Intracorrelation Coefficient producing unexpected results..can you help me understand

#1
Hi,

So i am trying to measure the reliability of measurements taken with a callipers by 4 different users on the same 10 samples. This my data in millimeters:


Barry Sarah Aoife Jen
Sample
1 2.18 2.15 2.27 1.62
2 1.695 1.82 2.07 1.33
3 1.76 1.46 2.20 1.18
4 1.83 1.94 3.00 1.51
5 1.54 1.63 1.86 1.14
6 2.105 2.15 3.10 1.72
7 1.175 1.78 1.76 1.12
8 1.966 2.18 1.88 1.57
9 1.591 1.40 1.95 1.07
10 1.625 1.38 1.80 1.23


When i calculate the intracorrelation coefficient in SPSS using a two way random model, absolute agreement i get the following result.

ICC single = .253 and ICC average = .575

This is fine, exactly what i expected, the reliability is quite poor.


I then performed the same experiment with the same users and the same instrument however the 10 samples they measured were different (thicker).

here are my results:


Barry Sarah Aoife Jen
Sample
1 3.15 3.21 2.88 1.87
2 3.76 1.92 2.90 1.29
3 2.39 3.04 2.57 1.87
4 3.14 2.82 3.14 1.8
5 4.05 3.19 3.47 3.21
6 2.96 3.20 2.59 1.89
7 1.89 3.18 1.70 1.58
8 2.94 2.83 2.63 2.11
9 3.20 4.32 3.58 2.45
10 2.10 2.13 1.83 1.48


When i calculate the intracorrelation coefficient in SPSS using a two way random model, absolute agreement i get the following result.

ICC single = .327 and ICC average = .660


Again very poor reliability as i expected however, comparing the two results it appears that the reliabilty between measures for the 2nd test was higher than for the 1st test
.327 (2nd) > .253 (1st). This baffles me as you can clearly see when you actually work out the differences in each dataset that the actual difference across users (mm) is much higher in the 2nd dataset than in the first. Looking at the average range 1.34 (2nd) >0.8 (1st) making it alot less reliable!

I am obviously missing something here,just wondering would anybody be able to shed a bit of light on my situations it is really baffling me!

Thanks

Siobhan