Embedded Multilevel Mixed Methods Design

The standard embedded design nests a secondary strand within a primary study without necessarily imposing a hierarchical data structure. The embedded multilevel variant adds the requirement that data are organized across nested organizational levels (e.g., individuals within groups within organizations), and the integration logic explicitly connects findings at different levels of that hierarchy. This makes the sampling and analytic demands considerably more complex.

No. Typically the dominant strand covers all levels relevant to the primary research question, while the secondary embedded strand is purposively targeted at the level where supplementary depth is needed. For example, the quantitative strand might span student, classroom, and school levels, while the qualitative strand focuses on the classroom level only.

Hierarchical linear modeling (HLM), also known as multilevel modeling (MLM), is the standard approach for quantitative data nested within groups. It correctly partitions variance across levels and produces unbiased standard errors. Software options include HLM 8, R (lme4, nlme), Stata (mixed), and SPSS (Mixed Models).

Use a parallel (or concurrent triangulation) multilevel design when both strands carry equal weight and are intended to provide independent, convergent answers to the same research question at the same level. Use the embedded variant when one strand is clearly primary and the other serves a supplementary, supportive role — typically to explain or elaborate a finding from the dominant strand.

State the integration point explicitly in the methods section: specify at which phase of the study and at which level of the hierarchy the strands are connected, and what the purpose of integration is (explanation, elaboration, or disconfirmation). In the results or discussion section, present the integrated interpretation with explicit reference to which level each claim derives from, and acknowledge any discrepancies between strands.