[R-sig-ME] PupilAbility

[David Duffy]

On Fri, 14 Sep 2007, Iasonas Lamprianou wrote:

> Dear friends, this is a question that has to do with both stats and 
> lmer. I hope somebody could respond. I gave a probabilities test to 
> around 300 Year 6 and Year 7 pupils in England. I also gave the same 
> test to their teachers (14 teachers). The test for the teachers consists 
> of two parts, only 8 of the teachers completed the second part, so only 
> 8 teachedrs and 230 pupils are in the analysis. I also gave the test to 
> other 100 (approximately teachers), so I know where those 14 teachers 
> stand compared to the other 100 teachers that completed the test.  I am 
> using the ability of the teachers on the two parts of the test as 
> predictors of the ability of the pupils. This is the model:
>
> ab2 <- lmer(PupilAbility ~ 1+TeacherAbilityPart_A * TeacherAbilityPArt_B +(1|TEACHER),mix,method="ML")
>

I would think you need to fit a structural equation model that includes a
measurement model for TeacherAbility.  Then you can also use the teachers who
only completed part A but not part B.  Unfortunately, the R sem package
doesn't handle multigroup or missing data type problems, as far as I know.


    TeacherAbility ------> PupilAbility
         /  \
        /    \
       V      V
      A        B

>
>  Is it right to use second level variables (teacher's variables) 
> as fixed effects where there are 230 rows/pupils and 8 teachers (around 
> 25 pupils per teacher - is there enough variance)?
>

I would think so.

David Duffy.
-- 
| David Duffy (MBBS PhD)                                         ,-_|\
| email: davidD at qimr.edu.au  ph: INT+61+7+3362-0217 fax: -0101  /     *
| Epidemiology Unit, Queensland Institute of Medical Research   \_,-._/
| 300 Herston Rd, Brisbane, Queensland 4029, Australia  GPG 4D0B994A v