Repeated Measures

[mpickard]

Hi -

Can SmartPLS handle repeated measures? My study is a 2x2x2 within subjects design. My data looks like this (where the first column is the subject ID):

1010 0 4.26 2.4 1 0.1 171 1 1 1 0
1010 0 4.26 2.4 1 0.1 170 1 1 0 0
1010 0 4.26 2.4 1 0.1 151 1 0 1 0
1010 0 4.26 2.4 1 0.1 187 1 0 0 0
1010 0 4.26 2.4 1 0.1 181 0 1 1 0
1010 0 4.26 2.4 1 0.1 171 0 1 0 0
1010 0 4.26 2.4 1 0.1 162 0 0 1 0
1010 0 4.26 2.4 1 0.1 141 0 0 0 0
2012 1 5.5 2.84 3 -0.7 178 1 1 1 1
2012 1 5.5 2.84 3 -0.7 134 1 1 0 1
2012 1 5.5 2.84 3 -0.7 151 1 0 1 1
2012 1 5.5 2.84 3 -0.7 -29 1 0 0 1
2012 1 5.5 2.84 3 -0.7 173 0 1 1 1
2012 1 5.5 2.84 3 -0.7 150 0 1 0 1
2012 1 5.5 2.84 3 -0.7 140 0 0 1 1
2012 1 5.5 2.84 3 -0.7 1 0 0 0 1
2014 0 5.65 2.32 10 -0.4 92 1 1 1 1
2014 0 5.65 2.32 10 -0.4 0 1 1 0 1
2014 0 5.65 2.32 10 -0.4 63 1 0 1

Any help on how to handle repeated measures would be greatly appreciated.

Thanks,

Matt

[gdavidgarson]

Some authors state that observations need not be independent in PLS (see Lohmoller, 1989: 31; Chin & Neusted, 1999; Urbach & Ahlemann, 2010), apparently based on the statement by Herman Wold (1980: p. 70) that, "Being distribution-free, PLS estimation imposes no restrictions on the format or on the data." However, being distribution-free does not mean data independence can be ignored or that use of repeated measures data is not problematic. Ignoring the assumption of independence is a form of measurement error and since PLS is relatively robust in the face of measurement error, in this sense only is it true that PLS does not require independent observations. More to the point, PLS is a single-level form of analysis, not a form of multilevel analysis. If there is some grouping variable that might be handled via multilevel analysis (ex., time, for repeated measures), and if it has a limited number of levels, it may be possible to handle it within PLS by creating separate factors for each level (ex., time1, time2, time3). Note, though, that PLS is less powerful than covariance-based structural equation modeling for such repeated measures models since SEM can model correlated residual error and PLS cannot. Multilevel and repeated measures PLS is, as far as I know, academic discussion at this point and not implemented by leading PLS software. Feedback on this welcome.