Projects

This page contains short descriptions of projects I have worked on, in reverse chronological order, roughly.  Some of my earlier projects are here.


 

Religious affiliation, education and Internet use

Using data from the General Social Survey, we measure the effect of education and Internet use on religious affiliation. We find that Internet use is associated with decreased probability of religious affiliation; for moderate use (2 or more hours per week) the odds ratio is 0.82 (CI 0.69–0.98, p=0.01). For heavier use (7 or more hours per week) the odds ratio is 0.58 (CI 0.41–0.81, p<0.001). In the 2010 U.S. population, Internet use could account for 5.1 million people with no religious affiliation, or 20% of the observed decrease in affiliation relative to the 1980s. Increases in college graduation between the 1980s and 2000s could account for an additional 5% of the decrease.

Comments: 12 pages, 1 figure, 3 tables
Subjects: Applications (stat.AP)
Cite as: arXiv:1403.5534 [stat.AP]
(or arXiv:1403.5534v1 [stat.AP] for this version)

 

Estimating the age of renal tumors

We present a Bayesian method for estimating the age of a renal tumor given its size. We use a model of tumor growth based on published data from observations of untreated tumors. We find, for example, that the median age of a 5 cm tumor is 20 years, with interquartile range 16-23 and 90% confidence interval 11-30 years.

Comments: 3 pages, 3 figures
Subjects: Applications (stat.AP); Tissues and Organs (q-bio.TO)
Cite as: arXiv:1203.6890 [stat.AP]
(or arXiv:1203.6890v1 [stat.AP] for this version)

A novel changepoint detection algorithm

We propose an algorithm for simultaneously detecting and locating changepoints in a time series, and a framework for predicting the distribution of the next point in the series. The kernel of the algorithm is a system of equations that computes, for each index i, the probability that the last (most recent) change point occurred at i. We evaluate this algorithm by applying it to the change point detection problem and comparing it to the generalized likelihood ratio (GLR) algorithm. We find that our algorithm is as good as GLR, or better, over a wide range of scenarios, and that the advantage increases as the signal-to-noise ratio decreases.

Comments: 11 pages
Subjects: Applications (stat.AP); Computation (stat.CO)
Cite as: arXiv:0812.1237 [stat.AP]
(or arXiv:0812.1237v1 [stat.AP] for this version)

Leave a Reply