3. The Stepwise method
Here is where John can use the forward selection, the backwards elimination, or a combination of both.
In the forward selection, John tries out the variables one by one (starting with none at all) and including those that are statistically significant. In the backward elimination, the reverse occurs; John starts with all variables, tests them for significance and eliminates those that lack significance. John can also amalgamate both methods for greater security.
This method is advantageous for a study that yields a large number of possible explanatory variables, but there is no underlying theory which John can use to base his selection. Given the huge amount of potential candidates, screening out the variables that are significant can be helpful to John. He may be able to see some pattern from doing this.
The problems include that John needs an appreciably large study to do this. Moreover, since several ANOVAs are used to determine the inclusion or exclusion of variables and since these ANOVAs are carried out on the same data you may have bias and multiple comparisons.
Advice to John
John would use the default Enter method if his study has not indicated anything about the importance of the order of the variables or of their relation to the constant.
If his research does indicate a certain order for warbles or importance of some above others, he would use the sequential method.
If he would like to screen his data, he would use the stepwise.
Discuss logistic regression. You may want to discuss such aspects as the logic of the method, the primary purposes of the method, the various steps involved, different methods of performing the method, and so on.
Logistic regression (LR) is used when the dependent variable is binary or ordinal. Lets say when you want to know if someone will live or die -- you want to know the odds...
This is binary (I..e either choice or more can occur - 1 or 2).
LR is useful for predicting whether something will, or will not happen. For instance, whether certain patients in a hospital may die from a certain disease. These are binary outcomes. LR is particularly useful when the dataset is large and the outcome is unpredictable and difficult to assess.
Advantages and disadvantage of logistic regression:
It is more robust than another system in that the dependent variables don't have to be normally distributed
It does not assume a necessary linear relationship between IV and DV
It can also deal with nonlinear effects
The DV need not be normally distributed
You can also add power terms
The independent needs not be interval
The independent need not be unbounded.
On the other hand, logistic regression requires that you have a lot of data in order to be able to obtain meaningful, consistent results. You need to have at least 50 data points! LR therefore is not for small studies.
Technicalities of logistic regression
The dependent variable in the LR is called the odds of the log variable or logit. Independent variables are regressed against this logit.
LR is different to standard regression in that the IVs regress against the logit rather than against the DV itself. Accordingly, therefore, to find the probability of an odd occurring given the huge amount of DVs, you have to convert the predicted logit back to a predicted probability. Various formulas are used for this.
Once interpreted, users will want to know the effect size (or "R value"). R close to 1 is a strong probability of even occurring, whereas R. close to 0 is a weak probability. Cox and Snell's R-Square is one of the most important measures of effect size for LR are:
