hcfa questions
4 views
Skip to first unread message
mhil...@gmail.com
Oct 2, 2008, 6:10:28 AM10/2/08
to CorpLing with R
Hi,
I have two basic questions regarding the use of hierarchical
configural frequency analysis.
First, is there a rule of thumb of how many factors are acceptable,
given a certain number of datapoints? To illustrate, I have around 350
examples and would like to analyze 5 factors (2-4 levels each). I
worry that there are too many fields with very small expected
frequencies.
Second, is it ok to use factors that are not independent? To
illustrate, one of my factors categorizes a constituent into NP, PP,
and ADJP. Another one distinguishes between grammatical functions,
i.e. SUB, OBJ, and OTHER, of that same constituent. Of course, PP and
ADJP can only be OTHER. I could just recode the two variables into one
(NPsub, NPobj, PP, ADJP), but does this make a difference?
Many thanks! --Martin
I have two basic questions regarding the use of hierarchical
configural frequency analysis.
First, is there a rule of thumb of how many factors are acceptable,
given a certain number of datapoints? To illustrate, I have around 350
examples and would like to analyze 5 factors (2-4 levels each). I
worry that there are too many fields with very small expected
frequencies.
Second, is it ok to use factors that are not independent? To
illustrate, one of my factors categorizes a constituent into NP, PP,
and ADJP. Another one distinguishes between grammatical functions,
i.e. SUB, OBJ, and OTHER, of that same constituent. Of course, PP and
ADJP can only be OTHER. I could just recode the two variables into one
(NPsub, NPobj, PP, ADJP), but does this make a difference?
Many thanks! --Martin
daniel wiechmann
Oct 2, 2008, 1:44:26 PM10/2/08
to corplin...@googlegroups.com
Hi Martin,
I read somewhere (either in von Eye 1990, or Krauth & Lienert 1973) that as a rule of thumb your sample size should be roughly N = 5*2^d (with d = number of dimensions ).
hth,
--daniel
____________________________________
daniel wiechmann
department of british and american studies
friedrich-schiller-university, jena
www.daniel-wiechmann.eu
I read somewhere (either in von Eye 1990, or Krauth & Lienert 1973) that as a rule of thumb your sample size should be roughly N = 5*2^d (with d = number of dimensions ).
hth,
--daniel
____________________________________
daniel wiechmann
department of british and american studies
friedrich-schiller-university, jena
www.daniel-wiechmann.eu
daniel wiechmann
Oct 2, 2008, 2:30:45 PM10/2/08
to corplin...@googlegroups.com
--daniel
____________________________________
daniel wiechmann
department of british and american studies
friedrich-schiller-university, jena
www.daniel-wiechmann.eu
Reply all
Reply to author
Forward
0 new messages