ISSN 1671-3710
CN 11-4766/R
主办:中国科学院心理研究所
出版:科学出版社

Advances in Psychological Science ›› 2020, Vol. 28 ›› Issue (8): 1392-1408.doi: 10.3724/SP.J.1042.2020.01392

• Research Method • Previous Articles    

Data aggregation adequacy testing in multilevel research: A critical literature review and preliminary solutions to key issues

ZHU Haiteng()   

  1. Department of Military and Ideological Basic Education, PLA Army Academy of Artillery and Air Defense, Hefei 230031, China
  • Received:2019-12-06 Online:2020-08-15 Published:2020-06-28
  • Contact: Haiteng ZHU E-mail:prettypig1990@sina.com

Abstract:

The measurement of shared unit property constructs is ubiquitous in multilevel organizational research, of which the most frequently used approach is to aggregate the ratings of several unit members to the unit level. The data aggregation adequacy testing (DAAT) is a statistical hurdle to ensure the validity and representativeness of aggregated scores. Well-established indicators of DAAT include within-group agreement index, rWG, and within-group reliability indices, ICC(1) and ICC(2); nonetheless, some key issues are still open to debate, for instance, the superiority of the two families of indicators, the null distribution and data screening decision of rWG, and appropriate cut-off values. To address the above questions, the current research firstly conducted a content analysis of 166 studies adopting DAAT procedure published on 9 Chinese journals in the field of management and psychology since 2014, coupled with 85 studies from Journal of Applied Psychology as a comparison. Common problems in routine practice of DAAT were identified and related suggestions were proposed as follows: (1) Disentangling and differentiating the role of DAAT indicators; specifically, rWG should be used as the exclusive indicator of aggregation adequacy, whereas ICC(1) and ICC(2) should be deemed as indices of validity and reliability, respectively. (2) Making prudent and justifiable decisions in choosing null distributions when calculating rWG index, and excluding groups with low within-group agreement. (3) Applying more reasonable and moderately flexible cut-off values instead of arbitrary and rough practical standards. Last but not the least, researchers should always prioritize theoretical considerations in the process of framework building and DAAT, and unload disproportionate dependence on statistical results.

Key words: multilevel research, shared unit property, aggregation, within-group agreement, within-group reliability

CLC Number: