Data Visualization and Data Summary - Text Mining Techniques for Healthcare Provider Quality Determination - page 41

Figure 7 appears to have the most optimal level of smoothness. For a bwm<1, the curve becomes
more jagged; for bwm>1, it becomes smoother. However, it can be too smooth. Figure 8 has a bwm
of 25. If the curve is too smooth, the depiction of the population shows more variability than actually
exists in the population. It is optimal to vary the bandwidth and then to choose the best representation
of the population.
Although the Central Limit Theorem will allow us to estimate the average length of stay for patients
with diabetes, the confidence limits require the assumption of a normal distribution. Without this as-
sumption, we can use numerical methods to estimate the integral. We can use a simple Simpson's Rule
Figure 7. Kernel density estimate of length of stay for patients with diabetes and BWM=10.
Figure 8. Kernel density estimate of length of stay for patients with diabetes and BWM=25.