Preparing the Input Variables

Question 1

What does MCAR refer to?

Answer 1

missing completely at random

Question 2

What is missingness?

Answer 2

the probability that a value is missing

Question 3

What should you ask about missingness in your dataset

Answer 3

Is the missingness dependent on the data?

Question 4

What are lurking inputs?

Answer 4

unobserved variables affecting the probability that a value is missing

Question 5

What does it mean to select only cases that have no missing values?

Answer 5

Complete case analysis

Question 6

Why is complete case analysis the default behavior in PROC LOGISTIC?

Answer 6

using complete case analysis biases your inferences the least

Question 7

When is complete case analysis appropriate?

Answer 7

When you have a very small percent missing, and the values are missing completely at random

Question 8

Why is complete case analysis not a good choice for predictive modeling?

Answer 8

small numbers of missing values can cause an enormous loss of data in high dimensions, the probability for missing values is high, and those values are unlikely to be missing by random chance, and scorability (the model won’t be able to score any new cases with missing data)

Question 9

What is imputation?

Answer 9

The process of replacing missing values with reasonable substitutes

Question 10

What type of imputation is best for a binary variable?

Answer 10

The Median

Question 11

What is the hot decking method of imputation?

Answer 11

This method sorts all the cases in the sample by the values of several variables. A missing value is then imputed by taking the value from the case that is closest to it.

Question 12

What purpose does a missing value indicator variable serve?

Answer 12

they may be used to detect the relationship between the missingness and the target.

Question 13

Imputing a numeric using a median or mean value is generally effective when the missing values of a numeric input represent no more than _____% or all the input’s values.

Answer 13

50

Question 14

Name three important goals when handling missing values for predictive modeling

Answer 14

retain all the original data for model development, score all new cases, capture the relationship of missingness with the target.

Question 15

What quasi-complete separation?

Answer 15

Quasi-complete separation occurs when a level of the categorical input has a target event rate of either 0% or 100%

Question 16

What does it mean to collapse categories using the thresholding method?

Answer 16

This method requires a minimum number of cases in a level in order to create a dummy code input for that level. Any class value that doesn’t meet the criteria is collapsed into a new ‘Other’ category

Question 17

What PROC might you run to help determine your threshold?

Answer 17

PROC FREQ

Question 18

Which of the following statements is true regarding Greenacre’s method for collapsing the levels of contingency tables?

 a. The method is appropriate for any categorical input.
 b. The method does not account for the sample size in each level.
 c. Levels with similar marginal response rates are merged.
 d. At each step, the levels that give the largest reduction in the chi-square statistic are merged.

Answer 18

C