Attewell Chapter 2

Question 1

Is the statement, “… [the] conventional statistical approach focuses on the individual coefficients
for the predictors, and doesn’t care as much about predictive power” true or false?

Answer 1

True

Question 2

Does the likelihood of finding a statistical significance increase as more predictors are entered into a model?

Answer 2

Yes! One in twenty predictors will be significant at p ≤ 0.05 through chance alone.

Question 3

How could a researcher solve the problem of multiplicity?

Answer 3

By using a Bonferroni correction

Question 4

How does the Bonferroni correction for multiple comparisons work?

Answer 4

By simply dividing the conventional value of

0.05 by the number of predictors (If there are 5 predictors, then the new significance threshold would be 0.01)

Question 5

In what field is the problem of multiplicity especially prominent?

Answer 5

The problem of multiplicity has grown evermore acute in the medical research and analyses of gene sequences, because it is increasingly common for thousands of significance tests to be tried, before reporting the significant ones.

Question 6

How does the Data Mining approach avoid the multiplicity problem entirely?

Answer 6

By using a form of replication known as Cross-Validation

Question 7

When are “residuals” said to be “homoscedastic”?

Answer 7

When they are normally distributed, with a constant variance and a mean of zero, and are independent of one another

Question 8

What does the Greek word “homoscedastic” mean?

Answer 8

Having equal variances

Question 9

What does the Greek word “heteroscedastic” mean?

Answer 9

Having unequal variances

Question 10

Does Data Mining provide ways for circumventing the problem of heteroscedasticity?

Answer 10

Yes

Question 11

Are many Data Mining methods nonparametric?

Answer 11

Yes, because, “…they do not require the kinds of statistical assumptions about the distribution of error terms that underlie many conventional modeling methods”.

Question 12

Is bootstrapping a type of nonparametric technique?

Answer 12

Yes

Question 13

Does bootstrapping make assumptions about the shape of the sampling distribution?

Answer 13

No

Question 14

Does significance testing play a crucial role in the conventional statistical approach?

Answer 14

Yes

Question 15

When do the most serious problems with significance testing occur?

Answer 15

When modelers add many predictors to models, and especially when they search through hundreds of predictors before deciding which to include in a model

Question 16

Can Data Mining models outperform conventional statistical models?

Answer 16

Yes, they mostly due, because many conventional statistical models neglect interactions between predictors.