Hi
This is the data I came up with.
*Group 1*
*Group 2*
*Group 3*
*Group 4*
*Group 5*
*Group 6*
**
10.5
7
7.5
8
9
10
11
8
8.5
9.5
7
6
9.5
6.5
6.5
7
8
8.5
9
7
9
6
7
12
9
6.5
6.5
9
6
7.5
11
9
7.5
*Averages*
10.17
7.50
7.88
7.71
8.00
7.75
You can see that Group 1 shows a mean of around 10 and the other groups show
a mean of around 8. I would like to include the sample size of each group in
some regression modelling to show that the level is indeed roughly of 8, not
10.
I have looked into the suggestions so far (i.e., GLM modelling in Minitab
and heteroskedasticity) but I am trying to find out the level of the groups
using equations such as y=b, whilst including sample sizes as some sort of
weights. Parameter *b* will be estimated by the suggested procedure. I know
that in the ARIMA and structural modelling (i.e., StructTS in R), the level
is automatically modelled. However, once again, they do not include
frequency/sampling/case weights (thank you Anthony and Allan for the correct
terminology!) in their procedures. This could be a simple problem, maybe I
am missing something here... Thank you all for any further
suggestions/pointers and thank you very much for all the pointers suggested
so far.
Concerning my other problem, if we toss an unbiased coin, we know that the
theoretical outcome should be 50/50 if we toss the coin infinitely, e.g.,
say 10,000 times. However we know that if we toss it 20 times, we may get a
ratio of 45/55. Is there a way to obtain a prediction band for sample
size=20? This is just to satisfy my curiosity...
Thank you again everyone for any pointers.
Best Regards,
Chris Roberts
|