General Public Health
Permanent URI for this collectionhttps://hdl.handle.net/20.500.12503/21761
Browse
Browsing General Public Health by Author "Aryal, Subhash"
Now showing 1 - 2 of 2
- Results Per Page
- Sort Options
Item COMPARISON OF FREQUENTIST AND BAYESIAN TOLERANCE INTERVALS VIA SIMULATION(2014-03) Arora, Shivani; Aryal, SubhashPurpose (a): To compare via simulations frequentist and Bayesian methods for estimating the shape parameter of a gamma distribution and apply the estimates to construct gamma tolerance interval. Methods (b): We generated data for various sample sizes (10, 20, 50 and 100) from gamma distribution with shape parameter = 0.25 to 7 (with an increment of 0.25). The scale parameter was held constant at 1. We obtained parameter estimate for the shape parameter via maximum likelihood method, method of moments and Bayesian approach. Next, we constructed 95% tolerance interval separately using each of the parameter estimates and evaluated the coverage probability. Results (c): All three methods failed to consistently provide 95% coverage. The coverage probability for the Bayesian approach was closer to 95% compared to the other two methods. For sample size less than 20, the coverage was close to 95% for the Bayesian approach and the method becomes progressively conservative when sample size becomes larger. This was observed consistently for all values of the shape parameter. Conclusions (d): Tolerance intervals are frequently used in environmental monitoring programs. Most monitoring programs for groundwater use samples of size 8 or 20 and our study shows that the Bayesian approach performs adequately for sample sizes less than 20.Item SAMPLE SIZE DETERMINATION IN MIXED-EFFECTS ZERO INFLATED POISSON LONGITUDINAL DATA(2014-03) Aryal, Subhash; Anne, Sruthi; Vemulapalli, AbhilashTo determine sample size for a mixed-effects zero-inflated Poisson regression model via simulation. Zero inflated Poisson data was simulated first and then a mixed-effects ZIP regression model was fitted to evaluate the significance of the time trend parameter using SAS software. Sample size was estimated to test the time trend parameter. Using simulation approach we determined sample size for testing both the Binomial and Poisson component separately as well as simultaneous testing of both the parameters. The results from Likelihood-Ratio-Test (LRT) indicate that different sample size estimates are required for the Binomial and Poisson components of model.We suggest zero inflated data can be best explained using ZIP model. It is recommended to use the larger of the two estimates from Binomial or Poisson model while designing any clinical study. Purpose (a): To determine sample size for a mixed-effects zero-inflated Poisson regression model via simulation. Methods (b): Zero inflated Poisson(ZIP) data was simulated first and then a mixed-effects ZIP regression model was fitted to evaluate the significance of the time trend parameter using SAS software. Sample size was estimated to test the time trend parameter.Results (c): Using simulation approach we determined sample size for testing both the Binomial and Poisson component separately as well as simultaneous testing of both the parameters. The results from Likelihood-Ratio-Test (LRT) indicate that different sample size estimates are required for the Binomial and Poisson components of model. Conclusions (d): We suggest zero inflated data can be best explained using ZIP model. It is recommended to use the larger of the two estimates from Binomial or Poisson model while designing any clinical study.