Note: This message is sent on behalf of Owen Murphy and replies should be
made to him at [log in to unmask]
---
Dear List members,
1. I need to pair then combine a number of predictor continuous variables, as a
prelude to Multiple Regression, and I would welcome advice on the main criteria
that need to be satisfied to ensure compatibility and avoidance of bias.
2. The variables in question are all counts (1,2,3, 4...), running from low to
high, and I propose to pair them on the basis that, in each pair, each of the
variables measures a different aspect of the broader, underlying construct (one
construct per pair, of course)
3. What other criteria, if any, should be applied? Can the degree of
compatibility between members of pairs best be measured by such means as
the level of correlation between each pair, or eg. the strength of loading on,
say, Component 1 in Principal Component Analysis, or it there a better way?
With advance thanks
Owen
|