Home › Forums › Default Forum › Hierarchical linear modelling (using SPSS)
- This topic has 3 replies, 3 voices, and was last updated 10 years, 6 months ago by
dani hugentobler.
-
AuthorPosts
-
17th August 2010 at 12:41 pm #4211
dani hugentobler
ParticipantHello!
I try to use hierarchical linear modelling, since I have a mixed within-between-subjects design and 5 time-varying covariates I would like to hold constant.
The syntax-file I uses looks as follows:
MIXED dv BY between within1 within2 WITH cov_1 cov_2 cov_3 cov_4 cov_5
/CRITERIA=CIN(95) MXITER(100) MXSTEP(5) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0,
ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE)
/FIXED=between within1 within2 between*within1 between*within2 within1*within2 between*within1*within2 | SSTYPE(3)
/METHOD=REML
/RANDOM=INTERCEPT cov_1 cov_2 cov_3 cov_4 cov_5 | SUBJECT(sub) COVTYPE(VC).…and produces the following error-message:
Die endgültige Hesse-Matrix ist nicht positiv definit, obwohl sämtliche Konvergenzkriterien erfüllt sind.
I have two questions:
Is this the correct approach to test the impact of the 3 factors on the dependent variable holding 5 covariates constant?
What do I need to change in order not to get this error-message?I would be very happy, if anybody could help!
dani.h.10th September 2010 at 8:49 am #4214dani hugentobler
Participanthi dave,
thanks a lot for your message! i created the syntax-file myself.
– the outcome variable is dv
– the 3 factors are between, within1, and within2
– the covariates are cov_1 cov_2 cov_3 cov_4 cov_5
the goal of the analysis is to get the variance components in the dependent variable explained by the 3 factors, whereas cov_1 – cov_5 are treated as constant variables.
i would be very happy, if you could help!
thanks in advance, dani.h.13th October 2010 at 1:31 am #4213Jeremy Miles
ParticipantI’m guessing that that means that your hessian matrix is not positive definite, or can’t be inverted (same thing). That often means that your model is too complex, or that you don’t have a large enough sample.
You have a massive number of random factors, that is probably causing the problem. I think you need to start off with intercept on the random line. The covariates should be on the fixe factor line (I think – I’m not sure, ‘cos I don’t use SPSS for mixed models).
Don’t try to use the menus – they are a nightmare. Syntax is much more straightforward.
Don’t try the interactions yet.
I would start off with a simple model and build up by adding more parameters. Also, the criteria are all fine and can be left at their defaults. Start, for example, with:
MIXED dv
/METHOD=REML
/RANDOM=INTERCEPT | SUBJECT(sub) .Jeremy
14th October 2010 at 9:57 am #4212K.Kalyanaraman
MemberDear Friend
I feel that your syntax is okay. The problem that you are facing may be due to a very high correlation between any two of your variables used to regress on dv. Please check. -
AuthorPosts
- The forum ‘Default Forum’ is closed to new topics and replies.