Dear All,
I am new to ordinal logistic regression. Using ordinal regression within the R
Commander GUI, I have generated an independent variable that is significant,
but whose 95% confidence intervals slightly crosses "1". Is this
possible? Here is the syntax and output:
polr(formula = CDIcat ~ Employment, data = CDIallvariables, Hess = TRUE,
method = "logistic")
Coefficients:
Value Std. Error t value
Employment[T.Parttime] -0.2523 0.2289 -1.102
Employment[T.fulltime] -0.4425 0.2484 -1.781
Intercepts:
Value Std. Error t value
0|1 -6.2182 1.0045 -6.1903
1|2 -1.9272 0.1480 -13.0213
2|3 -0.9892 0.1177 -8.4068
3|4 -0.4036 0.1094 -3.6880
4|5 -0.0801 0.1078 -0.7426
5|6 0.3187 0.1088 2.9297
6|7 1.0335 0.1196 8.6445
7|8 1.3607 0.1290 10.5445
8|9 2.3018 0.1758 13.0909
9|10 6.0283 1.0028 6.0116
Residual Deviance: 1960.430
AIC: 1984.430
(1 observation deleted due to missingness)
When I convert this to OR's ( using the Epicalc package) here is the
problem I have with the variable "Employment[T.fulltime]" which is
significant ,but has too wide a confidence interval. How would I interpret
this ?
ordinal.or.display(OrdRegModel.5)
Ordinal OR lower95ci upper95ci P value
Employment[T.Parttime] 0.777 0.496 1.217 1.36e-01
Employment[T.fulltime] 0.642 0.395 1.045 3.78e-02
Russell " Skip" Barbour Ph.D.
Associate Director for Statistics
Center for Interdisciplinary Research on AIDS
Yale School of Medicine
135 College St. Suit 200
New Haven , CT. 06510
Tel: 203 764 4332
Fax: 203 764 4353
Email: russell.barbour@yale.edu<mailto:russell.barbour@yale.edu>
Doubt grows with knowledge.
Johann Wolfgang von
Goethe<http://www.brainyquote.com/quotes/quotes/j/johannwolf380192.html>
[[alternative HTML version deleted]]
