Multilevel Marketing Analysis in Graphical System
Abstract : Marketing scholars are increasingly recognizing the importance of investigating phenomena at multiple levels. However, the analyses methods that are currently dominant within marketing may not be appropriate to dealing with multilevel or nested data structures. We identify the state of contemporary multilevel marketing research, fi nding that typical empirical approaches within marketing research may be less effective at explicitly taking account of multilevel data structures than those in other organizational disciplines. A Monte Carlo simulation, based on results from a previously published marketing study, demonstrates that different approaches to analysis of the same data can result in very different results (both in terms of power and effect size). The implication is that marketing scholars should be cautious when analyzing multilevel or other grouped data, and we provide a discussion and introduction to the use of hierarchical linear modeling for this purpose.
? We explored existing literature in top marketing journals, which analyzed research questions where the answers clearly depended on assumptions regarding the nonindependence of cases at the micro level.
? Existing methodological research does shed light on this issue (e.g., Bleise and Hanges 2004), and can be consulted by the interested reader.
? There may be huge potential in the reanalysis of existing data, to look for new answers to old questions by utilizing the power of multilevel tools.
? Multilevel modeling should be considered an addition to, not a replacement for, our existing methodological repertoire.
? It is important to address the potential nonindependence problems that may result from the use of multiple responses from groups or organizations.
? Problematic in this approach is the violation of the assumption of independent observation, which is a central assumption of most classical statistical procedures.
? A further problem of disaggregation lies in the fact that variables concerning the higher level are analyzed on the basis of the larger lower-level sample size.
? Although multilevel modeling is likely to be of signifi cant use to the marketing researcher, we would issue a word of caution about overenthusiastic application of multilevel techniques to new and existing marketing problems.
• Our purposes, we focus on the Chinese data set. Due to the inherent difficulty in collecting data in China, AtuaheneGima and Li were forced to use multiple respondents from organizations.
• The purposes of our analysis, we selected four individual hypotheses that together exemplify the main types of multilevel hypotheses likely to be encountered by marketing researchers.
• The data are intrinsically structured on at least two levels—employees are naturally grouped into firms.
• However, there are a number of group-level factors likely to affect the variance of the constructs (e.g., manager or fi rm factors), and thus again it is not clear that cases from the same fi rm should be treated as independent.
? The use of multilevel analysis approaches in research can provide practitioners with a more accurate picture of what level of the organization they should directly target for any performance improvement efforts.
? In the studies reviewed above, only a small minority of studies tested the assumption of a suffi cient correspondence among the ratings within the micro level, by, for example, using intraclass correlation coeffi cients (ICCs) or within- and betweengroup analysis (WABA), before aggregating the data.
? The higher the intraclass correlation coeffi cient within the lower level group, the less appropriate it is to treat individuals within a group as independent data points.
|