HLM 7--- a good helper for Straight A student

In social science research and other areas, research data often has a hierarchical structure. In other words, a single topic in the study will be divided into different groups that affect the study. At this point, these individual research topics can be considered as Level-1 units of research, and their group is Level-2 units, and can be further extended: Level-2 units can be assigned to the third level of the unit group , And the Level-3 unit can also be assigned to the fourth level of the cell group. In the field of education, this type of example is very rich (students in Level-1, teachers in Level-2, schools in Level-3, school district at Level-4). While in the field of sociology (individual in Level-1, community in Level-2). Obviously, the analysis of this data requires professional software. The design of hierarchical linear and nonlinear models (also known as multilayer models) can be used to analyze each level of relationship in a study without neglecting the variability associated with each hierarchy in the hierarchical model.

Using the variables specified in each hierarchy, the HLM can fit the model into a result variable that can produce a linear model with explanatory variables, and the explanatory variables cause changes at all levels.

HLM can not only estimate the model coefficients at each level, but also predict the random effects of the sampling units in each layer.

Because the data in the field of education generally has a hierarchical structure, HLM is widely used in this field. Of course, it can also be used for other research areas with hierarchical structure data.

This includes longitudinal analysis. In this analysis, individual repetitive measurement data can be nested in the individual under study. In addition, although the above example implies that any member of the hierarchy at this level is completely nested at a higher level of membership, the HLM can also handle crossed member relationships, not only the nested relationships The For example, a student can be a member of multiple classrooms throughout the study period.

The HLM software can handle consecutive, counting, ordinal, and nominal result variables, and assume a functional relationship between the expected value and the linear combination of explanatory variables. This relationship is usually defined by the appropriate association function. For example, identity association (continuous result) or logit association (binary result)

The HLM 7 has unprecedented flexibility in modeling multilayer and vertical data. Like HLM 6, HLM 7 has a series of graphical programs and residual files (file), computing speed, a solid, friendly user interface. In addition, it has three new programs that can handle binary, count, ordinal and multiple (nominal) response variables, and can also handle the normal response of the normal-theory hierarchical linear model:

Four-tier nested model:

Cross-sectional data of the four-tier nested model (for example, different schools in different schools of different students of the project response)

Vertical data four-tier nested model (for example, different communities of different people at different points in the project)

Four cross-categorization and nested mix:

Repeated measurements of students of different teachers are constantly changing among schools, or projects that are nested in cross-categorized migrants by country of origin and destination.

Repeat measurements of people living in a particular community and studying at a particular school

Hierarchical model with stochastic effects:

Dependent on the spatial community effect

Social network interaction

The new function of HLM 7 is also to estimate the hierarchical generalized linear model by maximizing Adaptive Gauss-Hermite (AGH) and higher order Laplace approximation. AGH is very suitable for cluster size (cluster size) small and large variance component of the situation. Higher-order Laplace methods require larger clusters but allow any large number of random effects (which are important when the number of clusters is large)