Week Eleven

Question 0

What is Cook’s distance measure?

Answer 0

It shows the effect of each case on the fitted values for all cases in the model. If Cook’s distance is greater than 1 it must be investigated further.

Question 1

How can we identify outliers?

Answer 1

Via bloxplots, scatter graphs and residual plots after regression.

Question 2

What is a logarithm?

Answer 2

It explains how many of one number we need to multiply to get another number. It is the opposite of a power.

Question 3

What do we get if we take a log of y?

Answer 3

We log x - depending on the value of b (base).

Question 4

What value can base (b) take?

Answer 4

Any value that is positive, but not 1.

Question 5

What are the three particularly useful values of base?

Answer 5

1. Base 2
2. Base 10 - also known as the common log, denoted as log(x)
3. Base e - an irrational, number denoted as ln(x)

Question 6

What rule do some researchers use to keep zeros in when conducting a log?

Answer 6

Add one then take a log - ln(1) = 0

Question 7

What are logarithms useful for?

Answer 7

Compressing skewed variables and to interpret the coefficients in terms of percentage changes in X and Y rather than in units.

Question 8

What is the Ecological fallacy?

Answer 8

Where we make conclusions about individuals based on analyses of group data.

Question 9

What is the exceptional/individualistic fallacy?

Answer 9

Where we make conclusions about the group based on exceptional or individual cases.

Question 10

What is the log-log model?

Answer 10

LogY = log.a ➕ b.logX

Interpret the coefficients as the effect of proportional change in X on a proportionate change in Y.

Question 11

What is the Lin-log model?

Answer 11

Y = a ➕ b.logX

Effect of proportionate change in X on units of Y.

Question 12

What is the log-lin model?

Answer 12

LogY = a ➕ bX

The effect of an absolute change in X (in units) on Y in proportionate terms.

Question 13

How do we overcome the issue of interpreting a dummy variable when the dependent variable has been log-transformed?

Answer 13

We can take an antilog of dummy coefficients, subtract 1 from the variables and mutiny the difference by 100 to get the percentage change in y when a dummy goes from 0 to 1.

Question 14

What can interaction terms be?

Answer 14

A mixture of continuous and categorical, or just categorical data.

Question 15

What is hierarchical regression?

Answer 15

The practice of building successive, related linear regression models, each time adding more predictors.