EXAM PA – SOA QUESTIONS AND
ANSWERS
descriptive modeling - Correct Answers -understanding the relationships between
conditions and outcomes using modeling of past events
predictive modeling - Correct Answers -statistical models that use data mining and
probability analysis to foretell outcomes, where interpretability may not be as important
curse of dimensionality - Correct Answers -when the number of explanatory variables
(or number of factor levels) is large compared to the volume of the data
constant variance assumption - Correct Answers -at any value of x, the population of
potential error term values has a variance that does not depend on the value of x
OLS assumptions - Correct Answers -- regression model is linear, correctly specified,
and has additive error term
- error term has a zero population mean
- all explanatory variables are uncorrelated with the error term
- no serial correlation
- no heteroscedasticity/multicollinearity
- Error term is normally distributed
Heteroscedasticity - Correct Answers -when variances in y for the values of x are not
equal
multicollinearity - Correct Answers -when several independent variables are highly
correlated with each other, resulting in difficulty in estimating separate or independent
regression coefficients for the correlated variables
information gain - Correct Answers -a measure of the drop in impurity cause by a tree
split
variable importance - Correct Answers -a measure of how much a predictor contributes
to the overall fit of the model, it is calculated by aggregating across all trees in the
random forest the reductions in error that all splits on a selected predictor produce.
cost-complexity pruning - Correct Answers -growing a large tree and then pruning it
back by dropping splits that do not reduce the model error by a fixed value. the fixed
value determined by the cost complexity parameter is then optimized using cross-
validation for the lowest error
, descriptive - Correct Answers -focused on the past; aims to interpret observed trends by
identifying relationships between variables
predictive - Correct Answers -focused on the future; aims to make accurate predictions
prescriptive - Correct Answers -measuring the impacts of different decisions; aims to
answer "what if"s and "what is the best course of action" questions
What are two constraints of predictive models? - Correct Answers -- availability of easily
accessible/high quality data
- implementation issues (technical requirements/cost and effort)
random sampling - Correct Answers -randomly draw observations from the underlying
population withou replacement, where each record is equally likely to be sampled
stratified sampling - Correct Answers -dividing the underlying population into a number
of non-overlapping strata, then randomly sample from each stratum
sensitive variables - Correct Answers -predictors where differential treatment can lead
to unfair discrimination or raise equity concerns (ex. race, ethnicity, gender, age,
income, disability status)
proxy variables - Correct Answers -predictors that are closely related to prohibited
predictors (ex. occupation, geographical location)
target leakage - Correct Answers -when predictors in a model provide information about
the target variable that wouldn't be available when the model is deployed in practice
monotonic - Correct Answers -a steady increase or decrease in a feature along the x-
axis of a graph
Problem: difficult to separate out individual effects of different predictors on the target
variables. For GLMs, coefficients are widely varying in sign/magnitude, and are difficult
to interpret. - Correct Answers -- Drop one of the strongly correlated predictors.
- Use PCA to compress the correlated predictors into a few PCs.
Problem: Extreme values are exerting a disproportionate effect on model fit/distorting
visualizations. - Correct Answers -- Apply transformations to reduce right skewness
- Remove outliers that will unlikely have a material effect on the model
- Modify outliers to make them more reasonable (ex. change negative values to zero)
log transformation - Correct Answers -used to reduce right skewness, not for zero or
negative values
ANSWERS
descriptive modeling - Correct Answers -understanding the relationships between
conditions and outcomes using modeling of past events
predictive modeling - Correct Answers -statistical models that use data mining and
probability analysis to foretell outcomes, where interpretability may not be as important
curse of dimensionality - Correct Answers -when the number of explanatory variables
(or number of factor levels) is large compared to the volume of the data
constant variance assumption - Correct Answers -at any value of x, the population of
potential error term values has a variance that does not depend on the value of x
OLS assumptions - Correct Answers -- regression model is linear, correctly specified,
and has additive error term
- error term has a zero population mean
- all explanatory variables are uncorrelated with the error term
- no serial correlation
- no heteroscedasticity/multicollinearity
- Error term is normally distributed
Heteroscedasticity - Correct Answers -when variances in y for the values of x are not
equal
multicollinearity - Correct Answers -when several independent variables are highly
correlated with each other, resulting in difficulty in estimating separate or independent
regression coefficients for the correlated variables
information gain - Correct Answers -a measure of the drop in impurity cause by a tree
split
variable importance - Correct Answers -a measure of how much a predictor contributes
to the overall fit of the model, it is calculated by aggregating across all trees in the
random forest the reductions in error that all splits on a selected predictor produce.
cost-complexity pruning - Correct Answers -growing a large tree and then pruning it
back by dropping splits that do not reduce the model error by a fixed value. the fixed
value determined by the cost complexity parameter is then optimized using cross-
validation for the lowest error
, descriptive - Correct Answers -focused on the past; aims to interpret observed trends by
identifying relationships between variables
predictive - Correct Answers -focused on the future; aims to make accurate predictions
prescriptive - Correct Answers -measuring the impacts of different decisions; aims to
answer "what if"s and "what is the best course of action" questions
What are two constraints of predictive models? - Correct Answers -- availability of easily
accessible/high quality data
- implementation issues (technical requirements/cost and effort)
random sampling - Correct Answers -randomly draw observations from the underlying
population withou replacement, where each record is equally likely to be sampled
stratified sampling - Correct Answers -dividing the underlying population into a number
of non-overlapping strata, then randomly sample from each stratum
sensitive variables - Correct Answers -predictors where differential treatment can lead
to unfair discrimination or raise equity concerns (ex. race, ethnicity, gender, age,
income, disability status)
proxy variables - Correct Answers -predictors that are closely related to prohibited
predictors (ex. occupation, geographical location)
target leakage - Correct Answers -when predictors in a model provide information about
the target variable that wouldn't be available when the model is deployed in practice
monotonic - Correct Answers -a steady increase or decrease in a feature along the x-
axis of a graph
Problem: difficult to separate out individual effects of different predictors on the target
variables. For GLMs, coefficients are widely varying in sign/magnitude, and are difficult
to interpret. - Correct Answers -- Drop one of the strongly correlated predictors.
- Use PCA to compress the correlated predictors into a few PCs.
Problem: Extreme values are exerting a disproportionate effect on model fit/distorting
visualizations. - Correct Answers -- Apply transformations to reduce right skewness
- Remove outliers that will unlikely have a material effect on the model
- Modify outliers to make them more reasonable (ex. change negative values to zero)
log transformation - Correct Answers -used to reduce right skewness, not for zero or
negative values
