Can we test a multilevel moderation effect without significant τ of the unconditional model?
Locked
 
|
Hello list, Is there anyone
here who is expert in HLM? I have a basic question: I am
attempting to test a multilevel moderation model, in which the independent and
dependent variables are at level 1, while the moderator is at level 2.
According to what I learned from the textbook previously, I should run the unconditional
model firstly to examine whether the variance component (τ)
of the dependent variable’s slope is significant. Unfortunately, it was not significant.
It seems that it is not necessary for me to test further. However, I am curious
about the result so I did the multilevel moderation test and found it was
significant. Now here comes
my question: Maybe it is inconsistent
with the principle of HLM to test such a moderation test, but how to explain
the significant moderation effects since I got the statistic results? I would be grateful if any of you can give me some hopeful suggestions. Chu-Ding LING Ph.D. Student of Business Administration School of Management Zhejiang University |
Re: Can we test a multilevel moderation effect without significant τ of the unconditional model?
Administrator
|
I don't profess extraordinary expertise in HLM, but I'll jump in with a question and a comment.
Which book are you looking at, and what exactly does it say? Now the comment: Given that you appear to have an a priori interest in that particular cross-level interaction, my gut feeling is that you should go ahead and look at it regardless of significance (or not) of the variance of the random slopes. I see a parallel here to going ahead with planned orthogonal contrasts (in the ANOVA context) regardless of the significance of the omnibus F-test. There...maybe that will provoke the real HLM experts (Ryan, Alex...) to jump in with their comments. ;-)
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
|
|
If there is apriori theory that a cross-level interaction exists between a level-1 variable and level-2 variable, then it is perfectly reasonable to immediately fit a model which includes the interaction.
That said, since the OP couldn't resist fitting both models (I'm shocked! ;-)), we might as well discuss the probable reasons for the results. The scenario presented by the OP can likely be explained by: 1. A Type I error has been committed on the test of interaction (less likely) or 2. A Type II error has been committed on the test of slope in the model which excluded the interaction (more likely). Ryan p.s. I'm fairly certain this is discussed in one of the widely used and cited multilevel textbooks but I can't recall which one at the moment. ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
|
|
By the way, whenever posting a statistics question to SPSS-L, it is advisable to link it to SPSS in some way. Providing the MIXED syntax for both models would have been a great way to make this question related to SPSS. Moreover, including the syntax would help others learn how to fit such models in SPSS.
Ryan > On May 18, 2014, at 2:15 PM, GMAIL <[hidden email]> wrote: > > If there is apriori theory that a cross-level interaction exists between a level-1 variable and level-2 variable, then it is perfectly reasonable to immediately fit a model which includes the interaction. > > That said, since the OP couldn't resist fitting both models (I'm shocked! ;-)), we might as well discuss the probable reasons for the results. > > The scenario presented by the OP can likely be explained by: > > 1. A Type I error has been committed on the test of interaction (less likely) > > or > > 2. A Type II error has been committed on the test of slope in the model which excluded the interaction (more likely). > > Ryan > > p.s. I'm fairly certain this is discussed in one of the widely used and cited multilevel textbooks but I can't recall which one at the moment. ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
|
|
In reply to this post by 凌楚定
2014-05-28 9:25 GMT+08:00 凌楚定 <[hidden email]>:
|
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |