Quiz · Hard
Slope tests and intervals, ANOVA basics, multiple regression, multicollinearity, transformations, logistic regression and model checking.
Which transformation could straighten an exponential growth relationship?
Square x
Take the reciprocal of both
Take the log of y
Add a constant to y
If y grows by a constant percent, ln(y) is linear in x.
For ln(y) = 1.2 + 0.05x, each 1-unit increase in x changes y by about...
0.05 units
5%
50%
1.2%
A coefficient b changes y by approximately 100b% for small b.
In simple linear regression, the null hypothesis for the slope test is usually...
beta1 = 1
beta1 = 0
r = 1
b1 = 0
H0: no linear relationship in the population.
The degrees of freedom for the t test on the slope with n = 8 points is...
6
8
7
2
df = n - 2.
With b1 = 3.714 and SE = 0.389, the t statistic is about...
0.10
1.4
3.7
9.6
t = 3.714/0.389 = 9.56.
For b1 = 3.714, SE = 0.389 and t* = 2.447, the 95% interval for the slope is about...
(3.3, 4.1)
(0, 7.4)
(2.76, 4.67)
(3.4, 4.1)
Margin = 2.447 x 0.389 = 0.952; 3.714 ± 0.952.
A prediction interval for an individual y at a given x is...
Narrower than the confidence interval
Wider than the confidence interval for the mean response
The same as the confidence interval
Always centered at 0
It includes individual variation around the line.
SST = 617.5 and SSE = 38.07. R-squared equals...
0.938
0.062
0.50
0.969
1 - 38.07/617.5 = 0.938.
In multiple regression, a coefficient is interpreted as...
The correlation of that predictor with y
The p-value
The total R-squared
The change in predicted y per unit of that predictor, holding the others constant
Each partial slope controls for other predictors.
Adding a useless predictor to a multiple regression will...
Always lower R-squared
Raise adjusted R-squared
Never decrease R-squared, but may reduce adjusted R-squared
Make the F statistic 0
R-squared cannot fall; adjusted R-squared penalizes extra terms.
Strong correlation among predictors is called...
Heteroscedasticity
Multicollinearity
Extrapolation
Leverage
It inflates coefficient standard errors.
A categorical variable with 4 levels needs how many dummy variables?
3
4
2
1
Use k - 1 dummies, leaving one reference level.
A fan-shaped residual plot suggests...
Perfect linearity
A normal distribution
Multicollinearity
Non-constant variance
The spread of residuals changes with x.
Logistic regression is appropriate when the response is...
Binary (yes/no)
Continuous and normal
A count with no maximum
Ordered with 20 levels
It models the log-odds of the outcome.
In logistic regression, e raised to a coefficient is...
The probability
The slope of the line
R-squared
An odds ratio
It is the multiplicative change in odds per unit change in the predictor.
Adjusted R-squared for R-squared = 0.80, n = 30, k = 3 is about...
0.80
0.60
0.777
0.90
1 - 0.2(29/26) = 0.777.
The Spearman rank correlation is useful because it...
Requires normal data
Detects monotonic relationships and is less affected by outliers
Equals r^2
Measures causation
It is Pearson's r applied to ranks.
A significant slope but r-squared of 0.02 in a huge data set means...
The relationship is real but explains almost none of the variation
The model predicts very well
The slope is zero
There are no outliers
Statistical significance does not imply strong prediction.
Simpson's paradox happens when...
Residuals are normal
The correlation is exactly 0
Two predictors are identical
A trend in combined data reverses within subgroups
A lurking grouping variable can flip the overall association.