Lecture 8_Different Analytical Approaches in MLR

Question 1

What are the three different analytical approaches in MLR?

Answer 1

•Simultaneous Approach
•Sequential, or Hierarchical Approach
•Stepwise, or Statistical Approaches
     – Forward Selection
     – Backwards Elimination

Question 2

What is the difference between Explanation and Prediction?

Answer 2

Explanation – emphasis on understanding
Prediction – emphasis on practical applications
(an ideal explanation allows prediction, but the reverse not always true)

Question 3

What analytical approaches are useful for Explanation and Prediction?

Answer 3

Simultaneous, and Sequential/ Hierarchical

Question 4

What analytical approach is useful for Prediction only?

Answer 4

Stepwise/ Statistical (forward and backward)

Question 5

Simultaneous Approach

Answer 5

• All IVs are entered at once (i.e., in one block).
• Yields estimates of each IV’s direct effect.
• Recall that a direct effect refers to the unique influence of a IV controlling for all other IVs in the model.
• Also known as “forced entry” approach
(This is the approach we have been using so far.)

Question 6

Sequential or Hierarchical Approach

Answer 6

• Requires/Allows the researcher to control the advancement of the regression process.
• Can provide information for explanatory purposes (such as identifying the contribution of one group of theoretically related variables relative to another group of variables)
• The IVs are assigned roles, differing in importance, by the researcher according to logic or theory.

Question 7

In the Sequential or Hierarchical Approach, how are explicit hypotheses tested about the relation between the DV & certain IVs while controlling for the influence of other IVs?

Answer 7

• put in confounding variables 1st then variable(s) of interest
• focus on the change in R² (ΔR²)

Question 8

Stepwise, or Statistical Approaches: Forward Selection

Answer 8

– the IV with the largest zero-orderrwith DV enters first.
– IVs entered sequentially based on size of nth-order semipartial r with the DV (guarantees next variable will increase R² the most).
– “data driven” meaning it proceeds based on the size of the correlations (zero-order, 1st-order semipartial, 2nd-order semipartial, etc.) of the IVs with the DV.
– Helpful for predictive model building, but not well suited for explanatory purposes.

Question 9

Stepwise, or Statistical Approaches: Backward Elimination

Answer 9

– begin with all IVs in equation
– IVs deleted sequentially based on size of nth-order semipartial r with DV (guarantees next variable will decrease R² the least).
– also data driven (but works in reverse)
– Helpful for predictive model building, but not well suited for explanatory purposes.