Discussion: View Thread

  • 1.  Double-dipping

    Posted 04-07-2013 17:52

    The LMX scale measures the strength of the Unique Strategic Alliance (USA) between direct, hierarchical reports as seen from the position of less position power.  It has been found to vary in strength within business units and reflect relative advantage.  As in most competitions, the stronger performances receive the more attractive outcomes.  Thus, those earning strong USA's are winners relative to those who do not.

    Unfortunately, many studies have been published which use the deviation score of LMX from the business unit average as a predictor along with LMX.  These studies contain an analytical problem.  Because the LMX is negatively skewed, the mean and the deviation score (score minus the mean) are highly correlated.  This indicates that these measures share too much common covariance for linear regression and their independent contribution to any criterion is confounded.  Only one can be interpreted in a regression.  Clearly, when the mean is nearly a constant and most all of the variation is due to LMX scores, the raw and deviation scores must be highly correlated.  They share the same sources of variance.  This means that both scores should not be used as different variables.  Any statistical interaction of LMX and group mean deviation LMX is an artifact, but the main effect of one is not.  The original discovery of LMX research was that business unit averages disguised the power of the dyadic alliances, and it was not average leadership style but dyadic alliances that predicted leadership outcomes.

    George Graen

    jag



  • 2.  Double-dipping

    Posted 04-08-2013 17:22
    One good point George makes, amongst several, is editors and reviewers need to require reporting and discussion of the implications of skewness and kurtosis statistics. These are important indicators of what's going on in the data, and maybe even what's going on with the latent variables.
    Regards,
    Romie Littrell
     
    Romie Frederick Littrell, PhD, FIAIR
    Editor, AIB Insights; International Management Area Editor, International Journal of Emerging Markets
    AUT Business School, Tel. extension 5805, Mail stop B-31

    From: "Lmxlotus@AOL.COM" <Lmxlotus@AOL.COM>
    To: MG-ED-DV@AOMLISTS.PACE.EDU
    Sent: Monday, 8 April 2013, 9:52
    Subject: [MG-ED-DV] Double-dipping

    The LMX scale measures the strength of the Unique Strategic Alliance (USA) between direct, hierarchical reports as seen from the position of less position power.  It has been found to vary in strength within business units and reflect relative advantage.  As in most competitions, the stronger performances receive the more attractive outcomes.  Thus, those earning strong USA's are winners relative to those who do not.
    Unfortunately, many studies have been published which use the deviation score of LMX from the business unit average as a predictor along with LMX.  These studies contain an analytical problem.  Because the LMX is negatively skewed, the mean and the deviation score (score minus the mean) are highly correlated.  This indicates that these measures share too much common covariance for linear regression and their independent contribution to any criterion is confounded.  Only one can be interpreted in a regression.  Clearly, when the mean is nearly a constant and most all of the variation is due to LMX scores, the raw and deviation scores must be highly correlated.  They share the same sources of variance.  This means that both scores should not be used as different variables.  Any statistical interaction of LMX and group mean deviation LMX is an artifact, but the main effect of one is not.  The original discovery of LMX research was that business unit averages disguised the power of the dyadic alliances, and it was not average leadership style but dyadic alliances that predicted leadership outcomes.
    George Graen
    jag