Marker technique for testing common method bias in SmartPLS
Posted: Wed Jul 17, 2019 2:29 pm
I've looked at other posts on this forum about CMB, but was hoping for a clearer answer about how to use a theoretically unrelated marker factor to detect common method bias in SmartPLS. Here is some background: I have been reading about the marker technique to detect common method bias (e.g. Johnson, Rosen, & Djurdjevic, 2011; Podsakoff MacKenzie, Lee, & Podsakoff, 2003; Richardson, Simmering, & Sturman 2009; Williams Hartman, & Cavazotte, 2010) and was trying to figure out how to do this in SmartPLS. From my searches, it seems as though there are different ways to use the marker technique that vary by statistical program, with some being easily run by the program (e.g. Amos), but I wasn't sure how to do this in SmartPLS. I have a fairly complex model with many antecedents, 1 mediator, 2 dependent variables(most with multiple indicators) along with some control factors, some of which are single-item, binary factors (e.g. gender, age).
Specifically, I am wondering whether one links the theoretically unrelated marker factor to other factors in the model (i.e. paths between factors), or if it is necessary to link the marker factor to its own indicators as well as all indicators of all substantive factors in the model (i.e. generally follow the approach suggested for a common latent method factor per Liang, Saraf, Hu & Xue's 2007 MIS Quarterly article, but also including the marker variable's indicators for the marker factor)? Some seem to suggest that both the marker and latent method factor should be included in the same model (e.g. Eichhorn paper for testing for CMB in SAS https://www.lexjansen.com/mwsug/2014/AA ... 4-AA11.pdf), but I ran out of space running this model in SmartPLS (even after implementing the solution for "the java heap space error").
Also, I was not sure what results from SmartPLS need to be used to determine if there were CMB issues in my data using the theoretically unrelated marker variable technique. There seem to be different recommendations. Would one look at the significance levels between the marker factor to others (either indicators or factors) and possible changes to significance in hypothesized factor relationships only? And/or are factor loadings squared to see if this was an issue per Liang et al's (2007) suggestion for the common latent method factor way to check for CMB in PLS Graph? And/or are correlations between factors squared per https://www.youtube.com/watch?v=pUKT-QvQYhM (see minute 3:53)? And/or is something else generally done to assess common method bias using the marker technique? Some articles suggest constraining values, but to my knowledge we can't do this in SmartPLS.
Any specific advice, tips, or videos related to checking for CMB with the marker technique in SmartPLS would be greatly appreciated. Many thanks to the members of the forum in advance for your help.
Specifically, I am wondering whether one links the theoretically unrelated marker factor to other factors in the model (i.e. paths between factors), or if it is necessary to link the marker factor to its own indicators as well as all indicators of all substantive factors in the model (i.e. generally follow the approach suggested for a common latent method factor per Liang, Saraf, Hu & Xue's 2007 MIS Quarterly article, but also including the marker variable's indicators for the marker factor)? Some seem to suggest that both the marker and latent method factor should be included in the same model (e.g. Eichhorn paper for testing for CMB in SAS https://www.lexjansen.com/mwsug/2014/AA ... 4-AA11.pdf), but I ran out of space running this model in SmartPLS (even after implementing the solution for "the java heap space error").
Also, I was not sure what results from SmartPLS need to be used to determine if there were CMB issues in my data using the theoretically unrelated marker variable technique. There seem to be different recommendations. Would one look at the significance levels between the marker factor to others (either indicators or factors) and possible changes to significance in hypothesized factor relationships only? And/or are factor loadings squared to see if this was an issue per Liang et al's (2007) suggestion for the common latent method factor way to check for CMB in PLS Graph? And/or are correlations between factors squared per https://www.youtube.com/watch?v=pUKT-QvQYhM (see minute 3:53)? And/or is something else generally done to assess common method bias using the marker technique? Some articles suggest constraining values, but to my knowledge we can't do this in SmartPLS.
Any specific advice, tips, or videos related to checking for CMB with the marker technique in SmartPLS would be greatly appreciated. Many thanks to the members of the forum in advance for your help.