Wage Determinants Analysis by Quantile Regression Tree

Chang, Young-Jae The Korean Statistical Society 2012 Communications for statistical applications and me Vol.19 No.2

Quantile regression proposed by Koenker and Bassett (1978) is a statistical technique that estimates conditional quantiles. The advantage of using quantile regression is the robustness in response to large outliers compared to ordinary least squares(OLS) regression. A regression tree approach has been applied to OLS problems to fit flexible models. Loh (2002) proposed the GUIDE algorithm that has a negligible selection bias and relatively low computational cost. Quantile regression can be regarded as an analogue of OLS, therefore it can also be applied to GUIDE regression tree method. Chaudhuri and Loh (2002) proposed a nonparametric quantile regression method that blends key features of piecewise polynomial quantile regression and tree-structured regression based on adaptive recursive partitioning. Lee and Lee (2006) investigated wage determinants in the Korean labor market using the Korean Labor and Income Panel Study(KLIPS). Following Lee and Lee, we fit three kinds of quantile regression tree models to KLIPS data with respect to the quantiles, 0.05, 0.2, 0.5, 0.8, and 0.95. Among the three models, multiple linear piecewise quantile regression model forms the shortest tree structure, while the piecewise constant quantile regression model has a deeper tree structure with more terminal nodes in general. Age, gender, marriage status, and education seem to be the determinants of the wage level throughout the quantiles; in addition, education experience appears as the important determinant of the wage level in the highly paid group.

Tumor Regression Patterns Based on Follow-up Duration in Patients With Head and Neck Squamous Cell Carcinoma Treated With Radiotherapy or Chemoradiotherapy

공문규,홍성언 대한이비인후과학회 2015 Clinical and Experimental Otorhinolaryngology Vol.8 No.4

Objectives. We describe patterns of tumor regression based on follow-up duration after radiotherapy (RT) or chemo-RT in patients with head and neck squamous cell carcinoma. Methods. Thirty-one patients with head and neck squamous cell carcinoma were included in this study and received definitive RT or chemo-RT. The pattern of primary tumor regression after treatment was evaluated every 1 to 2 months. Predictive factors for the length of time to full regression were also analyzed. Results. Among all patients, 27 patients showed regression of the primary tumor, 24 patients showed >50% regression, and 15 patients showed total regression. The primary tumor gradually regressed during the course of follow-up. The median time to full regression was 5.2 months (range, 1.3 to 17.9 months). In the 24 patients who showed >50% regression, the rate of >50% regression increased over time as follows: 25.0% at 1 month, 62.5% at 2 months, 75.0% at 3 months, 91.7% at 4 months, and 95.8% at 5 months. Higher total RT dose and shorter RT duration were associated with longer time to full regression. Conclusion. A substantial number of patients showed continuous regression of the primary tumor for more than 2 months after treatment. The timing for evaluation of tumor regression must be greater than 2 months from the completion of RT or chemo-RT in patients with head and neck squamous cell carcinoma.

Shapley value regression을 활용한 한국남자프로농구의 경기력 요인 분석

김봉석,최경호 한국스포츠학회 2018 한국스포츠학회지 Vol.16 No.1

프로스포츠의 경기력 관련 요인을 탐색함에 있어 그동안 많이 활용된 방법 중의 하나가 바로 회귀분석이다. 그런 데 회귀분석은 모형예측 등을 수행함에 있어서, 설명변수들에 대하여 독립성 가정이 요구되는 등 통계적인 측면에서 많 은 문제점이 있는 방법이기도 하다. 따라서 회귀분석을 실제로 적용함에 있어서는 이러한 문제점을 극복할 수 있는 대안 이 요구되는 바, 본 연구에서는 그 대안으로 Shapley value regression을 활용할 것을 제언하였다. 이를 위해 2016~2017 한국남자프로농구 데이터를 이용하여 Shapley value regression을 이용한 상대적 중요도를 구해 보았 다. 그 결과 회귀분석에서는 통계적으로 유의하지 않았던 ‘어시스트’나 ‘3점 성공률’ 그리고 수비 관련 변수에서는 ‘GD’ 등의 변수가 나름 상대적으로 중요한 것으로 나타났다. 본 연구는 프로농구 자료에 대하여 경기력에 영향을 미치는 변수 를 탐색함에 있어 기존의 회귀분석과는 다른 접근을 통해 보았다는점에서 의의가 있으며, 향후 프로스포츠 관련 양적 연구를 수행하는 연구자들에게 보다 정확한 결과도출 및 해석을 하는데 도움을 줄 수 있을 것으로 설명 할 수 있다. Regression analysis is one of the most popular ways to explore the performance related factors of professional sports. Regression analysis is a method that has many problems in terms of statistics, such as the need for an independent assumption about the described variables, in carrying out model predictions, etc. Therefore, when applying regression analysis in practice, alternatives to overcome these problems are needed, and the study used the Shapley value regression suggestion. To this end, data from the Korean Basketball League from 2016 to 2017 is used to identify the relative importance using Shapley value regression. As a result, offense-related variables such as ‘assists’ or ‘three-point field goal percentage’ and a defense-related variable including ‘GD’ were found to be relatively important, which were not statistically significant when using regression analysis. This study is meaningful in that it explores variables affecting the performance of professional basketball through a different approach from the existing regression analysis. It can also help researchers who perform quantitative research related to professional sports in the future to derive and interpret more accurate results.

고객반응성을 고려한 고객세분화 방법에 관한 연구 -농촌체험관광 만족도 영향요인을 중심으로-

서상윤 ( Sang Yun Seo ),이훈영 ( Hoon Young Lee ) 한국고객만족경영학회 2011 고객만족경영연구 Vol.13 No.1

본 연구에서는 시장세분화 방법으로써 Mixture Regression방법을 농촌체험관광에 적용하여 결과를 비교함으로써, 일반적으로 사용되는 K-means 방법에 대한 새로운 대안으로써 Mixture Regression방법에 의한 시장세분화 방법을 제시하고자 하였다. K-means과 Mixture Regression 방법을 적용하여 농촌체험관광객들을 세분하여 본 결과, 두 방법에 의해 도출된 결과는 서로 상이한 기술적, 통계적 특성을 나타내었다. 세분화된 집단의 수가 3개 이상인 경우 K-means방법에 의해 구분된 집단에서는 농촌체험관광 만족에 영향을 주는 체험품질요인들의 통계적 유의성이 낮아진 반면에, Mixture Regression에 의해 구분된 집단에서는 변수간의 통계적 유의성이 크게 낮아지지 않았다. 또한 모형의 설명력에 있어서도 수리적인 방법을 이용한 K-means 방법에 의해 형성된 집단의 모형 설명력은 상당히 저조한 반면에, Mixture Regression모형에 의해 형성된 집단의 모형 설명력은 상대적으로 높은 것으로 나타났다. 본 연구결과는, 단순히 시장을 기술적인 변수들을 이용하여 유사한 집단으로 구분하는 데는 K-means 분석방법을 이용한 세분화 방법이 비교적 용이하고, 효과적인 반면에, 세분화된 집단에서 시장반응성에 영향을 주는 요인들을 탐색하고, 이를 바탕으로 새로운 고객을 확보하는 데는 Mixture Regression을 이용한 방법이 보다 효과적임을 시사한다. 또한 실무적으로도 인구통계적, 경제사회적 기술변수들만을 이용하여 단순히 유사한 기술적 특성만을 가진 대상자들을 한데 묶음으로써 집단을 세분화 하는 경우에는 재구매도 의도나, 만족도, 구매빈도와 같은 반응성의 차이를 발견하기가 쉽지 않기 때문에, 시장세분화의 목적을 달성하기 위한 시장반응성의 차이, 즉 기준변수의 차이를 발견하기 위해서는 Mixture Regression 방법을 이용하는 것이 보다 효과적이라 할 수 있다. In this study, we suggest the mixture regression as a new method of segmenting a market. We discussed its benefits and shortcomings in comparison with typical K-means method. We applied both methods to segmenting the rural tourism market. Unlike K-means method, mixture regression method is a kind of objective oriented method. In this study, we conduct segmentation so as to classify the market into several segments in terms of the level of overall satisfaction. Variables measuring the quality of experience are used as the segmentation bases. The segmentation results of both methods are quite different. The basis variables are not significantly related with overall satisfaction in the segments made by K-means method. However, the variables in the segments made by mixture regression method are significantly related with the overall quality of experience. Since K-means method simply groups the tourists in terms of the similarity of basis variables, the variation of basis variables are not useful in predicting the level of overall satisfaction. K-means method does simply apply the Euclidean distance among variables without considering their associated relationship. However, the mixture regression method classify the market in terms of predictability of overall satisfaction, considering the association among segmentation variables. Thus, the segments done by the mixture regression method are easy to apply marketing strategy related with overall satisfaction. In sum, if there is an objective target variable, the mixture regression method is useful. Otherwise, the simple K-means method is valid for segmentation. However, since the segmentation is an marketing activity preceeding the marketing plan and strategy intended, the mixture regression method would be more appropriate than K-means method in general.

퍼지선형 회귀에서의 이상점과 영향력이 큰 관측점 검출

손방용 한국자료분석학회 2011 Journal of the Korean Data Analysis Society Vol.13 No.1

In general, when we have fitted the least squares regression line to obtained from usual linear regression model, the estimated results of least squares regression are misunderstood by irregular data, outliers and influential observations. On the other hand, there are two ways to inference of the fuzzy linear regression. One ways to obtain fuzzy estimate of regression parameters in regression model for imprecise input-out data. The other way to estimate least squares regression in fuzzy regression for fuzzy input-out data. Also, when irregular data exist in set of fuzzy data, fuzzy least square regression can be incorrectly interpreted. In this paper, we propose the method of detection of outliers and influential observations in fuzzy regression for fuzzy input-out data by using fuzzy residuals and fuzzy residuals deleted irregular points and fuzzy interval. Numerical example provided shows that fitted fuzzy least square regression fuzzy residuals and fuzzy intervals can be detected irregular fuzzy data. 선형회귀모형에서 최소제곱 회귀식을 적합시킬 때 자료에 포함된 이상점 또는 영향력이 큰 관측점 등 비정상적인 자료에 의해 회귀 추정량은 심하게 왜곡된다. 한편 퍼지자료의 선형함수관계를 추론하는 방법은 일반적으로 크게 두 가지이다. 하나는 비퍼지 입-출력 자료에서 퍼지 회귀계수를 추론하는 것이고, 또 하나는 퍼지 입-출력자료에서 퍼지 최소제곱회귀를 추론하는 것이다. 불행하게도 이들 퍼지 선형회귀의결과는 비정상적인 퍼지자료에 의해 매우 민감하며 나아가 왜곡된 퍼지 회귀식을 추론하게 한다. 본 논문에서는 퍼지 입-출력자료가 주어진 경우 퍼지 최소제곱회귀에서 이상점과 영향력이 큰 관측점을 검출하고 이들을 제거한 퍼지 최소제곱 회귀계수를 추론하여 이들 비전상적인 퍼지자료에 저항력을 지닌 안정된 퍼지 최소제곱회귀를 제안한다.

자연과학편 : 다중공선성 상태의 주성분회귀와 능형회귀

김진욱(JinUkKim) 한국체육학회 2006 한국체육학회지 Vol.45 No.4

회귀분석에서 설명변수들 사이에 선형적이거나 또는 거의 선형적인 종속관계에 있는 경우를 다중공선성이라고 하는데 이 같은 결과를 왜곡시키는 공선성이 존재하는 경우에는 일반적인 최소제곱 추정량의 대안을 고려해야만 한다. 대개 연구자들은 그 대안으로 주성분회귀나 능형회귀를 많이 사용한다. 본 연구의 목적은 다중공선성이 존재하는 경우의 두 회귀방법을 제시하고자 시뮬레이션 자료와 실제 자료를 사용하였다. 생성된 자료는 세 개의 설명변수와 하나의 반응변수를 가지는 다중회귀분석 모형을 설정하였고 설명변수 두 개는 서로 관련이 있도록 하였고 실제 자료는 유산소성 체력 자료를 이용하였다. 이 두 가지 모형으로 일반최소제곱, 주성분회귀, 능형회귀의 각 계수들을 추정하였다. 이론적인 측면을 같이 고려하면서 두 자료의 타당한 추정치를 얻고자 분석한 결과 첫 번째 자료는 능형회귀분석이, 두 번째 자료는 주성분회귀분석이 만족스러운 결과를 나타냈다. 다중회귀분석시 다중공선성이 존재하는 자료에서 회귀계수의 타당한 추정치를 얻고자 한다면 주성분회귀와 능형회귀가 권장된다. In regression analysis the alternatives of OLS (ordinary least squares) method for estimating regression coefficients should be considered if there is a severe multicollinearity among explanatory variables. It means that the multicollinearity distorts the results of study. Researcher, therefore, usually use prevailing two alternative methods, which are principal components regression (PCR) and ridge regression (RR). The purpose of this study is to propose the two alternative regression methods in a case of the multicollinearity problem and to give some discussions of the methods. For this study, two regression model were considered. One model contains three explanatory variables, two of three variables of highly correlated, and one response variable, by explanatory variables, which were created under the assumptions of regression analysis. The other model was oxygen consumption analysed by SAS OnlineDoc. From these variables, OLS, PCR and RR estimators were estimated and their variations were examined by means of ridge trace. PCR and RR showed results with satisfaction in a case of multicollinearity of a multiple regression.

선천성 사경 환자에서 목빗근 병변의 부피 또는 물리치료와 소멸 소요 시간 사이의 연관성

김성진,김동욱 대한영상의학회 2013 대한영상의학회지 Vol.68 No.5

Purpose: This study aimed to evaluate any correlation between the volume of a lesion or physical training, and regression time, in congenital torticollis patients. Materials and Methods: From January 2007 to December 2010, 63 infants and young children underwent neck ultrasound (US), as congenital torticollis was clinically suspected. We statistically analyzed the correlation between the volume of a lesion or physical training and the regression time, when successful regression was defined as no visible lesion on follow-up US, or no palpation of the lesion on clinical follow-up. Results: Among 63 patients, 48 were sonographically diagnosed with congenital torticollis: normal delivery (n = 39) and c-sec (n = 9). Among them, 24 were clinically followed up, and 22 showed complete regression. Among 24 patients, 9 underwent physical training, and 7 showed complete regression after physical training. There was a negative correlation between the volume of a lesion and the regression time (p = 0.011), but there was no statistical significance between physical training and the regression time (p = 0.15). Conclusion: In congenital torticollis patients, the volume of a lesion was in reverse proportion to the regression time, and physical training may be unhelpful for a decrease in regression time. 목적: 저자들은 선천성 사경 환자에서 목빗근(sternocleidomastoid muscle) 병변의 부피 또는 물리치료 여부와 성공적인 소멸의 소요 시간(regression time) 사이의 연관성을 알아보고자 하였다. 대상과 방법: 2007년 1월부터 2010년 12월까지 임상적으로 사경이 의심되었던 63명의 환자에서 한 명의 영상의학과 전문의가 경부 초음파를 시행하였으며, 실시간 초음파로 병변의 직경을 측정하였다. 병변의 성공적인 소멸은 추적 초음파에서 병변이 보이지 않거나 추적 초음파가 이루어지지 않은 경우에는 신체진찰에서 병변이 만져지지 않는 경우로 정의하였다. 통계학적으로 병변의 부피 또는 물리치료 여부와 성공적인 소멸 소요 시간과의 연관성을 분석하였다. 결과: 63명의 환자 중에서 48명이 선천성 사경으로 최종 진단되었으며, 분만형태는 39명이 자연분만, 9명이 제왕절개였다. 48명 중 24명에서 추적관찰이 이루어졌고, 22명에서 선천성 근성 사경 병변의 성공적인 소멸이 확인되었다. 24명 중에서 9명이 물리치료를 시행 받았으며 7명에서 성공적인 소멸을 보였다. 병변의 부피가 클수록 성공적인 소멸에 걸리는 시간이 짧았으며(p = 0.011), 물리치료와 소멸 소요 시간 간에는 통계학적으로 유의한 상관관계를 보이지 않았다(p = 0.15). 결론: 선천성 사경 환자에서 측정 가능한 목빗근 병변의 부피가 클수록 빠른 소멸을 보였으며, 물리치료는 병변의 소멸기간의 단축에 도움이 되지 않았다.

Predictable Marker for Regression of Barrett`s Esophagus by Proton Pump Inhibitor Treatment in Korea

( Hyun Jin Jo ),( Hye Seung Lee ),( Nayoung Kim ),( Ryoung Hee Nam ),( Hyun Chang ),( Min Soo Kim ),( Sung Eun Kim ),( Jane C Oh ),( Dong Ho Lee ),( Hyun Chae Jung ) 대한소화기기능성질환·운동학회 2013 Journal of Neurogastroenterology and Motility (JNM Vol.19 No.2

Background/Aims There has been no report regarding the regression of Barrett`s esophagus (BE) by continuous treatment of proton pump inhibitor (PPI). The aim of this study was to determine the regression rate of BE by PPI and predictable markers related to regression. Methods Thirty-five patients diagnosed as BE were consecutively enrolled and most of them took continuous PPI. The 25 patients underwent endoscopic surveillance and received biopsy. If the specialized intestinal metaplasia (SIM) was lost at any point of surveillance and did not recur, the case was regarded as the regression group. The proportion of SIM was graded and the mucin phenotype was decided using immunohistochemistry for MUC2, MUC5AC and MUC6. To assess the cell proliferation indexes and the degree of intestinal maturation, immunohistochemistry for Ki67 and CDX2 were performed. Results The regression of BE occurred in the 11 (44%) patients. The clinical and demographic factors showed no difference between the regression (n=11) and persistence group (n=14). The lower grade of SIM (P<0.001) and gastric predominant mucin phenotype (P=0.018) were more frequent, and the number of Ki67 positive cell per gland (P = 0.008) and the mean extent of CDX2 (P=0.022) were lower in the regression group than in the persistence group. Conclusions The regression of BE by PPI treatment was frequent in Korea. The immunohistochemical detection of mucin phenotype, grade of SIM, Ki67 and CDX2 expression in Barrett`s mucosa could be useful as a predictable marker for regression of SIM in BE.

의사결정나무와 로지스틱 회귀분석을 이용한 태권도 수련생 이탈 예측을 위한 비교 연구

권태원(Kwon Tae-Won),구유희(Koo Yu-Hoe) 한국체육과학회 2008 한국체육과학회지 Vol.17 No.2

The purpose of this study is to suggest the most appropriate prediction model for prediction of secession of trainees of Taekwondo gymnasium through decision-making tree technique among logistics regression analysis and data mining techniques. In order to accomplish the purpose of this study, I have distributed 1,500 questionnaires sheets to the trainees using by convenience sampling method among non probability sampling extraction methods by setting trainees of Taekwondo gymnasium located in Gyeonggi-Do and Incheon Results of this study derived from this procedure and method are as follows. 1. As the result of examining the difference of level of specialty between decision-making tree technique & logistics regression analysis, in the level of specialty predicting carry forward number of people with actual number of carry-over, logistics regression analysis was 92.9% and decision-making tree technique showed a little higher at 96.3%. 2. As the result of examining the difference of level of sensitivity between decision-making tree technique & logistics regression analysis, in the level of sensitivity predicting carry forward number of people with actual number of carry-over, logistics regression analysis was 64.0% and decision-making tree technique showed 44.5%. 3. As the result of examining the difference of level of accuracy between decision-making tree technique & logistics regression analysis, in the level of accuracy predicting carry forward number of people with actual number of carry-over, logistics regression analysis was 86.8% and logistics regression analysis showed 87.6%. 4. As the results of examining variables affecting secession of trainees of Taekwondo gymnasium between decision-making tree technique and logistics regression analysis, in case of decision-making tree technique, variables were training period, grade, satisfaction for instructor, recommendation will, and satisfaction of facility and in case of logistics regression analysis, variables were training period, grade, recommendation will, satisfaction for instructor, and per sex. Summarizing the above result, as the result of comparison analysis of secession prediction of trainees of Taekwondo gymnasium between decision-making tree technique and logistics regression analysis, the two models showed all high prediction rate in the level of accuracy and there was no difference of prediction rate between the two analysis models. This result is believed to be because the same variables affecting secession were used in the same.

제주특별자치도 토지 실거래가격 결정요인에 관한 연구

양영준(Yang Young Jun) 한국부동산학회 2015 不動産學報 Vol.61 No.-

본 연구에서는 최소자승법(OLS)에 의한 회귀분석과 분위회귀분석을 이용하여 제주특별자치도의 토지 실거래가격 결정요인을 분석하였다. OLS에 의한 회귀분석과 분위회귀분석에서 토지면적은 토지 실거래가격에 부정적인 영향을 미치는 것으로 분석되었고, 주거용도의 지목이 농업용도나 기타용도의 지목보다 가격이 높은 것으로 나타났고 상업지역의 가격이 다른 용도지역의 토지에 비해 가격이 높은 것으로 나타났다. 광대로가 접한 토지가 여타 도로에 접한 토지에 비해 토지가격이 높게 나타났고, 분위회귀분석에서는 고지가 분위로 갈수록 광대로가 접한 토지 대비 여타 도로가 접한 토지의 가격차이가 감소하는 것으로 나타났다. 본 연구의 자료에서 토지에 대한 지역 및 환경 요인에 대한 변수가 포함되지 않는 등 자료의 한계와 공간가기상관을 고려하지 않은 분석방법 등을 연구의 한계로 남겨놓는다. 1. CONTENTS (1) RESEARCH OBJECTIVES The purpose of this study is analyze Jeju Special Self-Governing land price determinants (2) RESEARCH METHOD In an empirical analysis, the officially assessed land price has been set as an dependent variable and the characteristic of lands have been set as an independent variable. I analyzed the land price determinants by using Ordinary Least Square(OLS) based regression analysis and quantile regression analysis. (3) RESEARCH FINDINGS According to OLS regression and quantile regression, land area negatively affected land prices. There was no significant difference between land category for residential purpose and the one for other purpose in quantile regression regarding the price while the prices of the ones for agricultural and other purpose was lower than the one for residential purpose in OLS regression. It turned out that the price of commercial area was higher than the residential area in land use zone regarding OLS regression and quantile regression, and the prices of green area and management area were lower. The area next to the roads influenced on the price positively and land shape showed the positive effects against the price as it got indeterminate. 2. RESULTS This study has a meaning of having analysing land price determinants through OLS regression and quantile regression, however leaves limitations on not considering omitted variables problem and spatial autocorrelation.