Search

Mediation analysis occupies a unique place in the social science literature. It allows researchers to go beyond testing whether an independent variable has an effect on a dependent variable; rather, it allows the researcher to examine the process or mechanism that enables that effect to occur. In its simplest form, a mediation model tests the effect that an independent variable, X, has on a dependent variable, Y, through its effect on an intervening variable, M. In this way, it provides a more complete and nuanced description of the relationship between X and Y. Researchers working within the field of education are often required to work with data that are not independent due to how the data are typically structured. The lack of independence inherent in the majority of educational data is caused by the multilevel structure of observations (e.g., students nested within classrooms, teachers nested within schools, and/or longitudinal observations, to name a few). This requires researchers to adjust their methods of analysis in order to properly model this lack of independence at the lowest level of the hierarchy. The estimation of mediated relationships becomes complicated when working in a multilevel context. Traditional approaches to multilevel modeling (MLM) can be used to model some mediated relationships, but it carries multiple limitations. A latent variable approach using multilevel structural equation modeling (MSEM) overcomes many of the limitations found in the MLM approach. This Monte Carlo simulation study examined the performance of MSEM when estimating two specific types of multilevel mediated relationships (the 2-1-1 and 2-1-2 mediation design configurations). Simulations were performed to investigate how well MSEM was able to estimate the indirect effect of X on Y through M in these two configurations under various ICCs, indirect effect sizes, number of clusters, and levels of cluster size imbalance. Model performance was assessed by model convergence rates, levels of bias in the parameter estimate or its standard error, estimate efficiency, confidence interval coverage, Type-I error rate, and statistical power. Results indicated that the design configuration had an effect on model performance through its interaction with the other design factors. Specifically, the 2-1-1 model tended to perform worse than the 2-1-2 model in terms of model convergence rates and bias in the standard error of the indirect effect under conditions of low ICC and low number clusters. In line with previous studies of multilevel mediation and of MSEM in general, low ICC was the most common design factor that led to poor performance on almost all outcome measures and was especially problematic when combined with a small number of clusters. The magnitude of the indirect effect, and often how that indirect effect was formed, also impacted model convergence, the amount of bias in the estimate of the indirect effect, estimate efficiency, confidence interval coverage, Type-I error rate, and statistical power. Additionally, results indicated that cluster size imbalance did not impact model performance under the selected study conditions.

The purpose of this study was to investigate whether a virtual reality intervention could serve as a relational maintenance tool for older adults living in a residential care community. The current study extends the literature on older adults and technology by suggesting that residents can provide one another relational maintenance and communal orientation, which can then potentially help promote such positive outcomes as vitality, social activity, and quality communication, and lessen such negative outcomes as depression, stress, and loneliness. The study took place in a residential care community in Santa Barbara, CA. Residents were randomly assigned to participate once a week for four weeks in either a virtual reality or television group. Analyses were conducted in R and consisted of multilevel mediation with bootstrapping. Results indicate that the virtual reality intervention was positively associated with higher levels of vitality and lower levels of loneliness. The study integrated novel ways of understanding older adults’ living in a residential care community, new methodological techniques, and new insights into important relational processes outlined in the theory of resilience and relational load (Afifi, Merrill, & Davis, 2016)

Researchers in the social sciences are often interested in explaining causal processes in which the effect of one variable is transmitted to another through one or more intervening (or mediator) variables. Statistical mediation analysis can be used by researchers to gain a better understanding of these causal processes. The methods traditionally used to quantify and test a mediation process, or indirect effect, are not directly applicable when the data are nested hierarchically, which is common in educational, organizational, and behavioral sciences where students may be nested within classrooms, employees may be nested within companies, or repeated measurements may be nested within individuals. Multilevel modeling (MLM) is a method for analyzing hierarchical data that allows for the simultaneous estimation of the effects of variables at multiple levels of the hierarchy. As such, several researchers have proposed methods for testing mediation using MLM. A majority of these methods have focused on models in which the independent variable is a level-2 variable, the mediator is either a level-2 or a level-1 variable, and the dependent variable is a level-1 variable. Although some (Kenny, Korchmaros, & Bolger, 2003; Bauer, Preacher, & Gil, 2006) have focused on models in which all three variables are measured at the lowest level, the methods proposed are not without their shortcomings. The aim of this thesis is to address and mitigate some of these shortcomings, as well as provide advancements in the formulation and testing of a number of interesting effects that can be modeled using MLM. The major contributions include demonstrating how to simultaneously estimate between-group and within-group indirect effects using traditional MLM software. A test for the difference between within-group and between-group indirect effects is also conceptualized and explained. Further, the model is expanded to include multiple mediators with either fixed or random slopes. In the case where one or more paths for one indirect effect covary with one or more paths from another, the covariance between the random indirect effects is derived. Finally, the index of moderated mediation (Hayes, 2015) is applied to the multilevel context to test the moderating effect of a level-2 variable on the within-group and/or between-group indirect effect. To demonstrate the methods discussed throughout this thesis, two real-world datasets are analyzed using MLmed, a free SPSS macro developed in conjunction with this thesis and designed specifically to fit multilevel mediation and moderated mediation models. The parameter estimates obtained using MLmed are comparable to those obtained using Mplus (Muthen & Muthen, 2015) and the lme4 package (Bates, Machler, Bolker, & Walker, 2015) in R (R Core Team, 2016), but with considerably less programming effort required.

Nostalgia is best described as a sentimental and bittersweet yearning for a positive and pleasant past, particularly when juxtaposed with an unsatisfying present and uncertain future. One's positive memories surely influence the evocation of nostalgia, and an individual's negative feelings for the present or future are also related to nostalgia, since a person cannot return to the past. In other words, both positive and negative feelings are associated with nostalgia, and it is called a bittersweet emotion. In many cases, people are influenced by their past memories when they decide to attend sports events. As a result, individuals have their own attitudes based on their past memories, and it may affect individual's behavioral intentions. The NCAA football game is one of the most popular and historic sporting events in United States. According to U. S. Census Bureau (n.d.), the NCAA College football drew the second largest number of spectators in the United States from 1990 to 2010. The population of this study is people who attended Clemson football games. Specifically, this study surveyed participants who have had a positive past experience at the Clemson football home games, and a systematic sampling technique was used for gathering the data. The goal of this study was to develop a comprehensive conceptual framework of nostalgia in the context of sport tourism and to provide a valid and reliable nostalgia scale for sport tourism (NSST) based on a suggested classification of nostalgia in sport tourism. Another aim of this study was to verify the developed nostalgia scale for sport tourism by testing the relationship among nostalgia (independent variable), attitude (mediating variable), and behavioral intentions (dependent variable). To clarify the group effects, this study uses multilevel structural equation modeling. The results of this study indicated that attitude mediates the relationship between nostalgia and behavioral intentions in multilevel structural equation model. In addition, this study discusses how nostalgia plays a role in sport tourism and suggests the direction for sustainable development of nostalgia sport tourism.

Handling of clustered or nested data structures requires the use of multilevel modeling techniques. One such multilevel modeling technique is multilevel structural equation modeling (MLSEM). While estimation of indirect effect parameters and standard errors based on the conventional multilevel model (MMM) has been assessed, this is not the case for the use of the MLSEM model for estimating indirect effects. This simulation study was designed to investigate the use of the MLSEM for estimating mediated effects for the “upper-level” mediation model as compared with the MMM. The following conditions were manipulated: number of clusters (G), within-cluster sample size (nj ), intra-class correlation, measurement error in the mediator, and the true value of the mediated effect derived from various patterns of true values for a and b. The generating model entailed an upper-level mediation model for a cluster-randomized trial that included a dichotomous level two independent variable, a cluster-level latent mediator and an individual-level latent dependent variable both with four indicators. Relative parameter and standard error bias, obtained using the MLSEM and the MMM were evaluated and compared. Percent coverage was calculated and compared when PRODCLIN was used to calculate the confidence interval estimates of the ab effect. Finally, Type I error rates for conditions when ab = 0 were assessed and compared. In addition, statistical power for detecting a truly non-zero mediated effect was tallied and compared across models. Results showed that use of the MMM provided inaccurate and misleading parameter and standard error estimates for the estimates of the mediated effect, especially when the true values of a, b and ab were not zero and the measurement error for M was large. However, the MLSEM estimates were also unacceptable in some of the conditions with small values for G and nj. Researchers are encouraged to use the MLSEM for assessing the multilevel mediated effects when either or both paths a and b are expected to be non-zero, if G is at least 40 and nj is also greater than 40. Results are presented and discussed along with implications for applied researchers intending to assess mediated effect with clustered data.

Handling of clustered or nested data structures requires the use of multilevel modeling techniques. One such multilevel modeling technique is multilevel structural equation modeling (MLSEM). While estimation of indirect effect parameters and standard errors based on the conventional multilevel model (MMM) has been assessed, this is not the case for the use of the MLSEM model for estimating indirect effects. This simulation study was designed to investigate the use of the MLSEM for estimating mediated effects for the “upper-level” mediation model as compared with the MMM. The following conditions were manipulated: number of clusters (G), within-cluster sample size (nj ), intra-class correlation, measurement error in the mediator, and the true value of the mediated effect derived from various patterns of true values for a and b. The generating model entailed an upper-level mediation model for a cluster-randomized trial that included a dichotomous level two independent variable, a cluster-level latent mediator and an individual-level latent dependent variable both with four indicators. Relative parameter and standard error bias, obtained using the MLSEM and the MMM were evaluated and compared. Percent coverage was calculated and compared when PRODCLIN was used to calculate the confidence interval estimates of the ab effect. Finally, Type I error rates for conditions when ab = 0 were assessed and compared. In addition, statistical power for detecting a truly non-zero mediated effect was tallied and compared across models. Results showed that use of the MMM provided inaccurate and misleading parameter and standard error estimates for the estimates of the mediated effect, especially when the true values of a, b and ab were not zero and the measurement error for M was large. However, the MLSEM estimates were also unacceptable in some of the conditions with small values for G and nj. Researchers are encouraged to use the MLSEM for assessing the multilevel mediated effects when either or both paths a and b are expected to be non-zero, if G is at least 40 and nj is also greater than 40. Results are presented and discussed along with implications for applied researchers intending to assess mediated effect with clustered data.