유전적 요인을 갖는 질병의 개인 및 가구요인에 대한 다수준분석의 의의 : 천식을 중심으로

이재인,이기재 한국자료분석학회 2019 Journal of the Korean Data Analysis Society Vol.21 No.4

Asthma is an allergic disease caused by a combination of genetic, environmental, and immunological factors. Previous studies did not reflect the correlation between the members of the family, although the patients were not at all independent in the family. A multilevel analysis model considering households was constructed and compared with the logistic regression analysis considering complex sample design and weights (complex sample-logistic regression) used in the previous studies. Data from the Korea national health and nutrition examination survey V and VI was analyzed. The models were compared with model 1 considering only the random effect of the household, model 2 considering the individual level variable additionally and model 3 considering the household level variable additionally. The estimates significant variables were compared between the optimal model of the multilevel analysis and the complex sample logistic regression. In multilevel analysis, age, sex, education level, anxiety or depression experience and family size were associated with asthma. In complex sample logistic regression analysis, age, anxiety or depression, and family size were associated with asthma with the same trends in multilevel analysis. Multilevel analysis in asthma, which has a genetic predisposition to correlate with family members, is considered to be a suitable model for finding more significant variables as much as complex sample - logistic regression. 천식은 알레르기 질환으로 유전적인 요인, 환경적인 요인, 면역학적 요인 등이 복합적으로 작용하여 발생하는 질환이다. 국민건강영양조사는 각 표본 조사구 안에 가구들이 포함되어 있고 가구 안에 개인들이 내재되어 있는 형태의 자료이다. 본 연구는 국민건강영양조사 제5기(2010-2012), 제6기(2013-2015) 자료를 대상으로 개인-가구 2수준을 고려한 다수준모형을 이용한 분석결과와 복합표본설계와 가중치를 반영한 로지스틱회귀분석(복합표본-로지스틱회귀분석, complex sample-logistic regression) 결과를 비교·분석하였다. 다수준모형은 가구 구분을 랜덤효과(random effect)로 하고, 개인-가구 수준의 설명변수를 고정효과(fixed effect)로 반영하였다. 다수준분석에서는 나이, 성별, 교육수준, 불안 또는 우울경험, 가구원수 등의 변수가 천식과 관련이 있었고, 복합표본-로지스틱회귀분석에서는 나이, 불안 또는 우울경험, 가구원수가 천식과 관련이 있었다. 유전적 경향으로 가구내 구성원의 상관성이 있는 천식에서 다수준분석은 복합표본-로지스틱회귀분석만큼 유의한 변수를 찾아내는데 적합한 분석 방법으로 볼 수 있으며, 추후 가중치가 반영된 다수준분석 방법을 적용할 수 있다면 좀 더 정확한 분석 결과를 도출해 낼 수 있을 것이다.

특수교육에서의 증거기반교수를 위한 다층모형 메타분석(Multilevel Meta-Analysis) 활용 방안 탐색

여승수,홍성두 한국특수교육학회 2011 특수교육학연구 Vol.46 No.1

It is necessary for special education teachers to use evidence based practices in order to improve academic achievement for students who need special education. In other words, special education teachers should take into account the use of evidence based practices in teaching students with disabilities. For that reason, evidence based practices have been considered as an important issue in the area of special education. With the importance of evidence based practices, many researchers and teachers are interested in a meta analysis that allows for evaluating the effectiveness of interventions used in special education. However, it is true that some meta analysis studies have incorrectly used a meta analysis. Therefore, the main purpose of this study was to provide a guideline that helps select an appropriate method that is suited for data structures used in a meta analysis. Especially, this paper provided an explanation on a multilevel meta analysis which has been received great attention. 특수교사는 특수교육을 필요로 하는 학생들의 교육적 성취를 위해 과학적으로 효과성이 증명된 증거기반교수를 사용하여 지도해야만 한다. 실제로 증거기반교수의 사용은 현재의 특수교육분야에서 중요한 주제로 부각되고 있다. 이런 증거기반교수의 중요성이 강조됨에 따라 증거기반교수를 평가하는데 있어서 사용되는 중요한 연구방법인 메타분석에 대한 관심 또한 증가하고 있다. 하지만 전통적인 메타분석연구를 보면 자료의 특성에 적합하지 않은 방법과 절차를 사용하여 연구를 진행함에 따라 심각한 문제점을 나타내고 있다. 이에 본 연구에서는 전통적인 메타분석방법의 문제점을 제시하고, 그 대안으로서 최근에 대두되고 있는 다층 메타분석(multi level meta-analysis)의 개념을 소개하고 관련된 구체적 사례를 제공함으로서 다층 메타분석에 대한 이해를 증진시키고자 하였다. 마지막으로 증거기반교수를 위한 다층메타분석의 활용과 관련된 시사점을 서술하였다.

