4

I am currently in the process of analyzing a data set comprised of manager-subordinate dyads. Data were collected cross-sectionally and the data set contains some of the same variables collected from both members of the dyad (e.g., age, relationship tenure, interpersonal trust) as well as some unique variables for each member.

Unfortunately, I have a large number of participants whose partners did not fill out their surveys. I was wondering if these unpaired cases could be of any utility in my analysis? For example, when conducting an EFA or CFA on a scale, could I include all managers (including the unpaired individuals), or is this bad practice since the unpaired individuals will not be a part of my regression and/or path analyses?

I have been reading Dyadic Data analysis by David A. Kenny et al., but I have been unable to find an answer to this question so far.

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Nicholas Bremner
  • 43
  • 3

1 Answers1

3

Maybe my answer comes a bit late ... When you do Actor-Partner-Interdependence Models (APIMs) with multilevel modeling, single cases still improve the estimation of the intercepts. Hence, these data are not necessarily lost.

Felix S
  • 4,432
  • 4
  • 26
  • 34
  • 1
    You are right (+1). However, you (probably) know that using MLM for analyzing dyadic data is based on the assumption of indistinguishable dyads (sure, one can include an interaction term but the model gets "more complicated"). – Bernd Weiss May 16 '12 at 15:43
  • Yes, if you do the "double intercept approach", you can also model distinguishable dyads. – Felix S May 17 '12 at 07:14