I have a problem that is confounding me. I conducted two surveys that collected data about individual perceptions (dependent variable) and about project team level actions (independent variable) in a straightforward two-level design. I also had some individual level and team level control variable.
The survey for the team level data was designed in such a way that the mean score for all the survey questions for each team was a measurement of the overall construct, but there were subsections of the survey that measured each of four subsets within the construct for each team. In effect, I had five independent variables – one for the overall mean and one for each subset for each team.
This caused me to analyze the data in what may be considered a non-standard way. I used HLM 7.0. Using the overall score I found a significant relationship. This result was, in my mind, the equivalent of the final result of a step-up approach as if I was adding all of the four subset IVs to the equation in combination. To delve deeper to understand the relationship between each of the four subset independent variables and the dependent variable, I examined each of the four separately with the dependent variable (not in combination). I found significance for each one.
But if I build it in reverse, that is, if I sequentially add each of the four to the equation in combination, I get significance with the first independent variable, then each time I add one there is no significance for any of them. This doesn't seem to make sense since the overall score was significant.
Thanks for any help.