- Prediction the value of one variable from another.
- You are in the head of line and there are three clerks each dealing with customer. What is the probability that of the three customers currently being served and yourself, you are the last one to finish buying a ticket?
- The training error decreases for smoothing splines as the tuning parameter lambda increases
- Shrinkage and Dimension Reduction are both methods with dealing with the problem of large bias that occurs in high dimensions
- The mean function for white noise is zero
- Which of the following pairs of distribution and link function is most appropriate to model if a person is hospitalized or notA. Normal Dist, Identity LinkB. Normal Dist, Logit LinkC. Binom Dist, Linear LinkD. Binom Dist, Logit Link
- For a data set containing one million observations, two models are tested. The models are nested, and the model with the fewest variables, model 1, has the smaller BIC, the larger AIC, the smaller R^2 and the larger adjusted R^2.On the basis of adjusted R^2, select model 1.
- AR(3) Process: Partial autocorrelation for lag 5 is always greater than zero
- For every time series, the acf at lag 0 is exactly 1
- AR(3) Process: Partial autocorrelation for lag three is always equal to zero
- Modeling Obstacles not treated by GLM: Modeling a non-linear relationship between the response mean and the predictors
- An acf that decays exponentially in height is characteristic of an autoregressive series
- For a large enough degree d, polynomial regression can produce an extremely flexible curve
- The weights for the Partial Least Squares score arise from the slopes of the single variable regressions of y onto each explanatory variable
- An acf with spike at lag k may indicate a repeating pattern of length k
- Modeling Obstacles not treated by GLM: Modeling a discrete response distribution in terms of a explanatory variable with a continuous distribution
- Estimating a value outside the range of measured data.
- To test a GLM with polynomial explanatory terms, only accept the model if both Type 1 and Type III Tests agree
- Regression works poorly in high dimensions because not all of the predictors are actually associated with the response
- Generalized Additive Models are sometimes referred to as parametric because no response distribution is assumed