I have one question which seems to be often ignored in standard textbooks in statistics. In my design there is (by definition) one cell left: I want to compare gender composition of siblings (same- or cross-gender) and zygosity (identical twins, fraternal twins, and non-twin siblings) of my participants.
As cross-gender identical twins do not exist I have an incomplete 2x3-design and in fact five groups:
identical twins / same-gender
fraternal twins / same-gender
fraternal twins / cross-gender
siblings / same-gender
siblings / cross-gender
The SPSS help (GLM command) recommends calculation of Type IV sums of squares, which was "specially designed for situations involving empty cells ...", but I guess that's not sufficient in this case. The DV is a repeated measurement (2 items), but I think this is less important here.
Any suggestions or remarks I should consider?
One approach would be to do create a single factor with the five levels you describe and then use an LMATRIX command (in syntax) to compare the specific contrasts you are interested in. The difficulty will be in interpretation, because you only ever observe identical twins with same-gender.
Suppose your factor is called GROUP and its levels are 1-5 in the order you listed in your question, than an LMATRIX command to compare dizigosity and gender compostion would be something like this:
\LMATRIX = "dizigosity comparison"
GROUP 1 -.5 -.5 0 0 ; GROUP 0 .5 .5 -.5 -.5
\LMATRIX = "gender composition comparisons"
GROUP 1/3 1/3 -1/2 1/3 -1/2
Again, the main difficulty is in interpretation because of the issue with identical twins.
NOTE: I didn't test the LMATRIX syntax above.
One approach would be to do create a single factor with the five levels you describe and then use an LMATRIX command (in syntax) to compare the specific contrasts you are interested in. The difficulty will be in interpretation, because you only ever observe identical twins with same-gender.
Suppose your factor is called GROUP and its levels are 1-5 in the order you listed in your question, than an LMATRIX command to compare dizigosity and gender compostion would be something like this:
\LMATRIX = "dizigosity comparison"
GROUP 1 -.5 -.5 0 0 ; GROUP 0 .5 .5 -.5 -.5
\LMATRIX = "gender composition comparisons"
GROUP 1/3 1/3 -1/2 1/3 -1/2
Again, the main difficulty is in interpretation because of the issue with identical twins.
NOTE: I didn't test the LMATRIX syntax above.
Can't you just make the IVs dummy variables of 0 or 1 and use those as multiple regression predictors against your dependent variable(s)? For identical twins, Identical = 1; for non-identical, Identical = 0. Etc. Use GLM and R^2 as final result instead of significance testing.
Not so elegant to partecipate into a discussion by excluding some other partecipants. (i.e. not using English as a common shared language).
Dank!
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS