The reading and practice we conducted in the last unit forregression easily covers the vast majority of
Question:
The reading and practice we conducted in the last unit forregression easily covers the vast majority of the regressionanalyses we typically need to conduct with healthcare data. TheKros and Rosenthal reading in this unit (Chapter 14) deals with themost common exception to that guideline, and it's an exception thatwe see more and more as we dive deeper into a broader array ofoutcome variables, many of which take only two (dichotomous) states(e.g. living/dead, admitted/discharged, success/failure,positive/negative). When these variables are our outcome ordependent variable, Kros and Rosenthal offer us a means tosuccessfully conduct regression analysis when the data doesn'tresolve well using chi-square analysis.
Consider the differences between the authors' use of OLS and WLSin their immunization example:
IMMUN = -.09035 + .039752(Age) +.17968(Educ) -.17968(Order) -- using Ordinary Least Squares (Figure14.4)
vs.
IMMUN = -.02844 + .00953(Age) +.105965(Educ) -.10994(Order) -- using Weighted Least Squares (Figure14.6)
The actual Regression Statistics from both runs are the same,including the 58% coverage (R Square). The authors explain why theR Square value requires some adjustment in the weighted situation,and provide the formulation for making that adjustment.Fortunately, this analysis usually ends up being embedded in ourstatistical tools, so we don't confront these issues actively mostof the time.
Discuss how you would explain what is happening in this arena toa manager or other co-worker who isn't versed in statistics. Whywould or should a statistical layperson care which of these basicmethods of analysis get used? What advantages are gained by themore involved analysis?
Auditing Cases An Interactive Learning Approach
ISBN: 978-0132423502
4th Edition
Authors: Steven M Glover, Douglas F Prawitt