You are here

pseudoreplication in Model Fittness BY SEM Softwares(Smartpls,Lisrel,...)

2 posts / 0 new
Last post
aliz's picture
Offline
Joined: 12/27/2016 - 11:46
pseudoreplication in Model Fittness BY SEM Softwares(Smartpls,Lisrel,...)

Dear SEM Experts,
This is a problem that I have searched for it a long time and got no reasonable answer. In 1984 STUART H. HURLBERT in his article introduced the concept of pseudo-replication. He warned us about not to use multiple measures for inferential statistics. His field is ecology and in fields like that we survey the behavior of natural systems. Based on his idea if you measure the blood pressure of a man during time intervals it won’t lead to new samples. But the case is that in this kind of example we are talking about just one person which is to be judged and -usually- one assessor to measure.

But I don't know if it is also true if -for model fitness- I make use of n experts to judge and k cases for to be judged. In this example, consider that all n experts are independent from each other and all k cases are independent from each other too. So, can I claim that multiplying all the n experts judgments and k cases I will have n*k cases that can be entered in SPSS as a csv file its fitness be analyzed by SEM software?
1- If someone believes that judgments of a case by different independent expert are not independent from each other, please tell me the reason of dependency.
2- If someone believes that different judgments of an expert about different cases are not independent from each other, please tell me the reason of dependency.

If there is one of those 2 kinds of independence we could say that we must only have 1 expert to judge about different cases or one case to be judged by different experts to get the samples' independence. It seems that there is something wrong with this kind of judgment.

Thanks in Advance
P.S.
I must add that Hulbert only talks about the pseudo-replication in inferential statistics. But I don't know if we just run a model in SEM Software to cehck the validity or reliability, what will be the inferential statistics.

neale's picture
Offline
Joined: 07/31/2009 - 15:14
Neither is independent

I've not read Hulbert but I might now... Anyway, in my opinion:

  1. The repeated judgements by an observer may contain observer variance - a rating style or something that differs from other observers' style. So these may not be independent observations.
  2. Multiple judgments of an individual are not independent because they may contain reliable variance due to the case. Indeed, if such judgments were independent (no variance due to the case) it would suggest that there is no reliable variance associated with the case, which is usually a Bad Thing.

There are published reports of such analyses. Multilevel analysis is one way to go. A structural equation model may be another. In my own work I have looked at ratings of twins by both their parents. Having multiple raters rate multiple cases can be a valuable research design, see, e.g.,
Neale, M.C. & Stevenson, J. (1989) Rater Bias in the EASI Temperament Scales: A Twin Study. Journal of Personality and Social Psychology 56: 446-455.
and
Hewitt, J.K., Silberg, J.L., Neale, M.C., Eaves, L.J., Erickson, M. (1992) The analysis of parental ratings of children's behavior using LISREL. Behavior Genetics 22: 293-317.