Introduction: Methodological issues pertaining to multilevel research are certainly complex. However, this complexity is fruitful  in many domains, such as collective and shared constructs. Methods of data analysis obtained from random tests when the unit of analysis corresponds to intervention unit are well recognized. When individual and random data are used via a group intervention, different statistical methods should be applied to evaluate treatment effects. The three introduced models are: I: individuals randomly assigned to individually administered treatments, II: preexisting groups are randomly assigned to group administered treatments and III: randomization is administered on an individual basis, while the treatment is administered on a group setting. This research shows that most research in social welfare corresponds to the second or third model while data were analyzed with the well – recognized first model. An example of managing workplace bullying by training employees of a holding company in citizenship behavior through different courses and using different sites illustrates the third model. Variance component of the exemplified research was computed for three models at level 1( ) and level 2( ). It was shown that the best ratio of these variance component appears in the third model.

Method: All (quasi) experimental articles in Social Welfare Quarterly (SWQ) in which group treatment was applied were investigated. .Also, the content analysis method was used to investigate connotations of group therapy in introduction, discussion and conclusion of articles. The mentioned studies  illustrate a distinction  between the units that are randomized at the cluster level while the intervention is delivered at the individual or group (cluster) level. Six articles  were chosen. The unit of randomization and the unit to which treatment is administered for these articles were determined and  correspondence between them was analyzed .  Only in one article the treatment was administered in clusters, but in three articles there is no correspondence between the unit of analysis in randomization and treatment. Results of content analysis indicate that only half of the articles  in social welfare journal referred to group treatment, although none of the groups are reflected in the statistical analysis. .

Findings: In all the  six articles of  SWQ, deficiency in statistical method was observed, because in all of them ANOVA methods were applied instead of hierarchical method regardless of  group therapy . Also, except for one article  there is was no  attention paid to group setting effects in the introduction and discussion. The third data analysis model received little methodological attention that raises  issues. The most important issue is the relationship between treatment and residual variance. Traditional regression models and multilevel models were compared. The advantage of multilevel models especially their explicit determination of  source of variability at both individual and group level was mentioned.

Discussion: The results  showed that method of hierarchical linear models were not applied in articles published by  SWQ. It is recommended to consider two articles which were written a decade ago byLee & Thompson (2005) and Roberts & Roberts (2005). The advantage of paying attention to the unit of randomization and treatment and between design of research and statistical method of analyzing data empowers researchers to control for contamination across individuals if the cluster was used in randomization or treatment. In cluster randomized research,  such as research of Ali Pouri Niaz et al (1388) which has been conducted  at different sites and communities, the third model in this article  was superior to traditional pre-post tests. The methods introduced in this article  for analyzing data can be extended to some of the more complex situations encountered in social welfare research.