To determine crude associations between the dynamic measures of vision and the ADVS outcomes, kernal smooth density plots with lowess running-line smoothers were created (Stata, ver. 7; StataCorp, College Station, TX). Lowess, or locally weighted scatterplot smoothing, is a robust nonparametric regression approach to fitting a line to a scatterplot.
19 The lowess method acts to smooth out random noise, which more easily allows the examination of trends.
Next, because the ADVS scores were severely skewed, the scores were trichotomized into the 60% highest (the least difficulty), 25% middle (some difficulty), and 15% lowest (the most difficulty) scores, similar to Rubin et al.
3
Then, polytomous logistic regression was used (the catmod procedure in SAS, ver. 6.12; SAS, Cary, NC) to determine the log odds of some difficulty or the most difficulty compared with little difficulty by measures of dynamic vision while adjusting for demographics, health factors, and other measures of visual function. The measures of vision were examined in a continuous fashion, although signs of nonlinearity were investigated by categorizing the vision measures, creating dummy variables, and examining the estimates for signs of nonlinearity. Factors included as confounders included age, sex, race, education, cognitive status, depressive symptoms, and general health status. These variables were included because it was decided a priori that they might be associated with vision scores and report of visual difficulty on the ADVS.