The nonparametric estimate of survival as a function of an individual covariate is superimposed on the model's prediction of the same function in the graph below. Each point on the nonparametric function is the CJS estimate (or binomial estimate for known-fate data), for all data with covariate values within a window around the given point. The window size varies from a minimum of eight data points up to one-fifth of the total number of data points.
Because the window of data used to create the separate CJS estimates is small, there is considereable variability about the individual estimates. Furthermore, the estimates from the moving window are correlated, resulting in localized trends that may not actually exist. Also, the window size is smaller at the lower and upper boundaries of the covariate.
