www.cmm.bristol.ac.uk/forum

Is there anyone could give some suggestions?

Thanks.

Diva

I asked Kelvyn Jones about this and he said the following:

The key is that not that the outcome specific slopes are significantly different from zero but significantly different from each other; you can set this up in the Intervals and test window by putting a 1 for one of the slopes and a -1 for the one to be compared with so you are testing a difference.

Here is an example using Beck’s Schizophrenia example (available from http://hedeker.people.uic.edu/SCHIZX1.DAT.txt, https://rdrr.io/github/lme4/lme4/src/mi ... s/schizo.R):

Model: Wald Test: (No significant difference)

He also suggested reading the at materials at http://www.bristol.ac.uk/media-library/ ... models.pdf, noting that beta3 on slide 14 should have the value 0.481 instead of -0.481.

Another possibility is using the MLwiN customised predictions on the logit link scale (effectively the left hand side of slide 18). The proportional odds assumption would be parallel lines on the logit scale.

Hi Chris,

Many thanks for your help. And I will read the documents you suggested carefully.

I have another problem when fitting a model using R2MLwiN when I explored a random effect of one individual variable (random slope random intercept model). And the error is shown in the bottom. I have read several related topic in the forum, and tried to solve it. However, it didn't work to my problem. There is no missing data in my dataset; I have reordered my data; the cons is numeric; I have assign initial values from the basic random intercept model. So could you please give me some suggestions? I have sent my sample data and code to your email box (c.charlton@bristol.ac.uk). Hope you could help me solve this problem. Many thanks.

Regards.

Diva


/nogui option ignored
ECHO 0

Echoing is ON
BATC 1

Batch mode is ON
MAXI 2
STAR
iteration 0
iteration 1

Convergence not achieved
JOIN -1.42275739304974 0.308439026824743 1.25395389725885 2.84483612421086 -0.251278520475903 '_FP_b'
JOIN 0.039749333555033 0.0346204857887299 0.0383359924738334 0.0332605244565208 0.0364639882753521 0.040004125270958 0.0318606652111456 0.0348067623359593 0.0377525848112598 0.0484698156008947 -0.0114819365563485 -0.0118878355577854 -0.0121166989349699 -0.0125251094721577 0.0181430152561681 '_FP_v'
JOIN 0 0 2.75701906789359 1 '_RP_b'
JOIN 0 0 0 0 0 0.181569740306571 0 0 0 0 '_RP_v'
TOLE 2
MAXI 20
NEXT

error while obeying batch file C:/Users/Diva/AppData/Local/Temp/RtmporJ1lh/macrofile_1ca015fe7001.txt at line number 88:
NEXT

design vector at level 2 is the wrong length

WARNING: macro file has aborted, your data may be in a transformed state

WARNING: macro file has aborted, your data may be in a transformed state

error while obeying batch file C:/Users/Diva/AppData/Local/Temp/RtmporJ1lh/macrofile_1ca015fe7001.txt at line number 88:
NEXT

design vector at level 2 is the wrong length
.
Execution completed

Error in read.dta(IGLSfile) :
unable to open file: 'No such file or directory'

Although the default for the MISR command (replace small numbers with missing values in residuals) while running discrete response models was changed in MLwiN 3.01 it appears not to have been applied to multinomial models. The attached test version of R2MLwiN changes this option globally so it will be applied in this case, and this prevents the error from occurring in your example.

In general we recommend fitting discrete response models with MCMC as the IGLS estimation uses quarsi-likelihood and can therefore have biases.

Hi Paul,

It works for me now. Thanks a lot! And when I install this new version of R2MLwiN in R3.4.0, it doesn't work. However, after updating R version into R3.4.2, it works now. Just want to mention this problem here for others' reference. Again many thanks for your help.

Regards.

Diva