Dyadic Regression in the Presence of Heteroscedasticity: An Assessment of Alternative Approaches
2010 (English)In: Social Networks, ISSN 0378-8733, Vol. 32, no 4, 279-289 p.Article in journal (Refereed) Published
Although the problem of heteroscedasticity has been the subject of much discussion in other areas of applied statistics the problem has received scant attention in the social network literature. This study attempts to remedy this situation by considering how traditional methods for significance testing in dyadic regression models, such as standard QAP tests, perform under conditions of heteroscedasticity. Moreover, the article presents two alternative methods to deal with heteroscedasticity that are both shown to perform rather well with typical social network data under conditions of both heteroscedasticity and homoscedasticity. Overall, the results of the study suggest that applied researchers using regression techniques to study dyadic data are well advised to correct for heteroscedasticity, by either of the two methods discussed here, whenever there is a reason to suspect heteroscedasticity.
Place, publisher, year, edition, pages
Elsevier , 2010. Vol. 32, no 4, 279-289 p.
Dyadic regression; Quadratic assignment; Autocorrelation; Heteroscedasticity; Two-way clustered standard errors
Political Science (excluding Public Administration Studies and Globalization Studies)
IdentifiersURN: urn:nbn:se:uu:diva-131053DOI: 10.1016/j.socnet.2010.04.002ISI: 000283402000003OAI: oai:DiVA.org:uu-131053DiVA: diva2:352663