• The Mathematical Education
• /
• v.45 no.2 s.113
• /
• pp.145-154
• /
• 2006

The main purpose of this study is to investigate the effect of departures from normality and equal variance on the two-sample test when the variances are unknown. We have found that type I error brought about a little bit change which is ignorable in relation to kurtosis. But the change of type I error was mainly based on the skewness of the parent population. In introductory statistics classes where data analysis includes techniques for detecting skewness of two populations, we recommend the two-sample t-test when maximal skewness of two populations is smalter than the value 4 when the variances seem equal. Furthermore, our simulations reveal that the two-sample t-test appears somewhat more robust than that of z-test if the assumption of equal variance is satisfied. In the case of unequal variance, the two-sample t-test appears somewhat more robust provided the t-statistic using Satterthwaite's approximate degrees of freedom.

• Journal of the Korean Data and Information Science Society
• /
• v.9 no.2
• /
• pp.159-164
• /
• 1998

The two-factor nested variance component model with equal numbers in the cells are given by $y_{ijk}\;=\;{\mu}\;+\;A_i\;+\;B_{ij}\;+\;C_{ijk}$ and the confidence intervals for the ratio of variance components, ${\sigma}^{2}_{A}/{\sigma}^{2}_{B}$ are obtained in various forms by many authors. This article shows the probability ranges of these confidence intervals on ${\sigma}^{2}_{A}/{\sigma}^{2}_{B}$ proved by the mathematical computation.

• Journal of Korean Society of Industrial and Systems Engineering
• /
• v.5 no.7
• /
• pp.25-27
• /
• 1982

In most introductory statistics courses and text, the two equal tail test is presented without justification. In the present paper, the two equal tail critical region will be discussed in the light of unbiasedness with some test examples for the mean and the variance based on the random sample $X_1$, $X_2$,....$X_n$ from N($\mu$, $\delta^2$) using only elementary mathematics.

• Proceedings of the Korean Statistical Society Conference
• /
• 2002.11a
• /
• pp.327-332
• /
• 2002

Variance estimation for the regression estimator for a two-phase sample is investigated. A replication variance estimator with number of replicates equal to or slightly larger than the size of the second-phase sample is developed. In these cases, the proposed method is asymptotically equivalent to the full jackknife, but uses smaller number of replications.

• Journal of the Korean Statistical Society
• /
• v.22 no.1
• /
• pp.39-54
• /
• 1993

In the three-factor nested variance component model with equal numbers in the cells given by $y_{ijkm} = \mu + A_i + B_{ij} + C_{ijk} + \varepsilon_{ijkm}$, the exact confidence intervals of the variance component of $\sigma^2_A, \sigma^2_B, \sigma^2_C, \sigma^2_{\varepsilon}, \sigma^2_A/\sigma^2_{\varepsilon}, \sigma^2_B/\sigma^2_{\varepsilon}, \sigma^2_C/\sigma^2_{\varepsilon}, \sigma^2_A/\sigma^2_C, \sigma^2_B/\sigma^2_C$ and $\sigma^2_A/\sigma^2_B$ are not found out yet. In this paper approximate lower and upper confidence intervals are presented.

• Proceedings of the Safety Management and Science Conference
• /
• 2012.04a
• /
• pp.579-583
• /
• 2012

The Design of Experiment (DOE) is a most practical technique when establishing an optimal condition for production technology in Six Sigma innovation project. This research proposes the assessment of properties of error terms, such as normality, equal variance, unbiasedness and independence. The properties of six nonparametric ranking techniques for checking normality assumption are discussed as well as run test which is used to identify the randomness, and to check unbiased assumption. Furthermore, Durbin-Watson (DW) statistics and ARIMA (p,d,q) process are discussed to identify the serial correlation.

• Journal of applied mathematics & informatics
• /
• v.23 no.1_2
• /
• pp.479-488
• /
• 2007

A sequential procedure is proposed in order to construct one-sided confidence intervals for a normal mean with guaranteed coverage probability and ${\beta}-protection$ when the normal mean and variance are identical. First-order asymptotic properties on the sequential sample size are found. The derived results hold with uniformity in the total parameter space or its subsets.

• Journal of the Korean Statistical Society
• /
• v.25 no.4
• /
• pp.457-469
• /
• 1996

In this paper, we propose a method for an exact test of H : $p_i$ = $r_i$ for all i against K : $p_i$ $\neq$ $r_i$ for some i in an unbalanced random effect linear model, where $p_i$ denotes the ratio of the i-th variance component to the error variance. Then we present a method to test H : $p_i$ $\leq$ r against K : $p_i$> r for some specific i by applying orthogonal projection on the model. We also show that any test statistic that follows an F-distribution on the boundary of the hypotheses is equal to the one given here.

• The Korean Journal of Applied Statistics
• /
• v.32 no.5
• /
• pp.693-702
• /
• 2019

Deep learning has gained popularity for the classification and prediction task. Neural network layers become deeper as more data becomes available. Saturation is the phenomenon that the gradient of an activation function gets closer to 0 and can happen when the value of weight is too big. Increased importance has been placed on the issue of saturation which limits the ability of weight to learn. To resolve this problem, Glorot and Bengio (Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, 249-256, 2010) claimed that efficient neural network training is possible when data flows variously between layers. They argued that variance over the output of each layer and variance over input of each layer are equal. They proposed a method of initialization that the variance of the output of each layer and the variance of the input should be the same. In this paper, we propose a new method of establishing initialization by adopting truncated normal distribution and truncated cauchy distribution. We decide where to truncate the distribution while adapting the initialization method by Glorot and Bengio (2010). Variances are made over output and input equal that are then accomplished by setting variances equal to the variance of truncated distribution. It manipulates the distribution so that the initial values of weights would not grow so large and with values that simultaneously get close to zero. To compare the performance of our proposed method with existing methods, we conducted experiments on MNIST and CIFAR-10 data using DNN and CNN. Our proposed method outperformed existing methods in terms of accuracy.

• Communications for Statistical Applications and Methods
• /
• v.14 no.3
• /
• pp.633-640
• /
• 2007

We develop noninformative priors for the ratio of the lognormal means in equal variances case. The Jeffreys' prior and reference priors are derived. We find a first order matching prior and a second order matching prior. It turns out that Jeffreys' prior and all of the reference priors are first order matching priors and in particular, one-at-a-time reference prior is a second order matching prior. One-at-a-time reference prior meets very well the target coverage probabilities. We consider the bioequivalence problem. We calculate the posterior probabilities of the hypotheses and Bayes factors under Jeffreys' prior, reference prior and matching prior using a real-life example.