1: What about the omitted Satisfaction
trust item? Having left it out of the aggregated Satisfaction scale,
we may leave it out permanently. However, if it reflects something
uniquely important, the mere fact it does not want to “play
with” other items (in the sense that they do not want to group
together) does not necessarily mean it should be left out of further
analysis. You might include such “lone rider” items
as sole variables that - by virtue of their lack of ”willingness”
to group with other variables - are unique.
[return]
2:Having
said this, the use of this technique for ordinal variables is debated.
Some people prefer keeping ordinal variables as numerical data, although
it is more common to use the dummy variable technique. There also
exist some more complex, although not widely-used techniques that
adjust the regression results for ordinal variables. The beginner
will probably be safe with using dummy variables.
[return]
3: For instance, say you have a single
Likert scale item of the form 1 = strongly disagree, 2 = disagree,
3 = neutral, 4 = agree, 5 = strongly agree as an independent variable.
You could choose “neutral” as the reference variable,
and then have two dummy variables (one to represent the “disagree”
options, i.e. the “1” or “2”, and the
other to represent the “agree” options, i.e. the “4”
or “5”).
[return]
4: The suggested cut-off for Leverage
scores is ±2 p/n = ±2 (IVs + 1)/n where p is
the number of independent variables + 1 and n is
the sample size (Belsley, Kuh & Welsch, 1980).
[return]
5: In
SAS 9 go to Help > SAS Help and Documentation > SAS Products
> SAS Procedures and choose ROBUSTREG, or reference SAS/STAT 13.2
User’s Guide.
[return]
6: An example of a
methodological design that can mitigate missing data include setting
online survey tools to force answers (although this has the downside
of reducing response rates..
[return]
7: For
instance, Allison (2012); Dong & Peng (2013); Graham, Cumsille
& Elek-Fisk (2003); Graham & Hofer (2000); Little & Rubin
(2002); Rubin (1987); Schafer (1997); Schafer & Graham (2002);
Wayman (2003).
[return]
8: SAS has a great multiple imputation
offering in the combination of PROC MI and PROC MIANALYZE
[return]