|
George Mason University |
|
Victoria Plamadeala |
|
Article I Estimating Parliamentary composition through electoral polls Frederic Udina Pedro Delicado Journal of the Royal Statistical Society: Series A (Statistics in Society), vol. 168, no. 2, 2005, pp. 387-99 The authors identify a problem in the Parliamentary composition (what political party gets how many seats in the Parliament) forecasts based on electoral polls in Spain. Independent polls conducted by newspaper agencies release biased forecasts of the final percentage of Parliamentary seats that competing parties would hold. To illustrate the bias, Uldina and Delicado simulate the year 2000 election results in Spain by generating 1000 samples of “election results” drawn from 52 (number of electoral districts) multinomial distributions. Each district is a multinomial distribution, because the votes are divided among 12 competing political parties. The sample sizes are targeted to match the sizes of various independent polls under scrutiny. When applying the forecasting rules of these independent polls to their simulated samples, they obtain a set of forecasted Parliaments that are different from the true election results of year 2000. This is the main way a Monte Carlo simulation experiment is used in this paper – to establish that there was bias in the forecast. Slight modifications of this simulation are used for a bootstrap method to correct this bias. Article II Computer-Intensive Methods for Tests about the Mean of an Asymmetrical Distribution Clifton D. Sutton Journal of American Statistical Association, Vol. 88, No. 423, pp. 802-810. An extensive Monte Carlo experiment was conducted to compare the power and type I error of tests about the mean of skewed distributions based on the Student’s t-test and the Johnson’s modified t-test. The simulation was conducted using 190 cases (10 sample size and 19 parent distribution combinations) at 40,000 random samples each. It was found that the Johnson’s modified t-test had truer type I error rate and was more powerful than the Student’s t-test. However, if the sample size is small and the skewness large the Johnson’s test can be inaccurate as well. The paper (Dr. Sutton’s paper) is an extensive Monte Carlo study evaluating the accuracy and power of the Student’s test vs. Johnson’s modified t-test about the mean of skewed distributions. Replicate the evaluation of the t-test and the Johnson’s modified t-test for positively skewed distributions: The study uses 19 parent distributions; perhaps cut down to main 10 skewed distributions; Use same variations (10 of them) of sample sizes; Generate 40,000 random samples for each of the10 x 10 cases (100 sample size and parent distribution combinations); Conduct a lower tail and an upper tail version of each test on each of the samples; Compare the type I error and the power of the tests at each version. |
