In my research I hypothised that team learning mediates between transformational leadership and innovative work behavior. To analyse the relations I use HLM 7.
I found a significant positive relation between transformational leadership (X) and innovative work behavior (Y). I also found a sign. positive relation between transformational leadership (X) and team learning (M) and the relation between team learning and innovative work behavior (Y) is also sign. positive.
So X to Y, X to M and M to Y are all sign. positive (all p = < 0.001)
But when I put transformational leadership (X) and Team learning (M) both in the model: transformformational leadership (X) is sign. but the sign. level of team learning (M) drops to p=0.30.
How to interpret this?
To summarise your question, you are investigating three pathways: i. translational leadership (X) to team learning (M) - path a, ii. team learning (M) to innovative work behaviour (Y) - path b, and iii. translational leadership (X) to innovative work behaviour (Y) - path c. When you analyse the pathways individually, they are significant, but when you add X and M as independent variables for Y; path b (from M to Y) is no longer significant.
It appears you are investigating M as a mediator for the relationship between X and Y. You are correct to investigate the individual relationships first; however, when you include variables X and M in the model, your focus should be on the regression coefficient of the path between X and Y (let's call it c'). If path c' is statistically significant, but less than path c (the coefficient from the individual regression of Y on X), then M partially mediates the relationship between X and Y. If path c' is no longer statistically significant, then M fully mediates the relationship between X and Y. Your question suggests that M partially mediates this X-Y relationship because path c' is still statistically significant (and I assume it reduced in magnitude). You should also carry out Sobel's test to determine whether the mediation effect of M is statistically significant.
I hope I have helped you.
Testing Mediation with Regression Analysis - PDX
PDFweb.pdx.edu › ~newsomj › ho_mediation
Wilfred, you mentioned about using HLM 7, but did not give any information about the level of your variables or the multilevel structure of your model. To simplify the matter, I am gonna assume that you have a single level model. In this case, one way to interpret these results is that team learning might not be mediating the relationship between transformational leadership and innovative work behavior as you hypothesized. Instead, transformational leadership (TL) might be the common cause of the (i) innovative work behavior (IWB), and (ii) team learning (L) variables. Maybe that is why the relationship between IWB and L becomes non-significant when the effects of TL is controlled.
But, there are three caveats that I must mention:
(i) It should also theoretically make sense in order to interpret these results as above.
(ii) Relying solely on p values is not a good practice, especially in statistical mediation analysis.
(iii) Non-significant direct effect(s) does not necessarily mean that the indirect effect between the IV and DV will also be non-significant (Please see https://www.researchgate.net/post/Can_you_have_mediation_if_either_the_a-or_b-path_are_not_significant ). You might need to consider using more contemporary methods of mediation analysis.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics