Skip to content

Crossed and nested factors in experimental designs: Are there any flowcharts for decision making?

4 messages · Ben Bolker, Mehdi Abedi, Jake Westfall

Original
Mehdi Abedi
# 
Dear all,

I am following your nice discussion in this group. Considering your high
level discussion in this group i was not sure to ask basic problems in the
mailing list!, therefore, i asked this question in researchgate.

I think several researchers have this kind of question which your guidance
can also help them using mixed models in their researches. I can imagine
that teaching advanced statistic with simple language is not easy. However
in addition to advanced statistic books and codes, introducing experimental
design and their link to mixed models using simple codes would be valuable
as well.

It would be my pleasure to have your suggestion here or in this link:

Warm regards

Mehdi:

https://www.researchgate.net/post/With_regards_to_crossed_and_nested_factors_in_experimental_designs_Are_there_any_flowcharts_for_decision_making

Ecological researches mainly have complicated statistical design. Mixed
models can test complicated designs with different crossed or nested
factors. For instance lme4 and nlme could be used in R. However, still
decision about statistical design is complicated for most researchers.

Mixed models are quit an advanced method for most researchers and recent
publications still use simple statistical designs. For instance most
publications in high ranked journals related to plant ecophysiology and
seed researches still apply only factorial design for the statistical
analysis and not mixed models including random and fixed factors.

Could you recommend some references or lectures introducing nested, crossed
design with simple examples and simple codes in R?. I mean some simple
explanation without details which start from simple design to complicated
design like http://conjugateprior.org/2013/01/formulae-in-r-anova/
<https://www.researchgate.net/go.Deref.html?url=http%3A%2F%2Fconjugateprior.org%2F2013%2F01%2Fformulae-in-r-anova%2F>

Graphical designs showing these crossed and nested would also be useful.
This information would be very helpful for both teachers and researchers
for application correct statistical analysis.
Ben Bolker
#
On 14-11-14 05:21 AM, Mehdi Abedi wrote:
It's pretty basic, but I like the section in Gotelli and Ellison's
_Primer of Ecological Statistics_.  They don't really get as far as
crossed random effects, but they do discuss nested, randomized block,
split-plot, and factorial (i.e. _crossed fixed effect_) designs.

  Ben Bolker
Mehdi Abedi
# 
Dear Ben,
Thanks for your kind introducing this reference. I also read
"Biostatistical Design and Analysis Using R
A Practical Guide" by Logan which was very useful for statistical design.

Linking classical ANOVAs with mixed model is still confusing for me because
i have to collect information from different books (design from classic
book and mixed model from new books). Due to importance of random effects
in mixed model, misunderstanding designs can fully affect on analysis. I
worry to wrongly connect the basic experimental designs with mixed models.
Therefore, i asked group to introduce references talking about both
experimental design and mixed model in the same book or text avoiding this
mistake.
Warm regards
Mehdi
On Fri, Nov 14, 2014 at 4:40 PM, Ben Bolker <bbolker at gmail.com> wrote:

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Jake Westfall
    

    

      
      #
    

  


  

      
Hi Mehdi,

I'm not sure if it's quite what you have in mind, but we have a recent paper on power / optimal experimental design for mixed models, which has some diagrams of different designs and discussion of when one might want to use one design vs. another. We focus very much on crossed random effects (specifically we talk about human subjects responding to random stimulus materials, but you can substitute in any other crossed random factors you like), but there is some discussion toward the end on nested and "partially crossed" designs. Could be useful perhaps.
http://jakewestfall.org/publications/crossed_power_JEPG.pdf

Also there is a very nice chapter by Raudenbush that takes about similarities and difference between classical mixed models (from ANOVA framework) and "modern" mixed models. It's been a while since I read it but I believe he focuses more on designs with nested random effects. I remember it being really useful and informative. Hopefully it will be for you too.
http://jakewestfall.org/misc/Raudenbush_1993.pdf

Jake