Validity analysis on merged and averaged data using within and between analysis: focus on effect of qualitative social capital on self-rated health

Sang soo Shin,Young-jeon Shin 한국역학회 2016 Epidemiology and Health Vol.38 No.-

OBJECTIVES: With an increasing number of studies highlighting regional social capital (SC) as a determinant of health, many studies are using multi-level analysis with merged and averaged scores of community residents’ survey responses calculated from community SC data. Sufficient examination is required to validate if the merged and averaged data can represent the community. Therefore, this study analyzes the validity of the selected indicators and their applicability in multi-level analysis. METHODS: Within and between analysis (WABA) was performed after creating community variables using merged and averaged data of community residents’ responses from the 2013 Community Health Survey in Korea, using subjective self-rated health assessment as a dependent variable. Further analysis was performed following the model suggested by WABA result. RESULTS: Both E-test results (1) and WABA results (2) revealed that single-level analysis needs to be performed using qualitative SC variable with cluster mean centering. Through single-level multivariate regression analysis, qualitative SC with cluster mean centering showed positive effect on self-rated health (0.054, p<0.001), although there was no substantial difference in comparison to analysis using SC variables without cluster mean centering or multi-level analysis. CONCLUSIONS: As modification in qualitative SC was larger within the community than between communities, we validate that relational analysis of individual self-rated health can be performed within the group, using cluster mean centering. Other tests besides the WABA can be performed in the future to confirm the validity of using community variables and their applicability in multi-level analysis.

Validity analysis on merged and averaged data using within and between analysis: focus on effect of qualitative social capital on self-rated health

신상수,신영전 한국역학회 2016 Epidemiology and Health Vol.38 No.-

OBJECTIVES: With an increasing number of studies highlighting regional social capital (SC) as a determinant of health, many studies are using multi-level analysis with merged and averaged scores of community residents’ survey responses calculated from community SC data. Sufficient examination is required to validate if the merged and averaged data can represent the community. Therefore, this study analyzes the validity of the selected indicators and their applicability in multi-level analysis. METHODS: Within and between analysis (WABA) was performed after creating community variables using merged and averaged data of community residents’ responses from the 2013 Community Health Survey in Korea, using subjective self-rated health assessment as a dependent variable. Further analysis was performed following the model suggested by WABA result. RESULTS: Both E-test results (1) and WABA results (2) revealed that single-level analysis needs to be performed using qualitative SC variable with cluster mean centering. Through single-level multivariate regression analysis, qualitative SC with cluster mean centering showed positive effect on self-rated health (0.054, p<0.001), although there was no substantial difference in comparison to analysis using SC variables without cluster mean centering or multi-level analysis. CONCLUSIONS: As modification in qualitative SC was larger within the community than between communities, we validate that relational analysis of individual self-rated health can be performed within the group, using cluster mean centering. Other tests besides the WABA can be performed in the future to confirm the validity of using community variables and their applicability in multi-level analysis.

다집단 다층분석에 의한 학교 효과 분석

이명애,박도영,박찬호,양승호,김동영,장의선,김경주 한국교육과정평가원 2012 교육과정평가연구 Vol.15 No.3

The purpose of this study was to investigate differential effects of student- and school-level predictors on student achievement by grouping schools according to their achievement levels. For this purpose, the schools were first divided into high, middle, and low groups by the means of their students' standard scores, and then a multiple-group multilevel (MGML) analysis model was fitted to the Korean language, mathematics B, and foreign language (English) section data obtained from a preliminary test of the 2012 college scholastic ability test. The results showed that the ratios of student-level and school-level variances explained by the predictors were evidently different among the high, middle, and low schools. Also, the MGML analysis in comparison to the single-group multilevel (SGML) analysis revealed differential effects of predictors among the high, middle, and low schools. The SGML analysis showed that schools in urban areas had higher scores on the Korean language, mathematics B, and foreign language (English) than those in rural areas. From the MGML analysis results, however, only high-level schools in urban areas had higher scores on the foreign language (English) section than those in rural areas. Also, the effects of education policies for rural schools and no private education schools were not statistically significant in the SGML analysis, but were significant for low-level schools in the MGML analysis. 본 연구에서는 다집단 다층모형을 적용하여 학교의 성취수준에 따른 학생 및 학교수준 독립변인들의 차별적인 효과를 분석하였다. 이를 위해 2012학년도 대학수학능력시험 모의평가의 언어, 수리나, 외국어 영역 표준점수 학교별 평균을 기준으로 학교들을 상ㆍ중ㆍ하위로 구분하여 다집단 다층분석을 실시하였다. 분석 결과, 학생수준과 학교수준에서 독립변인들에 의해 설명되는 분산 비율이 상ㆍ중ㆍ하위 학교별로 상이한 사실이 분명하게 드러났으며, 전체 학교를 대상으로 한 다층분석 결과와는 달리 다집단 다층분석은 상ㆍ중ㆍ하위 학교별로 독립변인들이 가지는 차별적인 효과를 보여주었다. 예를 들어 다층분석 결과에 따르면 대도시와 중소도시의 학교들은 읍면지역에 소재한 학교보다 언어, 수리나, 외국어 영역의 표준점수 평균이 더 높지만, 다집단 다층분석 결과를 보면 외국어 영역의 상위권 학교에서만 대도시와 중소도시 학교들의 평균이 읍면지역 학교들의 평균보다 높은 것으로 확인되었다. 또한, 농산어촌우수고, 사교육없는학교와 같은 교육 정책의 효과가 다층분석에서는 유의미하지 않았지만 다집단 다층분석에서는 하위권 학교에서 유의미한 것으로 밝혀졌다.

고정 효과 회귀분석 vs. 무선 효과 위계적 선형모델(HLM): 가족 구성 및 정보 프라이버시 염려가 비판적 정보 추구 행위에 미치는 영향

류성진 한국언론학회 2024 커뮤니케이션 이론 Vol.20 No.1

In the case of regression analysis, it is necessary to diagnose whether the assumptions such as linearity, normality, homoscedasticity, and independence of residuals, multicollinearity are satisfied before full-scale analysis. If some assumptions are not met, appropriate measures, such as data transformation, must be applied to the data to achieve valid and reliable analysis results only after full-scale analysis is performed after the assumptions are met. Textbooks or papers dealing with the diagnosis of residual linearity, normality, homoscedasticity, and multicollinearity among the various assumptions introduced earlier are not difficult to find in communication academia. However, papers and discussions on the assumption of independence are hard to find in those area. In this repect, the study aims to introduce the assumption of residual independence, diagnosis, and statistical analysis technique that can overcome violation of residual independence, that is, hierarchical linear model(hereinafter, HLM). If clustering occurs due to interactions among members within the group, residual independence may be violated, which leads to an increase in standard error, resulting in a problem of testing inappropriate statistical significance. HLM has the advantage of being able to test appropriate statistical significance in consideration of such residual non-independence. In this study, in order to clearly explain the characteristics of HLM, we tried to conduct a comparative analysis with the results of fixed effect regression analysis. Using 2023 media panel survey data, the two statistical analysis techniques presented above were applied to test the differences in the impact of information privacy concerns on media literacy according to family composition. As a result of the analysis, the effect on media literacy according to family composition was statistically significant in the fixed effect regression analysis, and the interaction between family composition and information privacy concerns on media literacy was significant. However, in the case of this regression analysis, Durbin Watson d statistics confirmed that the assumption of independence was violated, confirming that there is a possibility of committing a Type I error. To the contrary, HLM revealed that the average difference in random effects of influence on media literacy according to family composition and the difference in the slope of random effects of information privacy concerns on media literacy according to family composition were not statistically significant, either. This study confirmed that in the case of data that can violate the assumption of residual independence, that is, if group clustering is likely to occur, using HLM is an effect way to reduce Type I errors. 회귀분석의 경우 본격적인 분석에 앞서 잔차의 선형성, 정규성, 등분산성, 독립성, 다중공선성 등의 가정 충족 여부를 진단해야 한다. 만일 일부 가정이 충족되지 않을 경우, 데이터에 적절한 조치, 예를 들어, 데이터 변환 등을 통해 가정을 충족한이후분석을 실행할 때 타당하고 신뢰할 수 있는 분석 결과를 얻을 수 있다. 커뮤니케이션 학계에서는 앞서 소개한 잔차의 선형성, 정규성, 등분산성, 다중공선성에 대한 가정 진단을 다룬 교재나 논문을 어렵지 않게 접할 수 있다. 그러나 독립성 가정을 다룬 논문이나 논의는 의외로 찾기가 쉽지 않다. 이 점에 착안하여, 이 연구는 잔차의 독립성 가정 및 진단 방법을 소개하고자 한다. 더불어, 잔차의 독립성 위배를 극복할 수 있는 통계분석기법, 즉 위계적 선형모델(hierarchical linear model: HLM)도 함께 소개하고자 한다. 집단 내 구성원 간 상호작용으로 인한 군집화가 발생할 때 잔차 독립성은 위배될 가능성이 있으며, 이는 곧 표준오차의 증가로 이어져부적절한통계적유의성을검증하는문제가 발생한다. HLM은 이 같은 잔차의 비독립성을 고려하여 타당한 방식으로 통계적 유의성을 검증할 수 있다는 장점이 있다. 이번 연구에서는 HLM의 특징을 명확하게 설명할 목적으로 기존 연구에서 일반적으로실행하는고정 효과 OLS 회귀분석의 결과와 비교분석을 진행했다. 2022년 미디어패널조사데이터를활용하여 가족 구성에따라정보 프라이버시 염려가 비판적 정보 추구 행위에 미치는 영향의 차이를 검정하기 위해 앞서 제시한 두 가지 통계분석기법을 적용하여 비교분석했다. 분석 결과, 고정 효과 OLS 회귀분석에서는 가족 구성에 따라 비판적 정보 추구 행위에 미치는 영향이 통계적으로 유의미했으며, 비판적 정보 추구 행위에 미치는 가족 구성과 정보 프라이버시 염려 간 상호작용은 유의미한 것으로 드러났다. 그러나 잔차 간 독립성 가정을 진단한 결과, 이 가정은 위배되었음이 밝혀졌고, 따라서 유의성 검증에 대한 제1종 오류를 범할 위험이 있음을 확인할 수 있었다. 이를 극복하기 위해 HLM을 실행한 결과, 가족 구성에 따라 비판적 정보 추구 행위에 미치는 영향의 무선 효과 평균 차이 그리고 가족 구성에 따라 정보 프라이버시 염려가 비판적 정보 추구 행위에 미치는 영향, 즉 무선 효과 기울기 차이는 통계적으로 유의미하지 않은 것으로 드러났다. 이번 연구 결과를 통해 잔차의 독립성 가정을 위배할 수 있는 데이터의 경우, 즉 집단의 군집화가 발생할 가능성이 있는 경우, 향후 연구자들은 HLM 사용을 신중하게 고려해 보길 이번 연구는 제안하고자 한다.

노인의 건강 관련 삶의 질에 영향을 미치는 요인에 관한 다수준 분석

문현정,차선경 한국정신간호학회 2022 정신간호학회지 Vol.31 No.3

Purpose: This study identifies the individual and community level factors affecting the health-related quality of life (QOL) of the elderly based on an ecological model using multilevel analysis. Methods: As a secondary research, this study used the raw data, community health survey results, and database 1.0 data on community health outcomes and determinants from the same year. Survey participants were 62,710 respondents age ≥65 of 228,558, and database 1.0 data on community health outcomes and determinants were established targeting 254 regions nationally. Results: The health-related QOL showed differences according to the individual and community level factors. Additionally, estimating the effects of the community level factors through multilevel analysis was needed. In a model in which the individual and community level factors were simultaneously inserted, the significant com- munity characteristic factor controlling the individual level factors was the number of hospital beds per thousand population (p=.028). Conclusion: In the multilevel analysis considering the individual and community level factors, the number of hospital beds per thousand population, which is a factor concerning medical access, was identified as a factor significantly affecting health-related QOL. The results should be considered in allocating and developing community- based healthcare resources and establishing strategies to improve the health-related QOL.

Civic Participation and Self-rated Health: A Cross-national Multi-level Analysis Using the World Value Survey

유명순,김새롬,김창엽 대한예방의학회 2015 Journal of Preventive Medicine and Public Health Vol.48 No.1

Objectives: Civic participation, that which directly influences important decisions in our personal lives, is considered necessary for developinga society. We hypothesized that civic participation might be related to self-rated health status. Methods: We constructed a multi-level analysis using data from the World Value Survey (44 countries, n=50 859). Results: People who participated in voting and voluntary social activities tended to report better subjective health than those who did not vote or participate in social activities, after controlling for socio-demographic factors at the individual level. A negative associationwith unconventional political activity and subjective health was found, but this effect disappeared in a subset analysis of only the 18 Organization for Economic Cooperation and Development (OECD) countries. Moreover, social participation and unconventional political participation had a statistically significant contextual association with subjective health status, but this relationship was not consistent throughout the analysis. In the analysis of the 44 countries, social participation was of borderline significance, while in the subset analysis of the OECD countries unconventional political participation was a stronger contextual determinant of subjective health. The democratic index was a significant factor in determining self-rated health in both analyses, while public health expenditurewas a significant factor in analysis of 18 countries. Conclusions: Our investigation suggests that civic participation, including unconventional political activity at the contextual level, might be a significant determinant of health status of a country.

Civic Participation and Self-rated Health: A Cross-national Multi-level Analysis Using the World Value Survey

Kim, Saerom,Kim, Chang-Yup,You, Myung Soon The Korean Society for Preventive Medicine 2015 Journal of Preventive Medicine and Public Health Vol.48 No.1

Objectives: Civic participation, that which directly influences important decisions in our personal lives, is considered necessary for developing a society. We hypothesized that civic participation might be related to self-rated health status. Methods: We constructed a multi-level analysis using data from the World Value Survey (44 countries, n=50 859). Results: People who participated in voting and voluntary social activities tended to report better subjective health than those who did not vote or participate in social activities, after controlling for socio-demographic factors at the individual level. A negative association with unconventional political activity and subjective health was found, but this effect disappeared in a subset analysis of only the 18 Organization for Economic Cooperation and Development (OECD) countries. Moreover, social participation and unconventional political participation had a statistically significant contextual association with subjective health status, but this relationship was not consistent throughout the analysis. In the analysis of the 44 countries, social participation was of borderline significance, while in the subset analysis of the OECD countries unconventional political participation was a stronger contextual determinant of subjective health. The democratic index was a significant factor in determining self-rated health in both analyses, while public health expenditure was a significant factor in analysis of 18 countries. Conclusions: Our investigation suggests that civic participation, including unconventional political activity at the contextual level, might be a significant determinant of health status of a country.

An Analysis Method of Harmonic Transfer Characteristics From AC-side to DC -side of Modular Multilevel Converter

Pengfan Xu,Yue Wang,Taiyuan Yin,Shiyuan Yin,Zhang Wang,Kai Li 전력전자학회 2019 ICPE(ISPE)논문집 Vol.2019 No.5

With the development of new energy sources, more and more isolated islands are interconnected to grid through MMC. However, isolated islands systems are easily influenced by harmonic disturbances. The analytical analysis of the harmonic transfer characteristics between AC/DC side of the Modular Multilevel Converter (MMC) is of great significance for understanding the harmonic issues and stability analysis. In this paper, the quantitative analysis method of the harmonics on the DC-side excited by the AC-side harmonics of MMC is presented in this paper. First, the influence of harmonics on MMC electric system and control system is analyzed. Second, the equivalent MMC harmonic transfer model is established based on the average switching function. Finally, the analytical expression of DC-side harmonic transferred from AC-side is derived. A 232-level simulation model was built in MATLAB/Simulink. The accuracy and validity of the proposed analytical expression were verified by comparing the simulation results with the analytical results.