assign p-values for independent variables in the model"
D) None of the choices are correct.

Question 79

What is a method that can be used to minimize the penalized least squares criterion when using neural network modeling?

A) Fit the model 10-50 times using differing sets of randomly selected starter values to each fit.
B) Fit the model once with a predetermined starter value.
C) Use the set of parameters that yields the largest value of the penalized least squares.
D) Fit the model hundreds of times using differing sets of randomly selected starter values to each fit, to mimic using the entire population of data.

Question 80

What is a positive attribute of linear discriminate analysis?

A) The model is interpretable and prediction is easy.
B) It can be used with a constant p-value less than .05, so significance is implied.
C) It can be used in place of logistic and simple regression techniques in most cases.
D) It used the cubed distance between predictor variables, creating validation of the data.