Symptomatology throughout pandemic : network analysis and latent transition analysis of prolonged post-traumatic stress symptoms, depressive symptoms, and sleep disturbance

이유진 Graduate School, Yonsei University 2023 국내박사

연구배경: COVID-19이 팬데믹(Pandemic)으로 공식 선언됨으로 인해, COVID19은 집단 트라우마로서 일반인구의 정신건강에 영향을 끼치고 있다. 팬데믹으로 인한 정신적 영향이 신체적 감염 정도보다 빠르다는 사실과 충격적인 사건의 경험이 다양한 정신적 고통을 유발할 수 있다는 사실을 고려할 때, COVID-19 팬데믹의 장기화는 일반인구에게 외상 후 스트레스증상(Posttraumatic stress symptom, PTSS) 및 우울증상(Depressive symptom)과 같이 빈번하게 함께 발생하는 것으로 알려진 정신증상들의 동반이환(Comorbidity) 유병률을 가속화 시킬 수 있다. 이처럼, 지속되는 팬데믹은 외상 후 스트레스증상과 우울증상 사이의 장기적인 상호 작용을 야기할 수 있으며, 이는 이후의 동반이환 구조 및 패턴을 변화시킬 수 있다. 이러한 관점에서, 팬데믹 기간의 시간 흐름에 따른 외상 후 스트레스증상 및 우울증상의 동반이환 메커니즘을 이해하는 것은 일반인구의 정신 건강을 위한 개입점이 될 수 있다. 연구방법: 연구 대상자는 30세에서 64세로 구성되었으며, COVID-19 기간 동안 진행한 3차례에 걸친 심뇌혈관 및 대사질환 원인연구센터 온라인 정신건강 설문조사의 참여자가 연구에 포함되었다(1차: 1,925명, 2차: 1,754명, 3차: 1,595명). 외상 후 스트레스증상 및 우울증상은 각각 PCL-5 (Post-traumatic stress disorder Checklist for the DSM-5) 및 PHQ-9 (Patient Health Questionnaires-9)으로 측정되었다. 모든 분석은 Goldbricker 테스트를 통해 모든 증상 사이의 연관성 패턴 중복 정도를 테스트 후 네트워크분석(Network analysis)을 수행하였다. 네트워크분석을 통해 외상 후 스트레스증상 및 우울증상 동반이환의 증상 수준 메커니즘(Symptom-level mechanism)을 밝히고자 했으며, 분석은 1) 네트워크추정(Network estimation), 2) 네트 워크추론(Network inference), 3) 네트워크 견고성평가(Network Robustness), 그리고 4) 시점 간 네트워크비교(Network comparison) 네 단계로 진행되었다. 네트워크분석을 통해 발견된 외상 후 스트레스증상 및 우울증상을 동시에 유지시키는 동반이환증상(Bridge symptom)은 잠재전이분석(Latent transition analysis, LTA)를 통해 시간에 따른 변화 패턴 및 영향을 고려했다. 잠재전이분석을 통해 팬데믹 기간에 따른 전이 정도와 이후 이의 외상 후 스트레스증상 및 우울증상에의 영향을 연령 및 성별로 계층화 하여 측정하였다. 연구결과: 외상후 스트레스증상 및 우울증상의 동반이환 네트워크(Comorbidity network)는 각 정신증상의 정의에 부합하는 두개의 분리된 증상 커뮤니티를 구성하였다. 팬데믹 전 시점에서 수면문제(Sleep problems and sleep disturbance)는 두개의 증상커뮤니티를 가장 강하게 연결하는 가교증상(Bridge symptom)으로 확인되었지만, 중심증상(Central symptom)은 시점에 따라 변화하는 패턴을 보였다. 동반이환 네트워크는 모든 시점에서 안정적이었지만, 판데믹의 아급성기(Subacute phase) 및 만성기(Chronic phase)의 네트워크 구조 자체는 서로 유의하게 상이하였다. 동반이환 네트워크 모델에 대한 확정 후, 가교증상인 수면문제에 대하여 잠재전이분석을 진행하였다. 수면문제는 PSQI (Pittsburgh Sleep Quality Index) 문항 중 수면장애(Sleep disturbance) 섹션 9개 문항으로 정의하였으며, 잠재계층분석(Latent Class Analysis, LCA)을 통해 기준상태(Reference), 수면 연속성문제가 있는 상태(Sleep continuity problems), 그리고 전반적인 수면문제가 있는 상태(Overall sleep problems)의 세가지 잠재상태로 분류되었다. 잠재전이분석을 적용하여 분석 한 결과, COVID-19이 아급성기에서 만성기로 변환할 때 기준상태 및 전반적인 수면문제가 있는 상태는 수면연속성문제가 있는 상태로 전이되는 패턴을 보였다. 이러한 수면 연속성문제로의 전이는 외상 후 스트레스증상 및 우울증상에 유의한 부정적인 연관성을 도출했으며, 특히 이러한 경향은 여성 및 50- 60대에서 두드러졌다. 아울러, 수면 연속성 문제 상태는 팬데믹 이후의 외상 후 스트레스증상 및 우울증상 모두에 부정적인 영향을 끼칠 가능성을 보였다. 결론: 본 연구는 팬데믹이 장기화됨에 따라 변화하는 외상 후 스트레스증상 및 우울증상의 동반이환 메커니즘의 변화하는 패턴을 발견하였으며, 두 정신증상의 동반이환 가교증상은 수면문제였다. 또한, 팬데믹 상황이 아급성기에서 만성기로 전환되는 동안 수면문제 상태들의 수면 연속성문제로의 전이가 두드러졌으며, 이러한 전이는 이후 외상 후 스트레스증상 및 우울증상에 부정적인 연관성을 보였다. 이러한 결과를 공중보건(Public health)의 관점에서 적용한다면, 향후 COVID-19과 유사한 감염성 전염병(Next pandemic)이 발생하였을 경우 일반인구를 대상으로 수면장애가 있는 인구를 신속하게 선별하고 그 잠재적인 수면상태까지 고려하는 것이 일반인구의 정신건강에 도움을 줄 수 있는 개입점이 될 수 있으며, 이는 특히 팬데믹의 아급성기에서 만성기로의 변화 시점에 초점을 맞추어 진행되어야 한다. BACKGROUND After the declaration of Coronavirus disease 2019 (COVID-19) as a pandemic, it has been considered a mass trauma or collective trauma for the general population. Given that those who are psychologically affected by the pandemic have outnumbered those who are physically infected with COVID-19 and given that experiencing traumatic events could trigger a wide variety of psychological distresses, the prolonged COVID-19 pandemic could lead to a significant increase in the prevalence of psychiatric comorbidity, especially between post-traumatic stress symptoms (PTSS) and depressive symptoms among the public. In addition, the symptom-level interplay between PTSS and depressive symptoms during the pandemic could cause a tilting pattern for deterioration and improvement of comorbidity. Accordingly, capturing the comorbid structure between PTSS and depressive symptoms and understanding the mechanisms of comorbid symptoms over time might be an intervention point for the public. METHODS Participants aged 30–64 years completed an online mental health survey from the Cardiovascular and Metabolic Etiology Research Center across three phases of the COVID-19 pandemic (Ns= 1925, 1754, 1595). The PTSS and depressive symptoms were estimated by online survey versions of the post-traumatic stress disorder Checklist for the DSM-5 (PCL-5) and Patient Health Questionnaires-9 (PHQ-9). After the goldbricker test for redundancy, we conducted network analyses to find the symptom-level mechanism between PCL-5 and PHQ-9 through four steps: 1) network estimation, 2) network inference, 3) network robustness, and 4) network comparison. To capture the changing patterns of comorbid symptoms, we applied latent transition analysis and inferred a significant status transition throughout COVID-19 and a possible impact on PTSS/depressive symptoms over time, stratified by age and sex. RESULTS From the comorbidity network, the PCL-5 and PHQ-9 formed two separate communities. Moreover, sleep problems were consistently identified as the most influential bridge symptom throughout COVID-19, whereas the central symptoms varied. Each structure of the comorbidity network was stable but suggested a significant difference between the subacute and chronic phases of COVID-19. After building on the network model, we conducted latent transition analysis on the bridge symptom, which was sleep problems in our study. The sleep problems were defined by scores of nine items of sleep disturbance components among the Pittsburgh Sleep Quality Index. As a result of latent transition analysis, the sleep disturbance was categorized as three latent statuses: reference (status 1), sleep continuity problems (status 2), and overall sleep problems (status 3). When transitioning from the subacute to chronic phase, changing patterns from reference or overall sleep problems into sleep continuity problems were primarily reported. This pattern showed a significant negative association with PCL-5 and PHQ-9 during the pandemic, especially for women and participants aged 50-60 years. Further, the sleep continuity problems status had a possible negative impact on PCL-5 and PHQ-9 over time. CONCLUSION Our results suggest that there might exist a changing pattern in network structure throughout the COVID-19 phases, while the bridge symptom, and sleep problems in our study, remain constant. In addition, the transition to sleep continuity problems, primarily from other sleep disturbances occurred during the time flow from the subacute to chronic phases, and this transition has a negative association with PTSS and depressive symptoms. Hence, from a public health perspective, if an infectious epidemic situation (i.e., next pandemic) arises in the future, prompt screening of associated sleep problems and considering its status would be effective as an effort to alleviate both the comorbidity of PTSS and depressive symptoms and adverse psychological effects, especially by focusing on the tilting point from the subacute to the chronic phase of the pandemic.

Latent Profile and Transition Analysis of Job Crafting among Korean Employees: Focusing on the Effects of Regulatory Focus, Temporal and Job Characteristics

Kim, Sun Young 연세대학교 일반대학원 2025 국내박사

Experiences of Posttraumatic Stress in a Sample of Sexual Assault Survivors: A Longitudinal Latent Class Analysis Incorporating Structural and Functional Social Support

Goodman-Williams, Rachael Michigan State University ProQuest Dissertations & 2020 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

Sexual assault is a major public health crisis, with national epidemiological studies reporting that approximately one-in-five women will be sexually assaulted in their lifetime. Experiential and empirical data indicate that sexual assault causes significant distress for those who experience it, with approximately one-third of sexual assault survivors evidencing lifetime posttraumatic stress disorder (PTSD). In recent years, there have been increasing calls to expand recognition of PTSD beyond its binary framing and to incorporate contextual variables into the study of PTSD so that understandings of posttraumatic stress are more fully situated within the context of trauma survivors’ lives. The current study responds to these stated needs. This study utilized two waves of a longitudinal data set in which sexual assault survivors were surveyed annually about their posttraumatic stress symptoms and social support. Using this data, I conducted a latent transition analysis (LTA) to identify latent classes of posttraumatic stress experiences and model the probability of transitioning between latent classes over time. With the goal of highlighting existing strengths in survivors’ communities and better understanding optimal utilization of those resources, I also incorporated structural and functional social support as predictors of class membership and class transition. Four latent classes emerged at each time point and were named the High Severity, Depressed & Anxious, Avoidant & Reactive, and Low Severity classes. These classes demonstrated both quantitative (i.e., overall severity) and qualitative (i.e., symptom cluster severity) differences. Transition probabilities illustrated a general pattern of de-escalation. As would be expected, de-escalation was more likely when both types of social support were high compared to when both types of social support were low. Notable differences between the latent classes emerged, however, when one type of social support was high and the other was low. Specifically, while participants assigned to the High Severity class at Time 1 were more likely to de-escalate when structural support was high and functional support was low than the reverse, the opposite pattern was true for participants assigned to the Depressed & Anxious class. The intermediate classes identified in this latent class analysis highlight the limitations of categorical understandings of posttraumatic stress in which one either “has” or “does not have” PTSD. These findings support the calls that have been made by researchers and clinicians to introduce a “subclinical” level of PTSD to the Diagnostic and Statistical Manual, while also encouraging nuance beyond what even a subclinical designation would provide. Specifically, while a subclinical designation would reflect some amount of quantitative variation between the latent classes, it would be unlikely to reflect the qualitative differences found in the current study. These qualitative differences were especially relevant to understanding the longitudinal relationship between posttraumatic stress and social support, with functional support appearing to be especially relevant to those in the High Severity class and structural support especially relevant to those in the Depressed & Anxious class. The findings indicate that sexual assault survivors’ social support needs may vary based on the specific symptoms that define their experience of posttraumatic stress at a given point in time. These findings also provide guidance for clinical and non-clinical practitioners regarding how to help survivors engage their support networks most effectively in the aftermath of trauma.

A Comparison of Type of Random Intercept in Random Intercept Latent Transition Analysis with Small Sample Size

Joonwoo Lee 고려대학교 대학원 2023 국내석사

Latent transition analysis is one of the longitudinally expanded models of the mixed model and is used to estimate the change in type over time. An extended model that adds random intercepts to existing latent transition analysis has recently been introduced. If the perspective of the multilevel model is applied to the latent transition analysis, the time can be viewed at the level-1 and the individual at the level-2. Random intercept latent transition analysis (RI-LTA) aims to estimate pure transition probabilities over time by separating variance at the individual level, which is 2-level. To add a random intercept to the model, it is necessary to determine the type of random intercept and restrict the factor loading of the indicator variable. The types of random intercepts include continuous random intercepts and binary random intercepts, and random intercepts should be added in a way that constrains the factor loading of each indicator variable from the random intercept to be the same. This study aims to determine which random intercept type estimates the transition probability more stably and how much sample size and number of time points must be secured to estimate a stable transition probability when the sample collection environment is limited. In addition, this study aims to check whether the required sample size and number of time points vary depending on the conditions of the model. Therefore, it was confirmed through Monte Carlo simulation studies. The following simulation model was defined to conduct the study. Considering previous studies, the number of latent classes at all points in time was set to 2. The ratio at the first point in time was set at the same (50:50). The average of random intercepts was set to 0, and the variance was set to 1. The transition probability from class 1 to class 1 was set to 0.622, and the transition probability from class 2 to class 1 was set to 0.500. Simulation conditions are sample size (100, 300, 500), the number of indicator variables (5, 8), the number of time points (2, 3, 4, 5), and the type of random intercept (continuous, binary). Twenty-four thousand datasets were created to analyze this model and generated by 500 replications. As the evaluation criteria, bias, the standard deviation and average standard error across replications and the 95% coverage probability were used. As a result of the study, the bias was problematic only when the sample size was 100 at the second point, and five indicator variables with continuous random intercepts were used. The standard deviation and average standard error across replication were effectively reduced when continuous random intercepts with eight indicators were used. Standard deviation and average standard error increased as the number of time points increased with binary random intercept. The 95% coverage probability was found to have no problem with convergence when using continuous random intercepts, except that the sample size was 100 at two points with five indicators were used. However, when the binary random intercept was used, it was found that there was no problem with convergence only when the sample size was 500. Therefore, continuous latent intercepts are recommended to utilize random intercept latent transition analysis in a limited sample collection environment, and at least 300 sample sizes and at least 3 points should be collected. This study is meaningful in providing guidelines for collecting adequate sample sizes in a limited sample collection environment by exploring the sample size and the number of time points required for random intercept latent transition analysis. In addition, it is expected to contribute to the use of various mixed models by providing guidelines for more stable transition probability estimation by adjusting the types of random intercepts and the number of indicator variables in the area of modeling. 잠재전이분석은 혼합모형을 종단적으로 확장한 모형 중 하나로 시간에 따른 유형의 변화를 확인하고자 할 때 활용한다. 최근 기존 잠재전이분석에 랜덤 절편을 추가한 확장된 모형이 소개되었다. 잠재전이분석에 다층모형의 관점을 적용하면, 시간을 1-수준으로 개인을 2-수준으로 볼 수 있다. 랜덤 절편 잠재전이분석(RI-LTA)는 2-수준인 개인 수준의 분산을 분리하여 시간에 따른 순수한 전이확률을 추정하는 것에 목적이 있다. 랜덤 절편을 모형에 추가하기 위해서는 랜덤 절편의 종류를 결정하고 지표변수의 요인부하량을 제약해주어야 한다. 랜덤 절편의 종류로는 연속형 랜덤 절편과 이분형 랜덤 절편이 있으며, 랜덤 절편으로부터의 각 지표변수의 요인부하량은 동일한 문항끼리 같도록 제약하는 방식으로 랜덤 절편을 추가할 수 있다. 본 연구의 목적은 랜덤 절편 잠재전이분석에서 어떤 랜덤 절편 유형이 전이확률을 더 안정적으로 추정하는지 알아보고, 또한 제한된 표본수집 환경에서 최소한 어느 정도의 표본크기와 시점 수가 확보되어야 안정적인 전이확률을 추정하는 지 알아보는 것이다. 따라서, 몬테카를로 시뮬레이션 연구를 통해 확인하였다. 연구 수행을 위해 다음의 시뮬레이션 모형을 정의하였다. 선행연구를 고려하여 모든 시점의 잠재집단의 수는 2개로 하였으며, 첫 시점의 비율은 동일한 비율(50:50)로 설정하였다. 랜덤 절편의 평균은 0, 분산은 1로 설정하였으며, 전이확률은 집단 1에서 집단 1을 유지할 확률은 0.622, 집단2에서 집단1로 전이할 확률은 0.500으로 설정하였다. 시뮬레이션 조건으로는 표본크기 (100, 300, 500), 지표변수의 수 (5, 8), 시점 수 (2, 3, 4, 5), 랜덤 절편의 종류 (연속형, 이분형)으로 총 48개의 조건을 500번 반복하여 24,000개의 데이터 셋을 생성하였다. 준거 기준으로는 편향, 반복을 통해 나타난 표준편차와 표준오차의 평균, 95% 수렴 확률을 사용하였다. 연구결과, 편향은 2시점에 표본크기가 100개이면서 5개의 지표변수와 연속형 랜덤 절편을 사용한 경우에만 문제가 발생하였다. 반복을 통해 나타난 표준편차와 표준오차의 평균 연속형 랜덤 절편을 사용하였 때는 표본 크기의 증가가 표준편차 및 표준오차를 줄이는데 효과가 있었으며, 이분형 랜덤 절편을 사용하였을 때는 8개의 지표변수를 함께 사용한다면 시점 수가 증가함에 따라 표준편차 및 표준오차가 증가하였다. 95% 수렴확률은 연속형 랜덤 절편을 사용하였을 때는 2개 시점에 표본 크기가 100이면서 5개의 지표변수를 활용한 경우를 제외하면 수렴에 문제가 없었으나, 이분형 랜덤 절편을 활용하였을 때는 표본 크기가 500은 되어야 수렴에 문제가 없는 것으로 나타났다. 따라서 분석결과를 종합하면, 제한된 표본 수집 환경에서 랜덤 절편 잠재전이분석을 활용하기 위해서는 연속형 잠재 절편을 활용할 것을 권장하며 최소 300이상의 표본크기와 3 시점 이상의 반복 측정된 자료를 활용할 것을 권장한다. 본 연구는 랜덤 절편 잠재전이 분석에 필요한 표본 크기와 시점 수를 탐색하여, 제한된 표본 수집 환경에서 적절한 표본 크기를 수집하는 데 가이드라인은 제시하였다는 의의가 있다. 또한, 모델링의 영역에서 랜덤 절편의 종류와 지표변수의 수를 조정하여 보다 안정적인 전이확률 추정을 위한 가이드라인을 제공하여 다양한 혼합모형의 활용에 기여할 수 있을 것으로 기대한다.

A Comparison of Mixture Modeling Approaches in Latent Transition Analysis Model with Predictor

Seunghyeok Ha 고려대학교 대학원 2023 국내석사

Mixture modeling approaches of latent transition analysis with predictor have existed in 1-step, and 3-step ML approaches, with the recent addition of the BCH-LTA approach extending the scope of the latent transition analysis methodology. Therefore, available LTA (Latent Transition Analysis) methodologies are three approaches: 1. The 1-step approach is a method of simultaneously analyzing a measurement model that classifies each latent class and a structural model that examines the relationship between each allocated latent class and auxiliary variable. 2. The 3-step ML approach is a method of classifying latent classes as indicator variables, assigning individuals to divided latent classes, and then checking the relationship between assigned latent classes and the predictor. 3. The BCH-LTA approach is a methodology that applies the 3-step BCH approach to latent transition analysis. After saving the BCH weights of all latent class variables, the BCH weights are stored according to the measurement model, and each BCH weight is multiplied to obtain the combined BCH weights. This study aims to compare how the new approach, BCH-LTA, differs from the existing approach, 1-step, and 3-step ML approaches, and to find the minimum conditions for using the BCH-LTA approach. Therefore, it was examined through Monte Carlo simulation studies. The simulation model was designed to conduct the study. When considering previous studies, there are two latent classes at each of the two points, the same proportion (50%: 50%) at time 1 and a different proportion (70%: 30%) at time 2. At this time, measurement invariance was not considered because it was not an essential condition. Six dichotomous variables were set as indicators, and one continuous variable was set as the predictor. Two thousand seven hundred datasets were created to analyze this model and generated by 100 replications considering the mixture modeling approach (1-step, 3-step ML, BCH-LTA), sample size (200, 500, 1000), and entropy (0.5, 0.7, 0.9). As evaluation criteria, bias (parameter estimation, standard error), mean square error, and coverage probability were used. As a result of the study, the performance of the 3-step ML and BCH-LTA approaches was generally higher than that of the 1-step approach, resulting in a more accurate estimation. When analyzed with three mixture modeling approaches, the most bias occurred in the 1-step approach. In addition, when the 3-step ML and BCH-LTA approach was confirmed, the performance of the 3-step ML was slightly higher, but similar trends were often shown. Therefore, this study can propose using the BCH-LTA approach and the 3-step ML approach, mainly used for latent transition analysis. In addition, when the analyzed results were put together, it was possible to estimate accurately and efficiently from the BCH-LTA approach if the sample size was more than 500 and the entropy was 0.7 or higher. This study is meaningful in verifying which approach is more suitable for analysis from the performance perspective by comparing the newly introduced BCH-LTA approach and the previously performed 1-step and 3-step ML approaches in the latent transition analysis model with the predictor. In addition, if the BCH-LTA approach is utilized to analyze the latent transition analysis model, it is expected that it can contribute to the usage of the mixture modeling approach by providing guidelines to confirm the minimum data conditions. 예측변인이 있는 잠재전이분석의 혼합모형 접근법은 1단계와 3단계 ML 접근법이 존재했는데, 최근 BCH-LTA 접근법이 추가되면서 잠재전이분석 방법론의 범위가 확장되었다. 따라서, 사용할 수 있는 잠재전이분석의 접근법에는 세 가지가 있다. 첫째, 1단계 접근법은 각 잠재계층을 분류하는 측정모형과 보조변수와의 관계를 살피는 구조모형을 동시에 분석하는 방법이다. 둘째, 3단계 ML 접근법은 지표변수로 잠재계층을 분류하고 개인을 나뉜 잠재계층에 할당한 뒤, 각자 배정된 잠재계층과 예측변인 사이의 관계를 확인하는 방법이다. 셋째, BCH-LTA 접근법은 3단계 BCH 접근법을 잠재전이분석에 적용한 방법론으로, 모든 잠재계층변수의 BCH 가중치를 저장한 후에 측정모형에 따른 BCH 가중치가 저장되고, 결합한 BCH 가중치를 얻기 위해 각 BCH 가중치가 함께 곱해진 값이 추정된다. 본 연구의 목적은 새로운 접근법인 BCH-LTA가 기존의 접근법인 1단계 및 3단계 ML 접근법과 어떤 차이점이 있는지 비교하고, BCH-LTA 접근법을 사용하기 위한 최소한의 조건을 알아보는 것이다. 따라서, 몬테카를로 시뮬레이션 연구를 통해 확인하였다. 연구 수행을 위해 다음의 시뮬레이션 모형을 정의하였다. 선행연구를 고려하여 두 시점에 각각 두 개의 잠재계층이 존재하고, 시점1에는 동일한 비율(50% : 50%)이고 시점2에는 동일하지 않은 비율(70% : 30%)로 설정하였다. 이때, 측정동일성은 필수조건이 아니므로 고려하지 않았다. 지표변수는 이분형 변수 6개를, 예측변인은 연속형 변수 1개를 설정하였다. 본 모형을 분석하기 위해 데이터세트를 생성했으며, 혼합모형 접근법(1단계, 3단계 ML, BCH-LTA), 표본크기(200, 500, 1000), 엔트로피(0.5, 0.7, 0.9)를 고려해 100번 반복하여 총 2,700개의 데이터세트를 생성하였다. 준거 기준으로는 편향(모수 추정, 표준 오차), 평균 제곱 오차, 범위 확률을 사용하였다. 연구 결과, 1단계에 비해 3단계 ML과 BCH-LTA 접근법의 수행력이 대체로 높아 더욱 정확한 추정이 나타났다. 세 종류의 접근법으로 분석했을 때, 1단계 접근법에서 가장 많은 편향이 발생하였다. 또한, 3단계 ML과 BCH-LTA 접근법의 수행력을 확인했을 때, 근소하게 3단계 ML의 수행력이 높았으나, 주로 유사한 경향이 나타났다. 따라서, 본 연구를 통해 잠재전이분석에 주로 사용되는 3단계 ML 접근법과 더불어 BCH-LTA 접근법의 사용을 제안할 수 있었다. 그뿐만 아니라, 분석 결과를 종합했을 때, 표본크기가 500 이상이고, 엔트로피가 0.7 이상이면 BCH-LTA 접근법에서 정확하고 효율적으로 추정할 수 있었다. 본 연구는 예측변인을 포함한 잠재전이분석에서 기존에 수행되었던 1단계와 3단계 ML 접근법의 비교뿐만 아니라 새롭게 소개된 BCH-LTA 접근법을 함께 비교하여, 어떤 접근법이 수행력 관점에서 분석에 더 적합한 방법론인지를 검증하였다는 의의가 있다. 또한, 잠재전이분석 모형을 분석할 때 최소한의 데이터 조건을 확인할 수 있는 BCH-LTA 접근법의 가이드라인을 제공하여 연구자들의 혼합모형 접근법 활용에 기여할 수 있을 것으로 기대한다.

(A) latent transition analysis (LTA) of vocational identity statuses among Korean college students in school-to-work transition

이은정 Graduate School, Korea University 2014 국내석사

Identity formation with respect to career is one of the most crucial developmental tasks one faces (Erikson, 1968). Especially, in terms of industrialized countries, acquiring a vocational identity (Porfeli, Lee, Vondracek, &Weigold, 2011) is important for adolescents and young adults to move forward and achieve one’s career goal. In this context, the current study confirmed vocational identity statuses of Korean college students in school-to-work transition with the Vocational Identity Status Assessment (VISA), and examined the patterns of vocational identity transitions at a six-month interval. Moreover, underlying factors that may lead to or result from vocational identity transition were investigated. Statistical methods were used to derive results. First, to confirm the vocational identity statuses in school-to-work transition, a Latent Profile Analysis (LPA) was implemented separately before (Time 1) and after (Time 2) the last semester of college. As a result, numbers of vocational identity statuses of Korean college students at Time 1 and Time 2 were inconsistent. At Time 1, six different vocational identity statuses were yielded: Achieved, Searching Moratorium, Moratorium, Foreclosed, Diffused, and Undifferentiated. On the other hand, at Time 2, only three statuses were detected: Achieved, Searching Moratorium, and Undifferentiated. Second, to investigate the patterns of vocational identity transitions across time, a Latent Transition Analysis (LTA) was applied based on the results of LPA. LTA detected three different transition patterns, trajectories from the initial six vocational identity statuses to the Achieved, Searching Moratorium, and Undifferentiated status. Furthermore, the findings presented that individuals were biased to lower identity statuses (i.e. Diffused and Undifferentiated) at Time 1and concentrated to upper statuses (i.e. Achieved and Searching Moratorium) after a six months interval. Third and lastly, to reveal underlying mechanisms of transition patterns, an one-way analysis of variance (ANOVA) was conducted in terms of different aspects such as self-perception (i.e. career decision-making self-efficacy), attitude (i.e. planned happenstance skills), and behavior (i.e. occupational engagement) as well as outcome (i.e. change in career satisfaction). Results of the one-way ANOVA indicated that initial levels of planned happenstance skills, career decision-making self-efficacy, and occupational engagement were respectively higher in the individuals who progressed to the Achieved and Searching Moratorium status than those who transferred to the Undifferentiated status. Likewise, those who moved to the Achieved group reported higher level of change in career satisfaction than the participants who progressed to the Undifferentiated group. Some implications were noted. First of all, this study improved validity of the Vocational Identity Status Assessment as findings revealed that Korean young adults were properly assigned into six different statuses. Second, outcomes offered important information regarding early adulthood career development processes as well as guidelines for career theorists and career counselors. Finally, it was remarkable that the present research applied statistical methods as LPA and LTA that compensated problems of traditional methods. 직업세계에 대한 이해를 확립하고 진로정체감을 형성하는 것은 한 개인에게 있어 매우 중요한 과제이다. 특히, 청소년기와 초기 성인기의 진로정체감 형성은 진로목표 달성 여부와 관계가 깊으며 가장 주요한 발달과업 중 한가지로 꼽히기도 한다. 이러한 맥락에서 본 연구는 사회진출전횐기 대학생의 진로정체감 변화 및 특성을 살펴보고자 하였다. 이에 본 연구는 통계적 방법을 사용하여 다음과 같은 결과를 확인하였다. 먼저 사회진출전환기 대학생들의 진로정체감 지위는 6개월의 시간차를 두고 VISA(Vocational Identity Status Assessment)를 사용하여 알아보았다. 그리고 잠재프로파일분석(LPA: Latent Profile Analysis) 방법을 두 시점에 적용함으로써 각각 6개(성취, 유예, 탐색유예, 유실, 혼란, 미분화) 그리고3개(성취, 탐색유예, 혼란)의 정체감 지위를 도출하였다. 다음으로 시간에 따른 진로정체감 지위의 변화는 잠재전이분석(LTA: Latent Transition Analysis) 방법을 통하여 알아보았다. 그 결과 성취, 탐색유예, 그리고 미분화로 이동하는 세 가지 형태의 경로가 확인되었다. 마지막으로, 이동경로에 따른 자기인식, 태도, 행동, 그리고 진로만족 변화를 살펴보았을 때, 성취와 탐색유예로 이동한 학생들이 미분화로 이동한 학생들보다 진로선택과정에서 우연 요인에 대한 개방적 태도를 취하며, 진로 결정에 대한 자기효능감을 높게 인식한 다는 것을 알 수 있었다. 또 이들은 진로관여적 행동을 더 활발하게 하였다. 반면 진로만족 변화는 성취 집단이 미분화 집단보다 높은 것으로 나타났다. 이와 같은 결과는 몇 가지 연구적 그리고 상담학적 의의를 제시한다. 첫째, 본 연구는 VISA(Vocational Identity Status Assessment)라는 측정도구를 사용하였다. LPA의 첫 시점에서 개발 당시와 같은 형태의 결과가 도출됨으로써, VISA의 타당성이 일정부분 확보되었다. 둘째, 연구결과는 초기 성인기의 진로정체감 발달 과정에 대한 정보를 제공하며 진로연구가뿐만 아니라 진로상담가에게도 중요한 가이드라인을 제공한다. 마지막으로 본 연구는 기존의 통계적 방법의 여러 가지 한계를 보완한 분석방법인 잠재프로파일분석과 잠재전이분석을 사용했다는 점에서 의미를 갖는다.

Detecting the changes of latent profiles in academic burnout of middle school students

신효정 Graduate School, Korea University 2012 국내박사

This study used latent transition analysis (LTA) to examine changes in academic burnout profiles during middle school times (grade 7th-2, 8th-1 and 8th-2). In addition, this study was to examine the effects of the covariates (i.e., gender, parent education, and coping strategy) in LTA model. To do so, the first goal was to identify profile of academic burnout for middle school students at each time point based on Maslach Burnout Inventory ? Student Survey (MBI-SS). For first aim, the following research question was presented: What kinds of burnout profiles would be identified among middle school students? The second goal was to describe developmental pattern of academic burnout throughout middle school using the LTA model. For the second aim, the following research question was presented: How do those profiles change over time? Finally, the third goal was to confirm effects of covariates for modeling change in academic profiles in LTA model. For last aim, the following research question was presented: What are the effects of covariate variables (i.e., gender, parent education, and coping strategy) on changes of academic burnout profiles over time? Total of 413 middle school students (TI: N=381, T2: N=375, T3: N=388) in Seoul, Korea participated in this study. The data was collected from one middle school of Seoul region of South Korea during three waves. Korean version of MBI-SS and Ways of Coping Checklist (WCC) were used to measure the levels of students? academic burnout and coping strategy. All instruments in the present study measured in the final of the semester. Mplus 5.0 and SPSS 18.0 were used to analyze the data. The results of the present study showed that the middle school students could be best described by three latent profiles representing degree of academic burnout across three times: ?distressed (DS)?, ?intermediate (IM)?, and ?well-functioning (WF)?. DS is high scores on all subscales although cynicism was slightly higher score than emotional exhaustion and inefficacy. IM is maintaining medium score on all subscales. WF is low scores on all subscales although emotional exhaustion and inefficacy was slightly higher score than cynicism. The result of the current study proved evidence that academic burnout profiles are understood according to the degree, rather than type of profiles, of academic burnout during the middle school years. In the next, results indicated that as student developed throughout middle school they were likely to transition out of the WF group and into the DS and IM group. Specially, non burned-out group, that is WF group, transited to IM group and even DS group. In addition, distressed group of the academic burnout showed stability, that is, less movement to another groups than the IM and WF groups. Further, the result showed that both active coping and passive coping consistently differentiated the academic burnout profiles throughout the course of middle school. Active coping was associated with higher percentage being in the WF versus DS. Conversely, passive coping was associated with decreased odds of being in the WF. Based on the results of the present study, there are several implications. Specifically, the present study can be utilized for designing prevention and intervention programs for middle school students as well as teacher and parents? training program for student discipline in school and counseling settings. First, helping professionals such as school counselors, teacher could understand the diverse types of academic burnout as well as conduct academic burnout interventions (e.g., preventing or decreasing students? academic burnout). Second, school counselors, teachers and parents should keen attention to WF group during middle school period. Because WF group students transited intermediate group very much over middle school time. These results imply that although students are adapting easily at the moment helping professional need to conduct prevention program for general students. To do so, helping professionals could plan prevention strategies through classroom guidance focusing on WF group students. In addition, since DS group of the academic burnout showed stability, professionals have to provide intensive treatment for being DS group students. Third, this result implies that helping professionals and parents should help to prompt active coping and refrain from passive coping. How to instruct effective coping strategy to manage academic burnout for students is very important. The study implies that teachers and parents? training to support students, as resources, could be good way to prompt active coping for middle school students. Moreover, it is crucial to implement prevention program to learn active coping and passive coping through the classroom guidance regularly. There are several limitations associated with the present study. First, it is important to note that academic profiles were different with previous study in this study. Thus, further research is needed in order to identify the academic profiles with other populations for middle school student. Second, in the present study, the results were obtained for predominantly one middle school students in an urban community. These results may differ for other samples with different characteristics (e.g., gender, grade, academic pressure). Thus, it would be useful to replicate the results in larger samples. Additionally, it would be interesting to explore the special samples of chronic academic burnout groups having high academic demanding for middle school students (e.g., international middle school) in longitudinal designs. Third, the time interval of the study was relatively short (six months). These reason resulted in little changes of levels of each burnout profiles in this study. Thus, future study need to extended periods. The lastly, all instruments in the present research were self-reported questionnaires. In the case of self-reported questionnaires, students who experience high level of academic burnout may report low levels of burnout symptoms to minimize or maximize their symptoms. Therefore, future studies should also consider including biological indicators (e.g., blood pressure, cortisol levels) of burnout and stress to help offset the use of self-reported instruments.

Longitudinal transition patterns of symptom clusters and health outcomes in patients with lung cancer surgery

김예솔 Graduate School, Yonsei University 2024 국내박사

폐암은 전 세계적으로 가장 많이 발병하는 암 중 하나이며, 치료를 위해 폐 절제술을 받아 생존율이 점차 증가하고 있다. 그러나 폐 절제술을 받았음에도 불구하고 환자들은 여전히 다양한 증상과 증상 군집으로 고통받고 있어, 시간의 경과에 따른 증상과 증상 군집의 변화를 파악하는 것에 대한 중요함이 강조되고 있다. 또한 증상과 증상 군집은 폐암 환자의 건강 결과에 부정적인 영향을 미치는 것으로 나타났다. 따라서 본 연구는 폐 절제술을 받은 초기 폐암 환자를 대상으로 시간 경과에 따른 수술 후 증상, 증상 군집의 유형 및 전이 패턴을 파악하고, 이들이 수술 12주 후 건강 결과에 미치는 영향에 대해 조사하고자 한다. 본 연구는 전향적 종단 연구 설계로서 퇴원 후 첫 외래 진료 방문 시(T1), 수술 4주 후(T2), 8주 후(T3), 12주 후(T4)와 같이 총 4회 자료수집을 실시하였다. T1에서는 병원에서 대면으로 자가보고 설문지를 통해 자료를 수집하였고, T2부터 T4는 대상자가 병원을 방문하지 않는 시기임을 고려하여 모바일 폰으로 접속 가능한 온라인 설문지를 통해 자료를 수집하였다. 연구 대상자는 서울 소재 3차 병원에서 병리학적으로 폐암을 진단받고 폐 절제술을 받은 만 19세 이상의 환자이다. 16가지 증상으로 구성된 MD Anderson Symptom Inventory - Lung Cancer를 이용하여 증상을 측정하였고, 점수가 높을수록 증상이 심함을 의미한다. 건강 결과는 삶의 질과 치료 복귀로 구성되었다. 삶의 질은 Functional Assessment of Cancer Therapy – General을 이용하여 측정하였고, 점수가 높을수록 삶의 질이 좋음을 의미한다. 치료 복귀는 전자 의무 기록을 검토하여 계획되지 않은 외래 방문, 응급실 방문, 재입원이 포함되었다. 잠재 프로파일 분석은 16개 증상을 기준으로 각 시점의 증상 군집을 분류하기 위해 수행되었고, 잠재 전이 분석은 퇴원 후 첫 외래 진료 방문 시점과 수술 12주 후 시점 사이 증상 군집의 전이 패턴을 파악하기 위해 수행되었다. 건강 결과와 관련된 요인을 파악하기 위해 선형 및 로지스틱 회귀 분석이 수행되었다. 본 연구에서는 SPSS Statistics version 26, Mplus version 8.8, R version 4.3.2 프로그램을 이용하여 자료를 분석하였다. 본 연구에 등록된 177명의 대상자 중 T1에 163명, T2에 138명, T3에 120명, T4에 101명이 자료 분석에 포함되었다. 대상자의 평균 연령은 61.99 ± 9.61세였고, 60.1%가 여성이었다. T1에서 심각도가 높은 5개의 증상으로는 피로, 기침, 통증, 숨가쁨, 졸음 순이었다. 이 중 피로는 네 시점 모두 심각도가 가장 높게 나타났다. 증상 군집은 각 시점의 증상 심각도 점수에 따라 낮은 증상 군집, 중간 증상 군집, 높은 증상 군집과 같이 세 유형으로 분류되었고, 네 시점 모두 낮은 증상 군집에 많은 대상자가 포함되었다. T1에서 낮은 증상 군집과 중간 증상 군집에 속한 대상자는 각각 T4에서 동일한 유형에 속할 가능성이 높았으나, T1에서 높은 증상 군집에 속한 대상자는 T4에서 중간 또는 높은 증상 군집에 속할 가능성이 높았다. T1과 T4 사이 증상 군집의 전이 패턴으로는 체류자(50.5%), 악화된 이동자(29.7%), 개선된 이동자(19.8%)로 분류되었다. T4에서 삶의 질은 108점 만점에 74.31 ± 17.15점이었고, 치료 복귀는 총 13건(8.4%) 발생하였으며, 그 중 계획되지 않은 외래 방문이 가장 많았다. 증상 군집 유형이 중간(B = -17.00, p < .001) 및 높은 경우(B = -26.85, p < .001) 삶의 질이 유의하게 낮게 나타났다. 증상 군집 유형의 전이 패턴이 악화된 이동자인 경우 삶의 질이 유의하게 낮게 나타났고(B = -12.03, p < .001), 개선된 이동자인 경우 유의하게 높게 나타났다(B = 11.58, p = .012). 그러나 치료 복귀와 관련된 유의한 요인은 없었다. 본 연구는 수술 후 회복 기간 동안 증상, 증상 군집, 전이 패턴 및 건강 결과에 대한 포괄적인 이해를 얻기 위해 종단적 접근 방식을 수행하였다. 본 연구는 상대적으로 증상이 경미할 것으로 예상되는 초기 폐암 환자를 대상으로 수술 12주 후까지 증상 군집을 개인 중심적 접근 방식으로 파악함으로써, 시간이 경과하여도 여전히 증상 심각도가 높고 증상 악화를 경험하는 환자가 있다는 것을 발견하였다. 또한 증상 군집 유형 및 전이 패턴은 삶의 질에 유의한 영향을 미치며, 이는 수술 후 회복 기간 동안 증상 관리의 중요성을 강조한다. 따라서 고위험 증상군을 조기에 선별하고 중재를 제공하는 것은 수술 후 폐암 환자의 증상 완화 및 삶의 질 향상에 기여할 수 있다. Introduction: Lung cancer is the most prevalent cancers worldwide. The survival rate of lung cancer is gradually increasing to lung resection surgery. Despite surgery, patients suffer from various symptoms and symptom clusters (SxCls), highlighting the importance of identifying these changes over time. Further, these symptoms and SxCls negatively impact health outcomes in patients with lung cancer. Therefore, this study aimed to identify postoperative symptoms, the types of SxCls, and their transition patterns over time, and to investigate their impact on health outcomes at 12 weeks after surgery in patients with early-stage lung cancer who underwent lung resection. Methods: This was a prospective longitudinal study conducted at the first outpatient clinic visit after discharge (T1), 4 weeks after surgery (T2), 8 weeks after surgery (T3), and 12 weeks after surgery (T4). At T1, data were collected through face-to-face self-reported surveys at the hospital, while from T2 to T4, data were collected using online surveys accessible via mobile phones. Inclusion criteria were age ≥ 19 years, with lung cancer diagnosed pathologically and underwent lung resection at a tertiary hospital in Seoul, South Korea. Symptoms were measured using the MD Anderson Symptom Inventory-Lung Cancer, which comprised of 16 symptoms, with higher scores indicating more severe symptoms. The health outcomes included quality of life (QOL) and return to care (RTC). QOL was measured using the Functional Assessment of Cancer Therapy-General, consisting of 27 items, with higher scores indicating a higher QOL. RTC included unplanned outpatient clinic visits, emergency department visits, and readmissions, as identified through the electronic medical records. Latent profile analysis was conducted to identify potential SxCls at each time point. Latent transition analysis was conducted to identify transition patterns of SxCls between T1 and T4. Linear and logistic regression analyses were conducted to identify factors associated with health outcomes. All data were analyzed using SPSS Statistics version 26, Mplus version 8.8, and R version 4.3.2. Results: Of the 177 participants enrolled in this study, 163 at T1, 138 at T2, 120 at T3, and 101 at T4 were included in the data analysis. Participants had a mean age of 61.99 ± 9.61 years; 60.1% were women. The top five symptoms with high severity at T1 were fatigue, coughing, pain, shortness of breath, and drowsy. Among the 16 symptoms, fatigue had the highest severity at all four time points. Three types of SxCls were categorized according to the symptom severity scores at each time point: low SxCl, medium SxCl, and high SxCl. Most participants were categorized in the low SxCl at all time points. Participants belonging to low and medium SxCls at T1 were likely to belong to the same type at T4. However, those with high SxCl at T1 were likely to belong to medium or high SxCls at T4. The transition patterns of SxCls between T1 and T4 were classified as stayers (50.5%), worsened movers (29.7%), and improved movers (19.8%). Participants’ QOL score at T4 was a mean of 74.31 ± 17.15 out of 108. RTC occurred in 13 cases (8.4%) until T4; among them, unplanned outpatient clinic visits occurred the most. Lower QOL was significantly associated with the medium (B = –17.00, p < .001) and high SxCls (B = –26.85, p < .001). Worsened mover was significantly associated with lower QOL (B = –12.03, p < .001); improved mover was significantly associated with higher QOL (B = 11.58, p = .012). However, no significant factors were associated with the occurrence of RTC. Conclusion: This longitudinal study was conducted by taking a person-centered approach to identify SxCls until 12 weeks after surgery in patients with early-stage lung cancer who underwent lung resection expected to have relatively mild symptoms. Some patients still experienced high-severity symptoms and worsening symptoms over time. In addition, SxCls and transition patterns significantly influenced the QOL. Therefore, early screening of high-risk symptom groups and provision of intervention may contribute to alleviating symptoms and ultimately improving QOL.

Detecting the Changes of Latent Profiles In Academic Burnout of Middle School Students

Shin, Hyo Jung 고려대학교 대학원 2012 국내박사

This study used latent transition analysis (LTA) to examine changes in academic burnout profiles during middle school times (grade 7th-2, 8th-1 and 8th-2). In addition, this study was to examine the effects of the covariates (i.e., gender, parent education, and coping strategy) in LTA model. To do so, the first goal was to identify profile of academic burnout for middle school students at each time point based on Maslach Burnout Inventory ? Student Survey (MBI-SS). For first aim, the following research question was presented: What kinds of burnout profiles would be identified among middle school students? The second goal was to describe developmental pattern of academic burnout throughout middle school using the LTA model. For the second aim, the following research question was presented: How do those profiles change over time? Finally, the third goal was to confirm effects of covariates for modeling change in academic profiles in LTA model. For last aim, the following research question was presented: What are the effects of covariate variables (i.e., gender, parent education, and coping strategy) on changes of academic burnout profiles over time? Total of 413 middle school students (TI: N=381, T2: N=375, T3: N=388) in Seoul, Korea participated in this study. The data was collected from one middle school of Seoul region of South Korea during three waves. Korean version of MBI-SS and Ways of Coping Checklist (WCC) were used to measure the levels of students? academic burnout and coping strategy. All instruments in the present study measured in the final of the semester. Mplus 5.0 and SPSS 18.0 were used to analyze the data. The results of the present study showed that the middle school students could be best described by three latent profiles representing degree of academic burnout across three times: ?distressed (DS)?, ?intermediate (IM)?, and ?well-functioning (WF)?. DS is high scores on all subscales although cynicism was slightly higher score than emotional exhaustion and inefficacy. IM is maintaining medium score on all subscales. WF is low scores on all subscales although emotional exhaustion and inefficacy was slightly higher score than cynicism. The result of the current study proved evidence that academic burnout profiles are understood according to the degree, rather than type of profiles, of academic burnout during the middle school years. In the next, results indicated that as student developed throughout middle school they were likely to transition out of the WF group and into the DS and IM group. Specially, non burned-out group, that is WF group, transited to IM group and even DS group. In addition, distressed group of the academic burnout showed stability, that is, less movement to another groups than the IM and WF groups. Further, the result showed that both active coping and passive coping consistently differentiated the academic burnout profiles throughout the course of middle school. Active coping was associated with higher percentage being in the WF versus DS. Conversely, passive coping was associated with decreased odds of being in the WF. Based on the results of the present study, there are several implications. Specifically, the present study can be utilized for designing prevention and intervention programs for middle school students as well as teacher and parents? training program for student discipline in school and counseling settings. First, helping professionals such as school counselors, teacher could understand the diverse types of academic burnout as well as conduct academic burnout interventions (e.g., preventing or decreasing students? academic burnout). Second, school counselors, teachers and parents should keen attention to WF group during middle school period. Because WF group students transited intermediate group very much over middle school time. These results imply that although students are adapting easily at the moment helping professional need to conduct prevention program for general students. To do so, helping professionals could plan prevention strategies through classroom guidance focusing on WF group students. In addition, since DS group of the academic burnout showed stability, professionals have to provide intensive treatment for being DS group students. Third, this result implies that helping professionals and parents should help to prompt active coping and refrain from passive coping. How to instruct effective coping strategy to manage academic burnout for students is very important. The study implies that teachers and parents? training to support students, as resources, could be good way to prompt active coping for middle school students. Moreover, it is crucial to implement prevention program to learn active coping and passive coping through the classroom guidance regularly. There are several limitations associated with the present study. First, it is important to note that academic profiles were different with previous study in this study. Thus, further research is needed in order to identify the academic profiles with other populations for middle school student. Second, in the present study, the results were obtained for predominantly one middle school students in an urban community. These results may differ for other samples with different characteristics (e.g., gender, grade, academic pressure). Thus, it would be useful to replicate the results in larger samples. Additionally, it would be interesting to explore the special samples of chronic academic burnout groups having high academic demanding for middle school students (e.g., international middle school) in longitudinal designs. Third, the time interval of the study was relatively short (six months). These reason resulted in little changes of levels of each burnout profiles in this study. Thus, future study need to extended periods. The lastly, all instruments in the present research were self-reported questionnaires. In the case of self-reported questionnaires, students who experience high level of academic burnout may report low levels of burnout symptoms to minimize or maximize their symptoms. Therefore, future studies should also consider including biological indicators (e.g., blood pressure, cortisol levels) of burnout and stress to help offset the use of self-reported instruments.

취업 준비 노력이 고용의 질에 미치는 영향 : 3단계 접근법을 통한 잠재 전이 분석의 적용

이한성 중앙대학교 대학원 2022 국내석사

본 연구의 목적은 3단계 접근법을 통한 잠재 전이 분석을 소개하고 대학졸업자를 대상으로 직업 관련 설문을 실시한 대졸자 직업이동 경로조사(GOMS)에 적용함에 있다. 구체적으로, 고용의 질 잠재 프로파일에 취업 준비 노력 잠재 계층이 미치는 영향, 그리고 고용의 질 유형에 따른 이직 의도 및 준비 행동의 차이를 파악하고자 하였다. 먼저, 고용의 질에 따른 잠재 프로파일 분석 결과, 고용의 질 하위 차원의 질적 및 양적 차이에 따라 이상적 일자리, 일반적 일자리, 불안정 일자리, 도구적 일자리, 좋지 않은 일자리로 총 5가지 유형이 분류되었다. 이후, 취업 준비 노력에 따라 잠재 계층 분석을 실시한 결과, 어학·인턴형, 인서울형, 학점·자격증형, 미비형으로 총 4가지 유형이 분류되었다. 도출된 측정 모형을 바탕으로 취업 준비 노력이 고용의 질에 미치는 영향을 파악하기 위해 두 측정 모형 간의 관계가 각 측정 모형의 모수에 영향을 주지 않는 3단계 접근법을 통한 잠재 전이 분석을 시행하였다. 분석 결과, 전반적인 전이 현상이 미미하였으며, 다항 로지스틱 회귀 분석 결과를 살펴본 결과, 미비형 집단에 속한 이들이 다른 집단에 비해 이상적 일자리와 비교하여 좋지 않은 일자리 또는 도구적 일자리에 속할 확률이 높았으며, 그 외의 관계는 유의하지 않았다. 마지막으로, 고용의 질 잠재 프로파일과 이직 의도 및 준비 행동과의 관계에서는 전반적으로 고용의 질 수준이 낮은 집단에 속할수록 이직 의도 및 준비 행동이 높은 경향을 보였다. 예외적으로, 좋지 않은 일자리와 도구적 일자리는 이직 의도에서는 차이가 유의하지 않았지만, 도구적 일자리에 속한 이들이 더 낮은 이직 준비 행동을 보였다. 본 연구는 3단계 접근법을 활용한 잠재 전이 분석을 횡단면 자료에 적용하여 유의미한 결과를 이끌어냈다는 점, 청년 실업 문제와 관련해 고용의 질 현황을 파악하고 취업 준비 노력과의 관계를 살펴보았다는 점, 더 나아가 고용의 질에 따른 이직 의도 및 준비 행동의 차이를 살펴보았다는 점에서 의의가 있다. The purposes of this study are to introduce latent transition analysis with a three-step approach and to apply it to the GOMS(Graduate Occupational Mobility Survey) data which surveyed the college graduates about a job-related issue. Specifically, we examined the effect of the latent class of job preparation efforts on the employment quality profiles, as well as differences in turnover intention and turnover preparation behavior in light of the employment quality profiles. First, as a result of latent profile analysis based on the employment quality, we found five profiles representing qualitative and quantitative differences in the lower dimensions of employment quality: Ideal job, Ordinary job, Unstable job, Instrumental job, and Bad job. After that, latent class analysis was conducted according to job preparation efforts. Four classes were derived: English- Internship class, Graduates in Seoul class, GPA-Certification class, and Unprepared class. Based on the derived measurement model, we conducted latent transition analysis with a three-step specification: the relationship between the two measurement models did not affect the parameters of either measurement model, thus enabling us to assess the effect of job preparation efforts on employment quality. In response to the latent transition analysis, given that the overall transition was not noticeable, we examined multinomial logistic regression. This showed that those in the Unprepared class were more likely to belong to Bad or Instrumental job compared to other classes, while other relationships were not significant. Finally, as regards the relationships between employment quality profiles and turnover intention or preparation behavior, our result is: The lower the overall employment quality of the profile, the higher the turnover intention and preparation behavior. Interestingly, there was no significant difference in turnover intention between Bad job and Instrumental job, but those belonging to Instrumental job exhibited lower turnover preparation behavior. This study has three implications. First, we applied latent transition analysis using the three-step approach to cross-sectional data in order to derive meaningful results. Second, we identified the current status of employment quality in relation to the youth unemployment problem, and its connections to job preparation efforts were examined. Furthermore, we examined the differences in turnover intention and preparation behavior as these relate to employment quality.