Research in social sciences is a human endeavor and hence is subject to human error. In management and organizational research one form of these errors are unsubstantiated beliefs about statistical methods that lead to questionable analysis practices. In this dissertation, I address several of these statistical and methodological myths and urban legends related to partial least squares (PLS) estimation and formative measurement in the context of structural equation modeling. This dissertation consists of five papers in which I analyze the use of these methods in management and organizational research and show that many of the beliefs that the current literature conveys do not hold when subjected to either formal analysis or Monte Carlo simulation. With respect to formative measurement, the dissertation challenges the long-held belief that all formative indicators are required for valid construct measurement. More generally, I argue that formative measurement is not an equally attractive alternative to more traditional reflective measurement and that PLS estimation is not a viable alternative for estimating these models. With respect to PLS estimation, I show that the apparent advantage of the method is a fallacy created by ignoring the effect of chance correlations. Furthermore, I show that the way in which PLS capitalizes on chance results in non-normal distributions of parameter estimates, which precludes null hypothesis significance testing.
|Translated title of the contribution||Tilastollisia ja menetelmällisiä myyttejä ja urbaaneja legendoja johtamisessa: Esseitä osittaisneliösummaestimattorista ja määrittävästä mittaamisestä|
|Publication status||Published - 2014|
|MoE publication type||G5 Doctoral dissertation (article)|
- statistical and methodological myths and urban legends
- partial least squares
- formative measurement
- structural equation modelling