Using Categorical Regression To Examine Long Term Care Ombudsman Data For
Predictors Of Institutional Abuse.
-
Kevin W. Borders
-
Kent School of Social Work
-
University of Louisville
-
Louisville KY 40292
-
(502) 896-6918
-
k.borders@Louisville.edu
-
-
Riaan van Zyl
-
Kent School of Social Work
-
University of Louisville
-
Louisville KY 40292
-
(502) 852-2430
-
mavanz01@gwise.Louisville.edu
The purposes of this oral paper presentation are two-fold: (i) to discuss
the categorical regression with optimal scaling data analysis method, a
relatively new promising technique of great relevance to social work research;
and (ii) to demonstrate the use of this method by reporting the results
of a study examining a six-state long term care ombudsman complaint database
for predictors of institutional abuse.
The regression with optimal scaling procedure considers categorical
variables to be quantified with numerical values, resulting in a linear
equation with transformed variables. The SPSS 10.0 CATREG procedure allows
nominal, ordinal and numerical variables to be scaled at the same time,
and considers the categorical variables the same as numerical variables.
Optimal scaling can be used to analyze categorical data when other standard
models do not perform well for data sets that feature too few observations,
too many variables, or too many values per variable. The last two difficulties
will be illustrated by analysis of the ombudsman database.
Ombudsmen represent and advocate for residents in long term care facilities
by investigating and resolving complaints given to them by family members
and friends, residents, facility staff, and others. Ombudsman data have
not been used to explore predictors of institutional abuse.
Data were collected by ombudsman programs in six states during 1996
(n=23,787 complaints). A model of facility and resident characteristics
accounted for 26% of predicted abuse. Facility characteristics of ownership
and beds (size of facility), and resident characteristics of race and,
to a lesser extent, pay status (resident ability to pay for care) were
the most influential predictors.
Optimal scaling allowed for the inclusion of a large number of data
points, but less variables so that regression analysis could
be used for the current model. Without optimal scaling, this level of data
interpretation would have been impossible.